Xinjun Chen - Application of Gray System Theory in Fishery Science-Springer-CAP (2023)

Xinjun Chen Editor
Application of Gray
System Theory in
Fishery Science
Application of Gray System Theory in Fishery
Science
Xinjun Chen
Editor
Application of Gray System

Theory in Fishery Science
Editor
Xinjun Chen
College of Marine Sciences
Shanghai Ocean University
Lingang New City, Shanghai, China
ISBN 978-981-99-0634-5 ISBN 978-981-99-0635-2 (eBook)

https://doi.org/10.1007/978-981-99-0635-2
Jointly published with China Agriculture Press
The print edition is not for sale in China (Mainland). Customers from China (Mainland) please order the
print book from: China Agriculture Press.
© China Agriculture Press 2023

This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface
Since the Gray system theory was founded in 1982 by Professor Deng Julong, a
renowned scholar in our country, its theory and methods have been continuously
developing. At the same time, its application in different industries and disciplines
has been deepening unceasingly, including fisheries science. As a result a series of
good results were obtained, which created favorable conditions for the development
of the Gray system theory. In 1998, the Shanghai Ocean University offered a
graduate course entitled “Lecture on Gray Systems” for its postgraduate students.
In 2003, the first textbook The application of Gray system in fishery science was
compiled and published by China Agricultural Press. This book is a revised edition
of Application of Gray system in fishery science. Based on the systematic introduc-
tion of the basic principles and methods of the Gray system, the book combines the
research results of the Gray system in fishery science at home and abroad in recent
years. The book is divided into eight chapters, covering the basic concept and theory
of the gray system, original data processing and gray sequence generation, gray
correlation analysis, gray clustering analysis, gray system modeling, gray prediction,
gray decision-making, and gray linear programming.
This book is highly readable and practical. Its re-publication will offer new
research methods and research tools for researchers engaged in fisheries science.
The monograph can be used by scientific workers and research units engaged in
fishery and marine biology; it is a good reference book and also can be used as
teaching material for undergraduate and graduate students of fishery.
However, due to the limitations of length and reference materials, as well as the
limited level of the authors, there may still be inappropriate points in this mono-
graph. Therefore, readers are requested to make corrections and suggestions.
This book is supported by the Top fisheries disciplines of China, the high-level
innovation team of local universities in Shanghai (the strategic innovation team of
Oceanic Fishery Science and Technology), and the outstanding scientific research
v
vi Preface
talents and innovation team of the Ministry of Agriculture (the sustainable develop-
ment of oceanic squid resources).
Lingang New City, China Xinjun Chen

Contents
1 Overview of Gray System Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Xinjun Chen
2 Raw Data Processing Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Xinjun Chen
3 Gray Correlation Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Xinjun Chen
4 Gray Cluster Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Xinjun Chen
5 Basic Principles of Gray Dynamic Modeling . . . . . . . . . . . . . . . . . . . 105
Xinjun Chen
6 Gray Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Xinjun Chen, Minyang Xie, and Jintao Wang
7 Gray Decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Xinjun Chen
8 Gray Linear Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Xinjun Chen
vii
Chapter 1
Overview of Gray System Theory
Xinjun Chen
Abstract Gray system theory is one part of the fields of control theory. It is the
product of the viewpoint and method of cybernetics applied to social economic
system and natural science system, and the combination of cybernetics and opera-
tional research. It takes the gray system as the research object, taking the whitening,
desalination, quantification, modeling, and optimization of the gray system as the
core and taking the prediction and control of the development of various gray
systems as the goal. Gray system is between white system and black box, in which
some information is known and some information is unknown. The gray system
theory is aimed at the problem of uncertainty with little data and no experience,
which is called “minority uncertainty.” The sequence of systematic behavior is often
irregular and varies randomly. For random variable and random process, people
often use the method of probability and statistics. The method of probability
statistics requires a large amount of data, so it is necessary to find statistical rules
from a large amount of data. Gray system theory is not from the angle of looking for
the statistical law and through a large number of samples to study, but with the
method of number processing, will be chaotic original data collated into a more
regular generating sequence. It is a kind of realistic law, not a priori law, to explore,
discover, and seek the inner law from the disorderly original data. The main research
contents of gray system are: the modeling theory of gray system, the relational
analysis theory of gray factors, gray prediction theory and decision theory, gray
system analysis and control theory, gray system optimization theory, and so on. In
1981, Professor Deng Julong, an expert of Chinese cybernetics, first put forward the
concept of gray system. Since 1982, the gray system theory has been successfully
applied in agriculture (including fishery), industry, meteorology, and other fields. In
this chapter, the concept, characteristics, research contents, and development status
of gray system theory are summarized, and the application of gray system in fishery
is briefly introduced.
X. Chen (✉)
College of Marine Sciences, Shanghai Ocean University, Lingang New City, Shanghai, China
e-mail: [email protected]
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023 1
X. Chen (ed.), Application of Gray System Theory in Fishery Science,
https://doi.org/10.1007/978-981-99-0635-2_1
2 X. Chen
Keywords Gray system theory · Gray correlation · Gray clustering · Gray

modeling · Fishery science
The trend of highly integrated modern science and technology on the basis of a high
degree of differentiation has led to the emergence of a group of disciplines of
systems science with methodological significance. Systematic science reveals a
deeper and more essential internal connection between things, which greatly pro-
motes the holistic process of science and technology. Many complex problems that
have long been difficult to solve in the field of science have been solved with
the emergence of new disciplines of systems science, and people’s understanding
of the evolutionary laws of nature and objective things has gradually deepened due
to the emergence of new disciplines of systems science. Systems theory, information
theory, and cybernetics, which were born in the late 1940s, emerged from dissipative
structure theory, synergetics, catastrophe theory, and fractal theory in the late 1960s
and early 1970s, as well as the supercycles that appeared in the middle and late
1970s. Theories, dynamic system theory, and pansystems theory are all new disci-
plines of system science with horizontality and a cross-cutting nature (Chen
2003, 2023).
In systematic research, due to the existence of internal and external disturbances
and the limitation of the level of understanding, the information obtained by people
is often uncertain. With the development of science and technology and the progress
of human society, people’s understanding of the uncertainty of various types of
systems has gradually deepened, and the study of uncertain systems has also become
increasingly in depth. In the second half of the twentieth century, in the fields of
systems science and systems engineering, various uncertain system theories and
methods continuously emerged to form a large landscape. Examples include fuzzy
mathematics created by Professor Lotfi Asker Zadeh in the 1960s (Syropoulos
2020), gray system theory created by Professor Deng Julong in the 1980s (Deng
1982), rough set theory created by Polish computer scientist Zdzisław I. Pawlak in
the 1980s, and unascertained mathematics created by Professor Guangyuan Wang in
the 1990s (Wang 1990). An emerging discipline in the study of uncertain systems.
These new disciplines discussed the theories and methods for describing and
processing various types of uncertain information from different angles and sides
(Chen 2003, 2023).
1.1 Basic Concepts of the Gray System
In people’s social economic activities, they often encounter incomplete or vague

information. Many systems, such as society, economy, agriculture (fisheries), ocean,
industry, and biology, have large uncertainties. For example, in fishery production,
even if the information on operating fishing vessels, number of fishermen, operating
hours, and operating fishing grounds is completely clear, due to unclear information
1 Overview of Gray System Theory 3
such as fishery resources, marine environment, fishermen’s technical level, sea

conditions, market conditions, etc., it is difficult to accurately predict fishing yield,
fishery output, and economic benefits. Similarly, due to changes in marine environ-
mental conditions, it is difficult to accurately predict the location, trend, and yield of
operating fishing grounds in fisheries forecasting.
The research object of gray system theory is the “poor information” uncertain
system with “some information is already known, and some information is
unknown.” It achieves an accurate description and understanding of the real world
through the generation and development of “partially known information.”
1.1.1 Gray Meaning and Gray Phenomenon
In cybernetics, people often use the shade of color to describe the degree of clarity of
information. For example, objects with unknown internal information are called
black boxes. Under normal circumstances, we use “black” to indicate that the
information is unknown, “white” to indicate that the information is completely
clear, and “gray” to indicate that some information is clear and some information
is not clear. The information is partially known and partially unknown, i.e., the
information is incomplete, which is the basic meaning of “gray.” In different
situations, “gray” can be transformed or extended to different meanings. In nature
and human society, the “gray” phenomenon is universal. The “gray” phenomenon
refers to a phenomenon whose information is partially known and partially
unknown. For example, a certain type of fishery resource is a gray phenomenon,
and we can roughly estimate it. However, the amount of fishery resources cannot be
accurately determined.
1.1.2 Gray System
The objective world is the material world and the world of information. However, in
the fields of engineering technology, society, economy, agriculture, fishery, envi-
ronment, ecology, and military, there are often situations of incomplete information.
For example, the system factors or parameters are not completely clear, the relation-
ship between the factors is not completely clear, and the system structure is not
completely known. The mechanism of the system is not fully understood.
We call the system with completely clear information the white system. For
example, in a circuit system, when the resistance value is given, there is a clear
relationship between voltage and current, which is a white system, and it is a white
system with a physical prototype.
A system with completely unclear information is called a black system. For
example, a distant planet can also be regarded as a system. Although it is known
4 X. Chen
to exist, it is completely unknown in terms of volume, mass, and distance from Earth.
This is a black system.
A system with partially clear and partially unclear information is called a gray
system. For example, in a fishery production system, fishery resources, water
temperature, salinity, ocean currents, plankton, fishing vessels, fishing vessel param-
eters, fishermen, and fishery management measures are all factors that affect fishery
yield. The mapping relationship between various factors and fishing yield is difficult
to obtain. Obviously, the fishery production system is a gray system without a
physical prototype.
1.2 Overview of the Development of Gray System Theory
1.2.1 The Scientific Background of Gray System Theory
According to the dialectical materialist view of science and technology, the emer-
gence of any new science and theory has two aspects: inevitability and contingency.
The law of the development of science and technology determines that in a certain
historical period and at a certain development stage, new science and new theories
will inevitably emerge. Gray system theory is also produced against a certain social
development background. At the branch point of scientific development, Professor
Deng Julong conformed to the needs of society and the law of scientific development
and created gray system theory with great success.
Professor Deng has been engaged in the study of “prediction and control of
economic systems” and “fuzzy systems” since the late 1960s and has been exposed
to a large number of systems with some known and some unknown information. Use
the method of fuzzy mathematics or probability theory to describe. Fuzzy mathe-
matics mainly focuses on the phenomenon of “cognitive uncertainty” and uses the
membership function to solve the problem based on experience. With no experience,
no typical distribution conditions, and a small sample size, Professor Deng has
conducted painstaking and fruitful research on this issue. Finally, in 1982, a new
theory of gray systems was developed, and the first paper on gray systems was
published in the journal Systems and Control Letters (Deng 1982). The editor-in-
chief of the magazine, Professor Roger Brockett of Harvard University, commented
on Professor Deng’s first paper on gray systems: The term “gray system” was first
created!
1.2.2 Overview of the Development of Gray System Theory
In 1982, the gray system theory established by the Chinese scholar Professor Deng
Julong was a new method for studying the problem of uncertainty with little data and
poor information (Deng 1982). Gray system theory uses “small sample” and “poor
information” uncertain systems with “partially known information” and “poor

information” as the research object. To achieve the correct description and effective
monitoring of system operation behavior and evolution pattern. Many systems, such
as society, economy, agriculture, industry, ecology, and biology, are named
according to the field and scope of the research object, while the gray system is
named according to the color.
The rapid development of gray system theory and its successful application in
many scientific fields have won the recognition and attention of the international
academic community. The English version of the international academic journal
“The Journal of Gray System,” founded in the UK in 1989, has become the core
journal of important international abstract institutions such as the British Scientific
Abstracts (SA) and the American Mathematical Review (MR). More than 300 aca-
demic journals around the world have accepted and published gray system papers.
The Proceedings of the American Computer Society, Kybernetes, and other inter-
national journals published the gray system special edition. At present, many well-
known scholars in the UK, the USA, Germany, Japan, Australia, Canada, Austria,
Russia, and other international organizations are engaged in gray system research
and have made important contributions to the development of gray system theory.
The application of gray system theory has been extended to many fields, such as
industry, agriculture, society, economy, energy, transportation, petroleum, geology,
water conservancy, meteorology, ecology, environment, medicine, education,
sports, military, law, finance, etc. It solves a large number of practical problems in
production, daily life, and scientific research. Since the 1980s, hundreds of cities,
counties, and provincial regions in China have applied the methods, models, and
techniques of gray system theory to study regional social, economic, and techno-
logical development issues, compile comprehensive development plans, and pro-
mote the health of the regional economy. Development: National, provincial, and
municipal science funds actively support gray system research, and a large number
of gray system theory or applied research projects are funded by various types of
funds each year. According to statistics, more than 200 achievements in gray
systems across the country have won national or provincial and ministerial awards.
In 2002, scholars in gray systems in China won the World Organization for System
and Control Award.
There are more than 100 universities in the world, such as Huazhong University
of Science and Technology, Renmin University of China, Tsinghua University,
Zhejiang University, Shandong University, Nanjing University of Aeronautics and
Astronautics, University of Maryland, Toyohashi University, Kanagawa University,
Vienna University of Economics, and France Aerospace Center, which have set up
gray system theory courses. Huazhong University of Science and Technology and
Nanjing University of Aeronautics and Astronautics in China recruit and train
doctoral students in the field of gray systems.
Many publishing institutions at home and abroad, such as Science Press, National
Defense Industry Press, Huazhong University of Science and Technology Press,
Jiangsu Science and Technology Press, Shandong People’s Publishing House,
Science and Technology Literature Press and others, have published more than
6 X. Chen
60 academic books on the gray system. The Blue Book of Science and Technology
in China (No. 8) compiled and published by the Ministry of Science and Technology
of China affirmed gray system theory as a new soft science method created by
Chinese scholars. At the same time, gray system theory has become a hot topic of
attention and discussion in many important international conferences and will
undoubtedly play a positive role in further understanding gray system theory in
the world system science community.
In the past 40 years, gray system theory has established itself in the forest of
science with its strong vitality, which has established its academic status as an
emerging cross-disciplinary discipline (Liu et al. 2014). The vigorous vitality and
broad development prospects of gray system theory are increasingly being recog-
nized and valued by all walks of life at home and abroad. As an emerging discipline
that is undergoing continuous development and improvement, gray system theory
still has many problems that need to be further studied:
1. The connotation and exact description of the gray concept and the basic princi-
ples of the gray system;
2. The operation of the gray number gray algebraic system;
3. The information content of simple gray numbers and composite gray numbers;
4. The modeling mechanism, function, and application scope of different gray
models;
5. The information and scientific basis for constructing the whitening weight
function of the gray number;
6. The gray relational axiom, gray relational degree, and stability of relational
order;
7. The construction, function, and qualitative and quantitative coupling points of
practical buffer operators;
8. The properties of the gray nonnegative matrix, gray matrix spectral drift, and
gray deepening research on input–output models;
9. Comparative research on uncertainty methods such as gray system theory,
inexact set theory, unascertained mathematics, probability statistics, fuzzy math-
ematics, and innovation of uncertain mathematical theories;
10. Gray system theory application in various scientific fields and systems analysis,
market forecasting, financial decision-making, asset evaluation, enterprise plan-
ning, and management decision-making at all levels of government.
1.3 Main Content of the Gray System Theory
After nearly 40 years of development, gray system theory has basically established a
structural system of an emerging discipline. Its main contents include a theoretical
system based on a gray hazy set, an analysis system based on gray correlation space,
a method system based on gray sequence generation, and a model system with a gray
model (GM) as the core. Evaluation, modeling, prediction, decision-making, control,
and optimization are the main technical systems. Gray hazy sets, gray algebraic
systems, gray equations, and gray matrices are the basis of gray system theory.
Starting from the beauty and perfection of the disciplinary system, there are many
issues worthy of further study. In addition to gray relational analysis, gray system
analysis also includes gray clustering and gray statistical evaluation. Gray sequence
generation is achieved through the function of the sequence operator. The sequence
operator mainly includes the buffer operator (weakening operator and strengthening
operator), the mean value generator, the ratio generator, the cumulative generator,
and the accumulative generator. The gray model is constructed according to the five-
step modeling idea. It weakens the randomness through the role of gray generation or
the sequence operator and mines the potential pattern. Through the exchange
between the gray difference equation and the gray differential equation, the discrete
data sequence is used to establish the continuous data. A new leap in dynamic
differential equations. Gray prediction is a quantitative prediction based on the GM
model. According to its function and characteristics, it can be divided into several
types, such as series prediction, interval prediction, catastrophe prediction, seasonal
catastrophe prediction, waveform prediction, and system prediction. Gray decision-
making includes gray target decision-making, gray relational decision-making, gray
statistics, clustering decision-making, gray situation decision-making, and gray
hierarchical decision-making. The main content of gray control includes the control
problem of the intrinsic gray system and control based on the gray system method,
such as gray relational control and GM (1, 1) predictive control. Gray optimization
techniques include gray linear programming, gray nonlinear programming, gray
integer programming, and gray dynamic programming.
Gray system theory mainly studies the “small sample uncertainty problem,”,
which is significantly different from the probability statistics of the “large sample
uncertainty problem” and the fuzzy mathematics of the “cognitive uncertainty
problem.” Probability statistics, fuzzy mathematics, and gray system theory are the
three most commonly used methods for studying uncertain systems. The study
subjects all have a certain degree of uncertainty, which is the common point of the
three. According to the results of Professor Deng’s research, there are significant
differences among the three (Table 1.1).
Fuzzy mathematics focuses on the problem of “cognitive uncertainty,” and its
research object has the characteristics of “clear connotation and unclear extension.”
For example, “young people” is a vague concept because everyone is very clear
about the connotation of “young people.” However, it is very difficult to delineate an
exact range, in which the young people are within this range and the young people
are not outside the range. The extension of the concept of young people is not clear.
For this type of “cognitive uncertainty” problem with clear connotations and unclear
extensions, fuzzy mathematics is mainly processed by the membership func-
tion based on the experience.”
The study of probability statistics is the phenomenon of “random uncertainty,”
which focuses on the historical statistical law of the phenomenon of “random
uncertainty” and examines the possibility of the occurrence of each of the “random
8 X. Chen
Table 1.1 Differences between gray systems, probability statistics, and fuzzy mathematics Chen
(2003, 2023)
Gray system Probability statistics Fuzzy mathematics
Connotation Small sample size Large sample size Cognitive
uncertainty uncertainty
Foundation Gray hazy set Kantoji Fuzzy set
Basis Information coverage Probability distribution Membership
function
Means Generate Statistics Boundary value
Characteristics Little data Multiple data By experience
Requirement Arbitrary distribution Typical distribution is Function
allowed required
Goal Law of reality Historical statistics Cognitive
expression
uncertainty” phenomena with multiple possible outcomes. The starting point is a

large sample, and the subject is required to obey a certain typical distribution.
Gray system theory focuses on the “small sample” and “poor information”
uncertainties that are difficult to solve by probability statistics and fuzzy mathemat-
ics. It is characterized by “less data modeling.” Different from fuzzy mathematics,
gray system theory focuses on objects with “clear extension but unclear connota-
tion.” For example, by 2050, China will control the total population between 1.5 and
1.6 billion. This “between 1.5 and 1.6 billion” is a gray concept, and its extension is
very clear. It is not clear which specific value is between 1.6 billion.
The main contents of gray system theory include the correlation analysis method
of the mutual influence of factors; the gray statistics and gray clustering method
based on the whitening weight function; the cumulative generation and the cumu-
lative generation method for data processing; and the gray modeling for the estab-
lishment of a differential equation model. The methods include the gray prediction of
series catastrophe prediction, seasonal catastrophe prediction, topological predic-
tion, and system prediction; the gray decision-making method, gray linear program-
ming, and gray dynamic control based on system behavior prediction.
1.3.1 Gray Relational Analysis
Gray relational analysis includes system factor analysis and system behavior anal-
ysis. Analysis of the factors that affect the main behavior of the system is called
system factor analysis, while the quantitative comparison of the behavior of different
systems is called system behavior analysis. For example, for the human–machine–
environment system, the factors that affect the safety of the system include human
physiological and psychological characteristics, operating skills, and health condi-
tions. Environmental factors such as humidity, noise, and vibration. Among the
above factors, it is necessary to analyze which factors are primary and which are
secondary, which is the factor analysis of system security.
The gray correlation analysis method is a method to measure the degree of
correlation according to the degree of similarity or difference between the factors
of the system or the behavior of each system. Because the gray correlation analysis is
based on the development trend, there is no excessive requirement on the size of the
sample, and there is no need for the typical distribution pattern. The calculation is
small. Even if there are more than ten variables, they can be calculated by hand, and
there will be no grayscale. The quantitative results of the correlation are inconsistent
with the qualitative analysis.
Probability statistics, fuzzy mathematics, and gray system theory are the three
most commonly used methods for studying uncertain systems. The study subjects all
have a certain degree of uncertainty, which is the common point of the three.
1.3.2 Gray Modeling
The mathematical model quantitatively expresses the mathematical relationship

between system factors (variables). However, the general method can only be used
to establish a difference equation model, which makes it difficult to analyze and
describe the entire process of the system. Some scholars have pointed out, “For life
science, economics, biomedicine, etc., we hope to establish differential equation
models.” Gray system theory has successfully solved the problem of differential
equation modeling, which has always been considered difficult to solve due to the
proposed new ideas and new methods, such as the generation number, the conver-
gence of the discrete function, the limit and smoothness, the gray derivative, and the
gray differential equation. The general formula of the dynamic differential equation
model is GM (n, h), where n represents the order of the differential equation and the
number of variables. Different values of n and h represent different system factor
relationships, and their descriptive function is quite strong. Commonly used gray
models are GM (1, 1), GM (2, 1), GM (0, 2), and GM (1, h).
Based on the premise of qualitative analysis and the backing of quantitative
analysis, gray system theory proposes a five-step modeling method: a language
model, network model, quantitative model, dynamic model, and optimization model.
It is worth noting that premodeling data generation and postmodeling residual
identification are two unique and effective methods in gray theory. The accumula-
tion of the original series is the most commonly used data generation method in gray
modeling. This method provides intermediate information for modeling and
weakens the randomness of the original data.
10 X. Chen
1.3.3 Gray Prediction
Gray prediction refers to the prediction made by the gray model GM (1, 1).
According to its functions and characteristics, gray prediction can be divided into
five categories: series prediction, catastrophe prediction, seasonal catastrophe pre-
diction, topological prediction, and system prediction.
The prediction of the magnitude of the development and change in the system
behavioral characteristics is called a series of predictions. The development and
change of the system are continuous in time and orderly in space. Sequence
prediction uses the time series or spatial sequence of the system to perform timing
or fixed spatial prediction of the system. The collection of behavioral eigenvalues
can be either equally spaced or nonequally spaced. In fact, sequence prediction
studies the variation in behavioral characteristics over time or space.
Prediction of the abnormal value when the system behavior characteristic quan-
tity will exceed a certain threshold is called catastrophe prediction. The feature of
catastrophe prediction is to predict the time of occurrence of “catastrophe” or the
occurrence time of anomalous numbers. The magnitude of the outlier is often a gray
number with given upper and lower limits. For example, the forecast of a harvest
year for a certain fishery resource is the forecast of the year when the annual average
catch is relatively high (annual yield is more than 1000 tons), which is called the
forecast of the harvest year, while the forecast of the poor year is that the annual
average catch is too small (e.g., less than 400 tons). The prediction of the occurrence
of a catastrophe in a certain season or a certain time zone of the year is called the
prediction of seasonal catastrophe. For example, the forecast of fishing season or
fishing season is the forecast of the occurrence of fishing in a specific time zone.
Topological prediction is the prediction of the characteristic data waveform of the
system behavior over a period of time. Because many points can form a waveform,
topological prediction specifies many given values. For each given value, a set of
point distribution data can be obtained on the given curve, and then GM (1, 1) is
established for each set of point distributions. The model predicts the time interval
for the future development and change of this set of given values.
Predicting the relationship between several variables (factors) included in the
system together and predicting the role of the dominant factors in the system is called
system prediction.
1.3.4 Gray Decision-Making
The so-called decision-making refers to the selection of an appropriate countermea-

sure to address the occurrence of an event to achieve the best effect. Gray system
theory proposes the gray decision because the establishment of the decision model is
based on the gray class, that is, according to the gray model, especially the GM (1, 1)
model.
Gray decision-making includes gray situation decision-making, gray-level

decision-making, and gray planning. Gray situation decision-making refers to the
decision made on the basis of the binary combination of events and countermea-
sures. Gray-level decision-making is a decision that coordinates, unifies, and com-
promises the intentions and functions of various decision-making groups (levels).
Gray planning is a decision that combines general planning with gray prediction and
gray number generation.
1.3.5 Gray Linear Programming
Linear programming is an important branch of operations research. It is a

mathematical model that is widely used and easy to implement in the study of
multivariable systems. It is also the most commonly used method for deterministic
decision-making. The main problem it solves is how to maximize the role of limited
resources (including human resources) and find effective ways to rationally use
human, material, and financial resources. However, general linear programming
has problems such as the inability to reflect the time-varying constraint conditions
and gray parameters. If the gray system idea and modeling method are used to solve
the problem, we will generally call it gray linear programming.
1.3.6 Gray Control
The execution of decisions is called control. The so-called gray control refers to the
control of the intrinsic gray system, the control of the gray parameters in the system,
or the predictive control composed of the GM (1, 1) model. The basic method of gray
control is to find the pattern of system behavior development and change through the
system behavior data series, predict the future behavior of the system according to
the mastered pattern, and make control decisions based on the predicted value of
future behavior.
Traditional control is control by judging whether the behavior that has occurred in
the system meets the requirements, which is a kind of ex-post control. Its shortcom-
ings are that it cannot be prevented in advance, it cannot be controlled in real time,
and its adaptability is not strong. Gray predictive control is a kind of advanced
control that can prevent problems before they occur, control them in a timely
manner, and improve adaptability.
12 X. Chen
1.4 The Status and Characteristics of Gray System Theory

in Scientific Development
1.4.1 The Position of Gray System Theory in the Discipline

System
People have different understandings and perspectives on objective things, and the
ways of dividing the disciplinary system are also different. In the seventeenth
century, based on the understanding that scientific classification should correspond
to human memory, imagination, and judgment, Bacon advocated dividing science
into three categories: history, poetry and art, and philosophy. Later, Saint-Simon and
Hegel proposed the idea of dividing disciplines according to metaphysics and
idealism, respectively. In the late nineteenth century, Engels proposed dividing
disciplines according to the different forms of material movement and their inherent
order and establishing a scientific system structure, which laid a solid scientific
foundation for the classification of disciplines.
In China, people usually divide science into natural science, social science, and
thinking science according to different research objects, as well as philosophy and
mathematics, which are summarized and run through the three fields. The basic
disciplines of natural sciences are accustomed to being divided according to the six
categories of mathematics, physics, chemistry, heaven, earth, and biology. Professor
Qian Xuesen advocated that the entire science and technology system should be
divided into six scientific fields: natural science, social science, systems science,
thinking science, human science, and mathematical science. In terms of disciplinary
division, we first classify scientific problems according to complexity and uncer-
tainty and then point out the corresponding cross-disciplines with methodological
significance according to the nature of various disciplinary problems, thus clarifying
the cross-disciplinary group of gray system theory.
Use box (Ω) to represent the set of all things in the world. Circles A, B, C, and D
are used to represent the set of simple things, complex things, deterministic things,
and uncertain things, respectively, and the four-ring diagram of the classification of
scientific problems can be obtained (Fig. 1.1). By marking the scientific methods for
solving various problems, the four-ring diagram of the cross-disciplinary classifica-
tion is obtained (Fig. 1.2).
By comparing Figs. 1.1 and 1.2, it can be seen that the grey system theory, as a
scientific method to solve uncertain and semi-complex problems, achieves a new
leap compared with pobability statistics and fuzzy mathematics to solve simple
uncertain problems. However, the solution of complex and uncertain problems
needs a new breakthrough in nonlinear science.
˞
Semi-deterministic
Deterministic complex problems Uncertain complex
complex problems problems
C BD
Uncertain semi-
Deterministic CB BD complex problem
semi-complex
˟ problemA CB B DA ˠ
AC AD
C AD
Deterministic simple Uncertain simple
problems Semi-deterministic problems
simple problems
˝
Fig. 1.1 Four-ring diagram of scientific problem classification (Chen 2003, 2023)
˞
Self-organization
Systems Science theory Nonlinear Science
C BD
Operations CB BD Gray system
Research
˟ B DA ˠ
A CB
AC AD
C AD
Mathematics Probability and
Logic and intuitive statistics, fuzzy
thinking mathematics
˝
Fig. 1.2 Four-ring diagram of cross-disciplinary classification (Chen 2003, 2023)
1.4.2 The Role of Gray System Theory in the Discipline

System
As a unique new theory, gray system theory has been recognized by the academic
community at home and abroad and has played a huge role in the development of
science. Its applications are widespread in agriculture, fisheries, industry, energy,
transportation, petroleum, geology, and meteorology. It has successfully solved a
14 X. Chen
large number of practical problems in production, life and scientific research in many
scientific fields, such as hydrology, ecology, environment, medicine, military, eco-
nomics, and society.
In the past 40 years of development, gray system theory has established itself in
the forest of science with its strong vitality, which has established its academic status
as an emerging cross-disciplinary discipline. Driven by gray system theory, “gray
hydrology,” “gray statistics,” “gray geology,” “gray breeding,” “gray medicine,”
“gray control theory,” “gray chaos theory,” “gray system analysis of regional
economy,” and a number of emerging interdisciplinary disciplines have been suc-
cessively produced, which promoted the development of science.
After nearly 40 years of development, gray system theory, as an emerging
discipline, stands on its own in the forest of science with its strong vitality. Professor
Xuesen Qian, the founder of fuzzy mathematics, Professor Lotfi A. Zadeh (USA),
and the founder of synergetics, Professor Herman Haken (Germany), also spoke
highly of the research on gray systems.
1.5 Research Progress on the Application of Gray System

Theory in Fishery Science
Gray system theory is a new discipline founded by the famous Chinese scholar
Professor Deng Julong in 1982. It is applied to an uncertain system with some
known information and some unknown information. It is a study of small data and
poor information and the movement of uncertain systems. In recent years, this theory
has achieved significant social benefits in various fields of natural science, social
science, and engineering technology, such as aerospace, metallurgy and petroleum,
mechanical and chemical engineering, electronics and electricity, medical and health
care, hydrometeorology, agriculture and forestry, and education and management.
The field of fishery science mainly includes fishery economy and fishing production.
Among them, fisheries resources, water temperature, salinity, ocean currents, plank-
ton, fishing vessel parameters, fishermen, and fishery management measures are all
factors that affect fishing yield. However, the fishery production system is a gray
system without a physical prototype. The traditional probability statistical method,
time series method, and linear regression analysis method require a large number of
samples and follow a typical distribution. In the field of fishery science, where
sample information is relatively scarce, the application of gray system theory can
effectively solve many problems.
Based on the quantitative analysis of the China National Knowledge Infrastruc-
ture (CNKI) literature, the application of gray system theory in fishery science is
divided into the following stages (Table 1.2): the initial stage (1988–1994), the
middle stage (1995–2005), and the current stage (after 2006). The main research
directions and progress of gray system theory in fishery science can be analyzed
from the keywords of these three periods. From Table 1.2, the main research
Table 1.2 High-frequency keyword analysis results of the application of gray system theory in
fishery science in different periods in China (Xie and Chen 2019)
Time Keywords (frequency)
The initial stage Gray system theory (4); fishery yield (3); Prediction Model (1); yield
(1988–1994) prediction (1); gray correlation analysis (1); time series (1); Mariculture
(1); marine fishery (1); Fishery production (1); marine fishing (1)
The middle stage Gray system theory (25); Prediction Model (17); marine fishing (11);
(1995–2005) gray correlation analysis (8); aquatic product yield (7); mariculture (6);
Gray Clustering Method (6); fishing intensity (6); lake eutrophication
(5); fishing yield (5); Marine Fisheries (5); fishery production (4);
Comprehensive Evaluation (4); fishery resources (4); pond fish culture
(4); lake water quality (3); yield relationship (3); fishery production (3);
fishery economy (3); structural adjustment (3);
The current stage Gray system theory (51); gray correlation analysis (49); Prediction
(after 2006) Model (34); Marine Fisheries (25); fishery economy (24); Mariculture
(23); marine fishing (22); aquatic product yield (20); aquatic product
processing industry (18); influencing factors (17); Industrial Structure
Adjustment (16); fishery production (13); fishery output (12); evaluation
index (12); freshwater aquaculture (11); pelagic fisheries (9); time series
(8); GM (1,1) Model (7); model precision (7); recreational fisheries (7)
direction in the early stage was the prediction of marine fishing or aquaculture
production and gray correlation analysis; in the middle stage, on the basis of
previous studies, the research on the aquaculture and marine fishing industry was
strengthened, resulting in the environmental assessment of fishery waters. The
current stage is the deepening of the fishery economic industry and the optimization
of the gray forecasting model.
Gray system theory has been widely applied and theoretically studied in fishery
science, mainly focusing on the following aspects: fishery economy, aquaculture,
environmental assessment of fishery waters, and forecasting of fishing conditions.
1.5.1 Fishery Economy Industry
The fishery economy industry involves many components, and the methods of gray
system theory applied are also diverse. It mainly includes industrial restructuring,
evaluation of sustainable use of fishery resources, comparison of industrial compet-
itiveness, analysis of factors affecting economic output, and regional economic
division. The analysis of industrial structure adjustment is mainly to calculate the
correlation between the total production of fisheries and the production of each part
of the fishery through the gray correlation method and establish the GM (1, 1) model
to predict the output of each part of the production, thereby making the proportion of
each part of the industry structure. Recommendations for adjustment. For example,
Song et al. (1999) and Song (2001) analyzed the correlation between the total fishing
16 X. Chen
yield of Zhejiang Province and the yield of various operation methods, established
the GM (1, 1) model for prediction, and obtained the correlation degree of various
operation methods from 1980 to 1990. The order of size was fixed net, drift net, and
trawl net, and trawl net, fixed net, and drift net in 1991–1997. In 2000, the total
fishing yield of Zhejiang Province reached 33.5–36.5 million tons. Subsequently, the
same analysis was conducted on the structural adjustment of the marine aquaculture
industry in Zhejiang Province. In addition, many scholars have also used this method
to analyze the industrial structure of fisheries and have achieved good results. Song
et al. (1998) analyzed the current situation of marine fishing vessels in Shandong
Province and used gray system theory to predict the fishing effort in 2000. Based on
these results, they optimized the configuration of marine fishing vessels in Shandong
Province. Song and Liu (2010) established the GM (1, 1) model using gray system
theory to study the development trend of the number, total power, and average power
of marine motor fishing vessels in China and found that the number of marine motor
fishing vessels will show a downward trend. The total power and the average power
showed an upward trend.
Since there are many uncertainties in the evaluation criteria for the sustainable use
of fishery resources, gray system theory can be used to quantify them to evaluate the
level of sustainable use, thereby understanding the development status of fishery
resources. Chen (2001, 2003) analyzed the development of fishery resources in the
East China Sea from 1978 to 1990 and selected a total of 23 indicators in the three
evaluation index subsystems of the resource environment, society, and economy,
and the optimal value of each sample point was formed into the mother. The index
sequence of each sample point was used as a subsequence, and the different weights
of each indicator were determined by the analytic hierarchy process. The correlation
between each sample point and the parent sequence was calculated as the evaluation
standard. The results showed the fishery resources in the East China Sea from
1978 to 1990. In particular, the sustainable utilization level was the lowest in
1983. Chen and Zhou (2004) proposed a comprehensive evaluation and evaluation
model for the sustainable use of fishery resources based on gray system theory and
combined it with the least squares criterion. This method has greater advantages than
the traditional bioeconomic model and comprehensively reflects the model. Regard-
ing various aspects of the sustainable use of fishery resources, the study used the
development of fishery resources in the East China Sea from 1978 to 1990 as an
empirical analysis. After 1978, the level of sustainable use of fishery resources in the
East China Sea showed a downward trend. The lowest level of sustainable use in
1990 was only half of that in 1978.
1.5.2 Aquaculture Industry
The aquaculture industry is an important industry in fishery science, and its devel-
opment trend can be used to measure the economic level of a country and affect the
fishery economy and the income of fishermen. The application of gray system theory
in the aquaculture industry mainly includes aquaculture yield prediction, aquaculture

input–output ratio optimization, fishing port discharge forecasting, analysis of fac-
tors affecting the aquaculture industry, and selective breeding. Xie et al. (1998) used
the five-step modeling method in gray dynamic control, i.e., the language model,
network model, quantitative model, dynamic model, and optimization model, to
analyze the investment and output value of grass carp and mud carp in the ponds of
the Pearl River Delta. It is concluded that if 18% of the funds are extracted from the
output of the system as the investment in the next year, the economic system can
maintain a good state of development. Xie et al. (2000) used the gray correlation
analysis method to analyze the relationship of multispecies polyculture fish in
Shunde high-yield ponds and found that the stocking amount of carp had the greatest
impact on the net yield of bighead carp, followed by grass carp stocking on the net
yield of carp. The minimum is the relationship between silver carp stocking and the
net yield of grass carp. The stocking of carp is the dominant parent factor, and the
harvest of bighead carp is the dominant daughter factor. The gray dynamic model
GM (1, 2) was used to establish the input–output relationship model for the main fish
in the high-yield pond. Peng and Chen (2017) analyzed the influencing factors of the
marine shellfish aquaculture industry in China’s coastal provinces from 2003 to
2015 and concluded that the most influential factors were the production of marine
shellfish aquaculture and the number of professional employees, and they also used
the approach of approximating the ideal solution. According to the Technique for
Order Preference by Similarity to Ideal Solution (TOPSIS), the ranking of each
coastal province was obtained, and the results showed that Shandong, Liaoning, and
Fujian were the top three. In addition, Liu et al. (2014) used the gray correlation
analysis method to analyze and evaluate the correlation between the growth shape
and body mass of Japanese flounder and found that the factors with the largest
correlation were the overall length, body length, tail stalk height, and trunk length,
which can be used as an important evaluation indicator for the cultivation of high-
yielding Japanese flounder.
1.5.3 Environmental Assessment of Fishery Waters
The application of gray system theory in the evaluation of fishery waters has
achieved good results. The environment of fishery waters, such as reservoirs,
lakes, and rivers, is a gray system. Due to the incomplete information provided by
the limited spatiotemporal monitoring data, the relationship between pollutants and
the environment is uncertain. The clustering analysis of the gray whitening weight
function in gray system theory satisfactorily solves the problem of fuzzy classifica-
tion of water quality grade evaluation and the inability to quantitatively evaluate
water quality grade. For example, Wang et al. (2006), Yang (1995), Xie (1997), and
Li et al. (2011) successively performed gray cluster analysis on the eutrophication
levels of Poyang Lake, Dongchang Lake, and Dianshan Lake. The types were
18 X. Chen
clustered, and good results were obtained. Zhao et al. (2017) analyzed the relation-
ship between the water quality indicators at the monitoring points and the water
quality evaluation criteria using gray correlation analysis and determined the water
quality evaluation grades at the monitoring points according to the degree of
correlation and the weight of the indicators.
1.5.4 Fisheries Forecasting
Fishery situation forecasting is an important branch of the field of fishery science. It

refers to the forecasting of various elements of fishery resources within a certain
period of time and within a certain range of waters in the future. There are uncer-
tainties in both, and the application of gray system theory can solve many problems.
For example, Li and Chen (2007) analyzed the spatiotemporal distribution and
abundance of offshore mackerel resources and selected the maximum value of the
total yield of mackerel at each longitude (latitude) between 1998 and 2003 as the
optimal vector, which was used as the parent sequence. The total yield composition
vector for each longitude (latitude) in each year was used as the subsequence.
Chen and Zheng (2007) used the gray correlation method to study the spatiotemporal
changes in skipjack resources of the tuna purse seine fisheries in the western and
central Pacific Ocean during the period 1990–2001 and found that in the 12 years
1998, 2000, 1994, 1995 and 1999, the abundance of skipjack resources in 1990,
1991, 1997, and 2001 was relatively low, and the abundance of skipjack resources in
1996, 1992, and 1993 was moderate. Xu et al. (2012) used gray correlation analysis
to analyze the proportion of the optimal surface water temperature range (PF) of the
operating fishing grounds and spawning grounds in the open waters of Peru between
2003 and 2010 as well as the Nino1 + 2 SSTA and fishing effort. Studies have shown
that the La Niña event will increase the proportion of the optimal surface water
temperature area of the operating fishing ground and form a wide upwelling current,
which is conducive to the feeding and growth of the fish and to the recruitment of
jumbo flying squid. Gao et al. (2017), Duan et al. (2018), and Wang et al. (2019)
used the gray correlation method to analyze the relationship between the resource
index and environmental and climatic factors. The different GM (1, N) models and
the GM (0, N) model are used to make relatively accurate predictions of resource
abundance for neon flying squid, Peruvian anchovy, and Argentine flying squid.
References
Chen XJ (2001) Evaluation of sustainable utilization of marine fishery resources. Nanjing Agricul-
tural University, Nanjing. (In Chinese)
Chen XJ (2003) Application of gray system theory in fishery science. China Agricultural Press,
Beijing. (In Chinese)
Chen XJ (2023) Application of gray system theory in fishery science. China Agricultural Press,
Beijing. (In Chinese)
Chen XJ, Zheng B (2007) Spatial and temporal distribution of skipjack resources in Tuna Purse
Seine fishery in the western and central Pacific Ocean [J]. Oceanogr Res 25(2):13–22. (in
Chinese)
Chen XJ, Zhou YQ (2004) Study on the synthesis assessment of sustainable use of fisheries
resources based on gray theory. J Fish Sci China 11(z1):91–95. (In Chinese)
Chen XJ, Tian SQ, Ye XC (2002) Study on population structure of flying squid in Northwestern
Pacific based on gray system theory. J Shanghai Fish Univ 11(4):335–341. (In Chinese)
Deng JL (1982) Control problems of gray systems. Syst Control Lett 1(5):288–294
Duan DY, Chen P, Chen XJ et al (2018) The construction of biomass forecasting model for the
anchoveta (Engraulis ringens) by the gray system model. J Shanghai Ocean Univ 27(2):
284–290. (In Chinese)
Gao X, Chen XJ, Yu W (2017) Forecasting model of the abundance index of winter-spring cohort
of neon flying squid (Ommastrephes batramii) in the Northwest Pacific Ocean based on gray
system theory. Haiyang Xuebao 39(6):55–61. (In Chinese)
Li G, Chen XJ (2007) Tempo-spatial characteristic analysis of the mackerel resource and its fishing
ground in the East China Sea. Period Ocean Univ China 37(6):921–926. (In Chinese)
Li ZL, Ma QM, Xu SQ et al (2011) Application of gray clustering analysis in Dongchanghu. Trans
Oceanol Limnol 3:139–144. (In Chinese)
Liu YX, Liu YJ, Zhou Q et al (2014) Gray relational analysis between main growth traits and body
weight in Japanese flounder (Paralichthys olivaceus). J Fish Sci China 21(2):205–213.
(In Chinese)
Peng DM, Chen PD (2017) Evaluation of Chinese marine shellfish aquaculture industry in
coastal provinces based on gray relationship and TOPSIS. Chin Fish Econ 35(3):78–83.
(In Chinese)
Song WH (2001) Application of gray system theory to structure adjustment of marine industry in
Zhejiang province. J Zhejiang Ocean Univ 20(2):91–111. (In Chinese)
Song XF, Liu L (2010) Forecast for fishing vessels developing trend based on gray system theory.
Fish Mod 37(1):56–59. (In Chinese)
Song XF, Qiu TX, Jiao ZG et al (1998) Linear optimization of marine fishing vessel distribution in
Shandong province. J Fish Sci China 5(4):82–88. (In Chinese)
Song WH, Chi HF, Yang JH (1999) Application of gray system theory to ocean fishing structure
adjustment in Zhejiang province. J Zhejiang Ocean Univ 18(4):296–300. (In Chinese)
Syropoulos A (2020) A modern introduction to fuzzy mathematics. Wiley, New York, NY
Wang GY (1990) Unascertained information and its mathematical treatment. J Haibin Archi Civ
Eng Inst 23(4):1–9. (In Chinese)
Wang XC, Wang LQ, Peng ZR (2006) Eutrophic status and water quality grade evaluations of
Lake Dianshan based on gray-clustering method. J Shanghai Fish Univ 15(4):497–502.
(In Chinese)
Wang YF, Chen XJ, Chen P et al (2019) Prediction of abundance index of Argentine shortfin squid
in the Southwest Atlantic Ocean based on gray system model. Haiyang Xuebao 41(4):64–73.
(In Chinese)
Xie J (1997) Application of gray system theory in evaluation of lake eutrophication degree in China.
J Hydrol 4:9–12. (In Chinese)
Xie MY, Chen XJ (2019) Advances in the application of bibliometrics-based gray system theory in
fisheries science. Trans Oceanol Limnol 5:117–126. (In Chinese)
Xie J, Xiao XZ, Huang ZH et al (1998) The comparative study on factors analysis and yield model
of high-yield fish-pond for the pearl river delta and Yangtze delta. J Shanghai Fish Univ 7(2):
20 X. Chen
Xie J, Huang ZH, Xiao XZ et al (2000) The relationships among polycultured fishes and an input–
output dynamics model in high-yield fish ponds. Acta Ecol Sin 20(2):317–320. (In Chinese)
Xu B, Chen XJ, Li JH (2012) Preliminary study on the influence of water temperature on the
recruitment of Dosidicus gigas. J Shanghai Ocean Univ 21(5):878–883. (In Chinese)
Yang H (1995) Application of gray clustering method in lake water eutrophication evaluation. Fish
Modern 6:36–39. (In Chinese)
Zhao LM, Lu Q, Wang N et al (2017) Application of improved gray correlation analysis method in
fishery water quality assessment. J North China Univ Sci Technol 39(2):110–114. (In Chinese)
Chapter 2
Raw Data Processing Method
Xinjun Chen
Abstract Data is the basic work of statistical analysis and modeling. The processing
of raw data is very important in data analysis and modeling. Different raw data come
from different sources and have different properties. The raw data usually include:
(1) scientific experiment and observation data; (2) socioeconomic statistics; (3) pro-
duction experience data; (4) decision-making and target data of relevant depart-
ments; (5) quantitative data of qualitative information, etc. These original data have
the following four main characteristics: (1) different dimensions, (2) different mag-
nitude, (3) most of the data have a certain randomness, (4) a large number of data
have a certain degree of gray. Therefore, strictly speaking, the majority of the data
collected are gray parameters, with varying degrees of gray. For most gray param-
eters, it is necessary to whiten or desalinate them in order to improve the whiteness
and reduce the gray degree. Because of the above characteristics and problems of the
original data, it is difficult and limited to build the mathematical model by statistical
analysis, so the original data should be transformed according to the classification of
the mathematical model. The main purposes of the transformation are: (1) to make
the index data as normal distribution as possible; (2) to unify the dimensionality of
the variables; (3) to transform the nonlinear relation of the two variables into linear
relation; (4) to replace a group of original variable indexes with a new group of
independent variables with a small number of indexes. The commonly used trans-
formation methods are standardization transformation, range transformation, mean
transformation, initial transformation, modularization transformation, moving aver-
age transformation, weakening operator, and strengthening operator transformation.
In this chapter, we will focus on introducing the source of the original data and its
characteristics and providing several methods of the original data transformation and
use some examples to demonstrate.
Keywords Original data · Data transformation method · Case analysis
X. Chen (✉)
22 X. Chen
2.1 Sources and Characteristics of the Original Data
The original data generally include the regional characteristics of natural resources,
such as sea conditions, meteorology, hydrology, topography, landforms, animals
and plants, and reflect the regional socioeconomic conditions and productivity
levels, such as population and population density, fishing labor, sea area, and
number of fishing vessels. The power of fishing vessels and the total fishery output
value, fishing output value, and aquaculture output value. According to their nature,
raw data can be roughly divided into (1) scientific experiment and observation data;
(2) socioeconomic statistical data; (3) production experience data; (4) decision-
making and target data of relevant departments; and (5) qualitative data and
quantitative data.
Different data have different sources (Chen 2003, 2023). However, in summary,
the main sources are (1) historical statistical data of national statistical departments
and industry departments, which are mostly social and economic indicators, and
(2) historical observation data and scientific experimental reports of relevant busi-
ness departments, which are mostly natural. Factor indicators, such as the observa-
tion data of fishery resources and the environment in the ocean; (3) data obtained
from typical field surveys by selecting representative units or years; (4) data accu-
mulated by regional planning departments through collection, survey, observation,
and calculation. (5) Data obtained by surveying and interviewing workers with
practical experience, production and technical personnel, scientific research person-
nel, and management personnel; (6) decision-making data such as development
plans and construction plans formulated by relevant national departments; and
(7) data in other aspects. The various information and data obtained above are called
raw data. These data sources are different, and their types are different.
From the perspective of utilization analysis, these data have the following main
characteristics (Chen 2003, 2023):
1. Different dimensions. For example, the fishery output value is RMB, the fishery
output is kg, the water temperature is °C, the operation time is days, the voyage is
nautical miles, the fishing effort is ton, kilowatt, boat, number of people, and the
catch per unit effort is ton/day, ton/hour, ton/kW, etc.
2. The order of magnitude is very different. Some numbers are only decimal, and
some numbers are as large as hundreds of millions. For example, the output value
of fisheries is calculated at hundreds of millions of yuan or 10,000 yuan, while
labor productivity is only tens to hundreds of yuan; the amount of fisheries
resources is tens of thousands or tens of thousands of tons.
3. Most of the data have a certain degree of randomness, especially the statistical or
observed time series or occasionally measured values, whether it is natural
indicators or economic data, all have random changes, and all have obvious
swings.
4. A large amount of data has a certain degree of gray, and most of the data collected
using the above methods are the average or statistical value of each sample point
in the region, which is not an exact white parameter in time or space but rather an
2 Raw Data Processing Method 23
exact white parameter in time or space. A gray number with upper and lower
limits. For example, in a fishery resource and environmental survey conducted by
a survey ship, the data obtained can only be the data value at a certain point at a
certain time, but due to the limitations of conditions and instruments and equip-
ment, the values will have errors, and the magnitude of this error value cannot be
known, resulting in a gray zone. For example, the amount of precipitation in a
certain area in a certain year is the average of each actual observation record in the
area, and it is impossible to know due to the difference in the measurement
method and the error caused by the time calculation. The same problem also
exists in some economic statistics. Therefore, strictly speaking, most of the
collected data are gray parameters with varying degrees of gray.
2.2 Several Methods of Raw Data Whitening and Initial

Transformation
2.2.1 Collection and Whitening of Raw Data
For the vast majority of gray parameters, whitening or lightening is needed to

increase the whiteness and reduce the grayness. That is, through the continuous
supplementation of information, the gray parameter gradually becomes relatively
close to the actual value. The main methods of data whitening are as follows (Chen
2003, 2023):
1. The multiyear average of the observation station closest to the sampling site was
directly used. For example, the average value of seawater temperature and salinity
can be used for many years or the average value of recent days.
2. According to the contour map of each factor, the interpolation method is used to
calculate its value. For example, for surface water temperature, salinity, seafloor
topography, and other indicators, each representative sample point does not have
ready-made accurate observation values, which can be used in the “iso-temper-
ature contour map,” “iso-salinity map,” and “topographic contour map.” The
whitening value was calculated using the interpolation method.
3. Determine reasonable data based on the actual situation and data characteristics.
For example, the indicators that reflect the quality of the marine environment
include organic matter, nitrogen content, and phosphorus content. When used, it
is impossible to input all the factors into the model. The convenience of calcu-
lation can be attributed to a comprehensive index. That is, the latest data in the
region are used to determine the various factor indicators of each sample point,
and then the transformed data are obtained using the “range transformation”
method, and a comprehensive number can be obtained by adding them. In this
way, the problems of different dimensions and large differences in magnitude
between various factors can be solved.
24 X. Chen
4. Prediction of economic indicators. The classification and zoning of the marine

economy should have relative stability, and the applied comprehensive indicators
should also reflect the dynamic characteristics of the marine economy. Therefore,
when using the corresponding indicator data, it is necessary to first carry out
the development forecast of the indicator data and then based on the forecast. The
values are classified and divided. There are many prediction methods, and the
gray system GM (1, 1) model and the exponential increasing rate model are
commonly used.
5. Relative values with the same weight are used. According to the principles of
statistics, each individual should have equal weights, but in fact, various statis-
tical objects have unequal products and unequal shapes; that is, there is a problem
of unequal weights. For example, the absolute values of indicators such as sea
area, population yield, and output value at various sampling points are very
different, and they are undoubtedly unequal weights. If we use relative numbers
such as the proportion of sea area, population density, average yield per unit, and
per capita yield within a specific range, we can solve the problem of unequal
weight of the same factor at each sample point.
2.2.2 Several Commonly Used Data Transformation Methods
Due to the above characteristics of the original data, there are certain difficulties and
limitations in establishing the mathematical model for statistical analysis. Therefore,
it is necessary to transform the original data according to the type of mathematical
model to be built. The purpose of the transformation is to (1) make the indicator data
as normal as possible; (2) unify the dimensions between the variables; (3) transform
the nonlinear relationship between the two variables into a linear relationship; and
(4) use a new set of independent variables with a small number of indicators to
replace a set of interrelated original variables (Chen 2003, 2023).
Different mathematical models have different requirements for indicator vari-
ables. Most multivariate statistical analyses require that the variables generally
follow a multivariate normal distribution and have consistent dimensions. For
example, discriminant analysis requires the variables to be normally distributed;
regression analysis requires the dependent variables to be normally distributed and
requires a close correlation between the respective variables and the dependent
variables. Cluster analysis requires the dimensions of each variable to be consistent
and independent of each other. Therefore, the data must be transformed in a targeted
manner according to the requirements of the mathematical model.
The commonly used transformation methods mainly include the following (Chen
2003, 2023):
2.2.2.1 Standardized Transformation
The calculation formula is
X ij - X j
X 0ij =
Sj
(i = 1, 2, . . ., the number of N samples; j = 1, 2, . . ., the number of P variables)

where X 0ij is the transformed data; Xij is the original data; X j is the arithmetic mean of
N
X ij
the jth variable, i.e., X j = i=1
N ; Sj is the standard deviation of variable j, that is,
N
ðX ij - X j Þ
2
Sj = i=1
N -1 .
After the transformation, the average value of each variable is 0, and the variance
is 1, showing a standard normal distribution. There is a unified dimension between
the variables, and the degree of correlation between the two variables before and
after the transformation is unchanged. In a geometric sense, the standardized trans-
formation is equivalent to moving the coordinate origin to the position of the center
of gravity (i.e., the average value). The standardized transformation is applicable to
continuous data with different dimensions and different orders of magnitude.
The relevant data in the empirical analysis of the doctoral dissertation “Evaluation
of the Sustainable Utilization of Marine Fishery Resources” by Prof. Chen (2001)
from Shanghai Ocean University are used for illustration. The resource and envi-
ronmental subsystems of the sustainable use system of fishery resources in the East
China Sea from 1978 to 1984 are shown in Table 2.1.where X1 is the trophic level of
the catch, and the unit is level; X2 is the proportion of the yield of high-quality fish in
the marine catch, and the unit is %; X3 is the proportion of the catch of nonselective
fishing gear in the marine catch, X4 is the average fishing yield per unit of motor
fishing vessels, in the unit of ton/vessel; X5 represents the average fishing yield per
tonnage of motor fishing vessels, in the unit of ton/tonnage; and X6 is the average
fishing yield per unit of motorized fishing vessels and nonmotorized fishing vessels
per kilowatt. The average fishing yield of the unit is ton/kilowatt.
Table 2.1 Data of the resource and environment subsystem of the sustainable use system of fishery
resources in the East China Sea (Chen 2001)
Year 1978 1979 1980 1981 1982 1983 1984
X1 2.64 2.72 2.73 2.72 2.64 2.63 2.54
X2 63.19 59.12 46.48 51.06 48.18 38.6 41.03
X3 43.60 41.10 56.90 58.50 62.20 64.50 67.70
X4 69.79 59.45 51.05 43.16 36.68 29.15 24.84
X5 2.61 2.24 1.55 1.48 1.44 1.30 1.26
X6 1.18 1.05 1.04 0.96 0.94 0.88 0.89
26 X. Chen
In the resource and environment subsystems shown in Table 2.1, the units of each
evaluation index are different and therefore need to be initialized. The mean values
and standard deviations of the sequences X1, X2, X3, X4, X5, and X6 were calculated.
X 1 = ðX 1978 þ X 1979 þ . . . þ X 1984 Þ=7 = ð2:64 þ 2:72 þ . . . þ 2:54Þ=7 = 2:66

X 2 = ðX 1978 þ X 1979 þ . . . þ X 1984 Þ=7 = ð63:19 þ 59:12 þ . . . þ 41:03Þ=7 = 49:67
X 3 = ðX 1978 þ X 1979 þ . . . þ X 1984 Þ=7 = ð43:6 þ 41:1 þ . . . þ 67:7Þ=7 = 56:36
X 4 = ðX 1978 þ X 1979 þ ⋯ þ X 1984 Þ=7 = ð69:79 þ 59:45 þ ⋯ þ 24:84Þ=7 = 44:87
X 5 = ðX 1978 þ X 1979 þ . . . þ X 1984 Þ=7 = ð2:61 þ 2:44 þ . . . þ 1:26Þ=7 = 1:70
X 6 = ðX 1978 þ X 1979 þ . . . þ X 1984 Þ=7 = ð1:18 þ 1:05 þ . . . þ 0:89Þ=7 = 0:99
N
2
X ij - X j
i=1 ð2:64 - 2:66Þ2 þ . . . ð2:54 - 2:66Þ2
S1 = = = 0:07
N -1 7-1
N
2
X ij - X j
i=1 ð63:19 - 49:67Þ2 þ . . . ð41:03 - 49:67Þ2
S2 = = = 8:98
N -1 7-1
N
2
X ij - X j
i=1 ð43:6 - 56:36Þ2 þ . . . ð67:7 - 56:36Þ2
S3 = = = 10:24
N -1 7-1
N
2
ðX ij - X j Þ
i=1 ð69:79 - 44:87Þ2 þ . . . ð24:84 - 44:87Þ2
S4 = = = 16:28
N -1 7-1
N
2
X ij - X j
i=1 ð2:61 - 1:7Þ2 þ . . . ð1:26 - 1:7Þ2
S3 = = = 0:52
N -1 7-1
N
2
X ij - X j
i=1 ð1:18 - 0:99Þ2 þ . . . ð0:89 - 0:99Þ2
S3 = = = 0:11
N -1 7-1
Then, the data are transformed into:
X 11 - X 1 2:64 - 2:66
X 011 = = = - 0:29
S1 0:07
Other data transformations are similar.

Table 2.2 Values of each indicator after transformation of the mean and standard deviation (Chen
2003, 2023)
Year 1978 1979 1980 1981 1982 1983 1984
X′1 -0.29 0.86 1.00 0.86 -0.29 -0.43 -1.71
X′2 1.51 1.05 -0.36 0.15 -0.17 -1.23 -0.96
X′3 -1.25 -1.49 0.05 0.21 0.57 0.79 1.11
X′4 1.53 0.90 0.38 -0.11 -0.50 -0.97 -1.23
X′5 1.75 1.05 -0.29 -0.43 -0.51 -0.77 -0.85
X′6 1.71 0.55 0.44 -0.31 -0.50 -1.05 -0.90
The sequences of the transformed resource and environment subsystems are

shown in Table 2.2.
2.2.2.2 Range Transformation
X ij - X j min
X 0ij =
X j max - X j min
where X 0ij is the transformed data; Xij is the original data; Xjmax is the maximum value
of the original data of the jth variable; Xjmin is the minimum value of the original data
of the jth variable.
After range transformation, the data have a unified dimension, with a maximum
value of 1 and a minimum value of 0, and all the data change between 0 and 1. The
degree of correlation between the two variables before and after the transformation is
unchanged, and its geometric meaning is equivalent to moving the coordinate origin
to the minimum value. Range transformation is suitable for the transformation of
continuous raw data with different dimensions and quantities.
The data in Table 2.1 were used for analysis, and the maximum and minimum
values of each indicator were first obtained. They are
X1max = 2.73 X1min = 2.54

X2max = 63.19 X2min = 38.60
X3max = 67.70 X3min = 41.10
X4max = 69.79 X4min = 24.84
X5max = 2.61 X5min = 1.26
X6max = 1.18 X6min = 0.88
Then, the data are transformed into:

28 X. Chen
Table 2.3 Index values after Year 1978 1979 1980 1981 1982 1983 1984
range transformation (Chen
X′1 0.53 0.95 1.00 0.95 0.53 0.47 0.00
2003, 2023)
X′2 1.00 0.83 0.32 0.51 0.39 0.00 0.10
X′3 0.09 0.00 0.59 0.65 0.79 0.88 1.00
X′4 1.00 0.77 0.58 0.41 0.26 0.10 0.00
X′5 1.00 0.73 0.21 0.16 0.13 0.03 0.00
X′6 0.99 0.57 0.53 0.25 0.18 0 0.04
X 11 - X 1 min 2:64 - 2:54

X 011 = = = 0:53
X 1 max - X 1 min 2:73 - 2:54
The rest of the data conversion is similar.

Then, the sequences of the resource and environment subsystems after the range
transformation can be obtained, as shown in Table 2.3.
2.2.2.3 Averaging Transformation
X ij
X 0ij =
Xj
where X 0ij is the transformed data; Xij is the original data; X j is the average of the jth
variable.
The transformed data have a uniform dimension, with values greater than 0 and
concentrated near 1. Its mathematical expectation value is 1, and the expectation
value of the difference between the variable and the mean is 0. This transformation is
applicable to proportional variables such as length, volume, and mass.
Using the data in Table 2.1 as an example for analysis, the average value of each
series is obtained, and the corresponding transformation value is
X 11 2:64
X 011 = = = 0:99
X1 2:66
The rest of the data are similar.

The mean transformation sequence of the resource and environment subsystem
can be obtained, as shown in Table 2.4.
2.2.2.4 Initialization Transformation

mean transformation (Chen
X′1 0.99 1.02 1.03 1.02 0.99 0.99 0.95
2003, 2023)
X′2 1.27 1.19 0.94 1.03 0.97 0.78 0.83
X′3 0.77 0.73 1.01 1.04 1.10 1.14 1.20
X′4 1.56 1.32 1.14 0.96 0.82 0.65 0.55
X′5 1.53 1.32 0.91 0.87 0.84 0.76 0.74
X′6 1.19 1.06 1.05 0.97 0.94 0.88 0.90
initial value transformation
X′1 1.00 1.03 1.03 1.03 1.00 1.00 0.96
(Chen 2003, 2023)
X′2 1.00 0.94 0.74 0.81 0.76 0.61 0.65
X′3 1.00 0.94 1.31 1.34 1.43 1.48 1.55
X′4 1.00 0.85 0.73 0.62 0.53 0.42 0.36
X′5 1.00 0.86 0.59 0.57 0.55 0.50 0.48
X′6 1.00 0.89 0.88 0.81 0.79 0.74 0.76
X ij
X 0ij =
X i1
where X 0ij is the transformed data; Xij is the original data; Xi1 is the initial value of the
ith variable (the first data).
The data after the initial value transformation have a unified dimension, and each
value is a multiple of the initial value, which is convenient for analyzing the
correlation between the series of factors, so it is suitable for processing the statistical
data of socioeconomic aspects.
The data in Table 2.1 are used as an example for analysis, and the above formula
is used for initial value transformation:
X 11 2:64
X 011 = = =1
X 11 2:64
X 12 2:72
X 012 = = = 1:03
X 11 2:64
...
X 17 2:54
X 012 = = = 0:96
X 11 2:64
Other calculations are similar.

The sequence of the initial value transformation of the resource and environment
subsystem can be obtained, as shown in Table 2.5.
30 X. Chen
2.2.2.5 Modular Transformation
j
X 0ij = X ik
k=1
where X 0ij is the transformed data; Xik is the kth data of the jth variable.
This transformation accumulates the time data series once a year to form a new
data series, i.e., generate a time series of numbers. This transformation can be used
for time series forecasting. This is the modeling mechanism and method of gray
system theory for establishing mathematical models, making predictions, and
performing dynamic analysis.
is used for modular processing:
1
X 011 = X 1k = X 11 = 2:64
k=1
2
X 012 = X 1k = X 11 þ X 12 = 2:64 þ 2:72 = 5:36
k=1
3
X 013 = X 1k = X 11 þ X 12 þ X 13 = 2:64 þ 2:72 þ 2:73 = 8:09
k=1
...
7
X 017 = X 1k = X 11 þ X 12 þ . . . þ X 17 = 2:64 þ 2:72 þ . . . þ 2:54 = 18:62
k=1

The sequence of the modularized transformation of the resource and environment
subsystem can be obtained, as shown in Table 2.6.
Table 2.6 Indicator values after modular transformation (Chen 2003, 2023)
Year 1978 1979 1980 1981 1982 1983 1984
X′1 2.64 5.36 8.09 10.81 13.45 16.08 18.62
X′2 63.19 122.31 168.79 219.85 268.03 306.63 347.66
X′3 43.6 84.70 141.60 200.10 262.30 326.80 394.50
X′4 69.79 129.23 180.28 223.43 260.12 289.27 314.11
X′5 2.61 4.85 6.40 7.88 9.31 10.61 11.87
X′6 1.18 2.23 3.27 4.22 5.16 6.03 6.92
2.2.2.6 Moving Average Transformation
X i - 1 þ X i þ X iþ1
Xi =
3
To avoid numerical cycles, the following formula can also be used:
X i - 1 þ 2X i þ X iþ1
Xi =
4
X i - 2 þ X i - 1 þ X i þ X iþ1 þ X iþ2
Or X i =
5
This transformation can weaken the randomness of time data, eliminate the errors
in collecting statistical data to varying degrees, and improve the reliability and
accuracy for further data processing.
is used for moving average transformation processing:
2X 11 þ X 12 2 × 2:64 þ 2:72
X 011 = = = 2:67
3 3
X þ X 12 þ X 13 2:64 þ 2:72 þ 2:73
X 012 = 11 = = 2:70
3 3
X þ X 13 þ X 14 2:72 þ 2:73 þ 2:72
X 013 = 12 = = 2:72
3 3
...
X 16 þ 2 × X 17 2:63 þ 2 × 2:54
X 017 = = = 2:57
3 3

Then, the sequence of the resource and environment subsystem after sliding
transformation can be obtained, as shown in Table 2.7.
Table 2.7 Index values after sliding transformation (Chen 2003, 2023)
Year 1978 1979 1980 1981 1982 1983 1984
X′1 2.67 2.70 2.72 2.70 2.66 2.60 2.57
X′2 61.83 56.26 52.22 48.57 45.95 42.60 40.22
X′3 42.77 47.20 52.17 59.20 61.73 64.80 66.63
X′4 66.34 60.09 51.22 43.63 36.33 30.22 26.27
X′5 2.49 2.13 1.76 1.49 1.40 1.33 1.27
X′6 1.14 1.09 1.01 0.98 0.92 0.90 0.89
32 X. Chen
2.2.2.7 Transformation of the Weakening Operator and Strengthening

Operator
Let X be the original data sequence and D be the buffer operator. When X is the
increasing sequence and the declining sequence, respectively:
1. If the buffer sequence XD has a slower growth rate (or decay rate) or a decrease in
amplitude than the original sequence X, then the buffer operator D is called a
weakening operator.
2. If the growth rate (or decay rate) of buffer sequence XD is faster or the amplitude
increases compared to the original sequence X, then buffer operator D is called a
strengthening operator.
2.2.2.7.1 Weakening Operator Transformation
Let the original data sequence X = (x (1), x (2)..., x (n)).

Let XD = (x(1)d, x(2)d, . . ., x(n)d)
where xðk Þd = n - k þ1 ½xðk Þ
1
þ xðk þ 1Þ þ ⋯ þ xðnÞ; k = 1, 2, . . ., n.
Then, when X is a monotonic increasing sequence, a monotonic decay sequence, or
an oscillation sequence, D is the first-order weakening operator, and XD is the
buffer sequence after the first-order weakening.
If XD2 = XDD = (x(1)d2, x(2)d2, . . ., x(n)d2),
where xðk Þd2 = n - k þ 1 ½ xð k Þ d
1
þ xðk þ 1Þd þ ⋯ þ xðnÞd; k = 1, 2, . . ., n.
Then, D2 is the second-order weakening operator for the monotonic growth, mono-
tonic decay or oscillation sequence, and XD2 is the buffer sequence after the
second-order weakening.
2.2.2.7.2 Enhanced Operator Transformation
Let the original sequence and its buffer sequence be X = (x (1), x (2)..., x (n)), and
XD = (x (1) d, x (2) d..., x (n) d), respectively.
xð1Þþxð2Þþ⋯þxðk - 1Þþkxðk Þ
where xðk Þd = 2k - 1 ; k = 1, 2, . . ., n–1, and x(n)d = x(n).
Then, when X is a monotonically increasing sequence or a monotonically declining
sequence, D is the first-order strengthening operator, and XD is the buffer
sequence after the first-order strengthening.
If XD2 = XDD = (x(1)d2, x(2)d2, . . ., x(n)d2),
xð1Þdþxð2Þdþ⋯þxðk - 1Þdþkxðk Þd
where x (n) d2 = x (n) d = x (n); x(k) d 2 = 2k - 1 ; k = 1,
2, . . ., n–1.
Then, D2 is a second-order strengthening operator for a monotonically increasing

sequence or a monotonically declining sequence, and XD2 is the buffer sequence
after the second-order strengthening.
Taking the fishery output data of a certain district in Zhejiang Province from 1983 to
1986 as an example for analysis, X = (10,155, 12,588, 23,480, 35,388), the unit is
100 yuan, and its growth rate of 51.6% is very strong every year from 1983 to 1986.
The average annual growth rate reached 67.7%, especially from 1984 to 1986.
However, due to the limitation of fishery resources and the limitation of fishing
capacity and technology, its fishery output value cannot continue to grow indefi-
nitely, and it is impossible to maintain such a high development speed in the future.
If the existing numerical values are used to directly model the prediction, the
prediction results are simply unacceptable. After analysis and discussion, it is
believed that the high growth rate is mainly due to the low base, while the reason
for the low base is the insufficient utilization of fishery resources in the past.
Therefore, in the forecast of fishery output in the future, it is necessary to adopt
the weakening sequence growth trend and introduce the second-order weakening
operator. The specific calculation process is as follows:
1
xð1Þd = × ð10155 þ 12588 þ 23480 þ 35388Þ = 20403
4-1 þ 1
1
xð2Þd = × ð12588 þ 23480 þ 35388Þ = 23819
4-2 þ 1
1
xð3Þd = × ð23480 þ 35388Þ = 29434
4-3 þ 1
1
xð4Þd = × 35388 = 35388
4-4 þ 1
Then, the first-order buffer sequence XD = (20403, 23819, 29434, 35388).
1
xð1Þd2 = × ð20403 þ 23819 þ 29434 þ 35388Þ = 27261
4-1 þ 1
1
xð2Þd2 = × ð23819 þ 29434 þ 35388Þ = 29547
4-2 þ 1
1
xð3Þd2 = × ð29434 þ 35388Þ = 32411
4-3 þ 1
1
xð4Þd 2 = × 35388 = 35388
4-4 þ 1
Then, the second-order buffer sequence XD2 = (27261, 29547, 32411, 35388) is
obtained. The GM (1, 1) model established using the second-order buffer sequence
XD2 shows that the average annual increase of 9.4% in the fishery output value from
1986 to 2000 is basically acceptable and consistent with the actual situation.
34 X. Chen
References
Chen XJ (2001) Sustainable utilization assessment of marine fisheries resources[D]. Nanjing

Agricultural University, Nanjing. (In Chinese)
Chen XJ (2003) Application of Gray system theory in fishery science. China Agricultural Press.
(In Chinese)
(In Chinese)
Chapter 3
Gray Correlation Analysis
Xinjun Chen
Abstract Correlation analysis is a simple and practical analysis technique, which is

to find the association or correlation existing in a large number of data sets and
describe the law and pattern that some attributes appear at the same time in a thing.
There are many methods of correlation analysis, among which the gray correlation is
one of the common methods. In the real natural system and social economy system,
there are many gray systems, and there are many subsystems (factors) that affect the
gray system; therefore, the gray system theory puts forward the concept of the gray
correlation degree analysis to each subsystem and intends to analyze the gray
correlation degree of each subsystem through certain methods, to seek the numerical
relations among the subsystems (or factors) in the system, this is the basic concept of
the gray correlation degree. Gray correlation analysis provides a quantitative mea-
sure for the development of a system, which is very suitable for dynamic process
analysis. The concrete calculation steps of gray correlation analysis are as follows:
(1) determining the reference series and the comparison series that influence the
system behavior, (2) dealing with the reference series and the comparison series
dimensionless, (3) calculating the gray correlation coefficient between the reference
series and the comparison series, (4) calculating the degree of correlation, (5) ranking
the degree of correlation. In this chapter, first, the concept and characteristics of gray
correlation are introduced, second, several common calculation methods of gray
correlation degree are described, it includes general calculation method, gray abso-
lute correlation degree, gray relative correlation degree, and gray comprehensive
correlation degree. Finally, the application of gray correlation in fishery science is
analyzed. The application cases in fishery science mainly include (1) adjustment and
analysis of industrial structure of fishery economy, (2) assessment of fishery
resources, (3) evaluation of sustainable utilization of fishery resources, (4) evalua-
tion of influencing factors of fishery water quality, (5) the application of aquaculture
(such as food selection), and (6) the evaluation of fish growth model.
X. Chen (✉)
36 X. Chen
Keywords Gray correlation · Gray relative correlation · Gray absolute correlation ·

Gray comprehensive correlation
3.1 The Concept of Gray Correlation and Its Application
There are many large and small systems in the objective world, which are composed
of many factors. The relationship between these systems and the internal factors of
the system is very complex. In particular, the randomness of changes in superficial
phenomena tends to confuse people’s intuition and obscure the essence of things,
making it difficult for people to obtain sufficient and comprehensive information
when understanding, analyzing, predicting, and making decisions, and it is difficult
to form a clear concept. Therefore, we believe that the relationship between different
systems is gray, and the relationship between various factors in the system is also
gray. It is difficult to identify the main contradictions and find the main factors when
it is not clear which factors are closely related and which factors are not closely
related. Therefore, gray system theory proposes the concept of correlation analysis.
The purpose is to analyze the main relationship between various factors in the system
through a certain method, to identify the most important factors affecting the system
and to grasp the main aspects of the contradiction. For example, which industry has
the most significant impact on the composition of the total fishery output, thereby
creating conditions for the healthy development of the fishery production system.
The measure of the correlation between two systems or two factors is called
the degree of correlation. It describes the relative changes between factors in the
development process of the system, that is, the relativity of indicators such as the
magnitude, direction, and speed of changes. The basic idea of gray correlation
analysis is to determine whether the relationship is close according to the degree
of similarity of the geometric shapes of the sequence curves. If the relative changes
of the two are basically the same in the process of system development, the closer the
curves are, the greater the degree of correlation between the two is considered;
otherwise, the degree of correlation between the two is smaller. Gray correlation
analysis is a quantitative description and comparison of the development and change
of a system. Only by clarifying the correlation between systems or factors we can
have a more thorough understanding of the system and distinguish which are the
dominant factors, which are the potential factors, which are the advantages, and
which are the disadvantages. Therefore, when analyzing and studying a gray system,
it is necessary to first determine how to find the correlation from the random time
series and calculate the correlation degree to provide a basis for factor discrimina-
tion, advantage analysis, and prediction accuracy testing and lay a good foundation
for system decision-making. Therefore, the correlation analysis between gray factors
is essentially the basis of gray system analysis, prediction, and decision-making.
The correlation analysis of gray system theory is different from the correlation
analysis of mathematical statistics, which is mainly manifested in the following
aspects. First, their theoretical bases are different. Correlation analysis is based on
the gray process of the gray system, while correlation analysis is based on the
3 Gray Correlation Analysis 37
stochastic process of probability theory. Second, the analysis method is different.

Correlation analysis is a comparison of time series between factors, while correlation
analysis is a comparison of groups of factors. Third, the data volume requirements
are different. Association analysis does not require too much data, while correlation
analysis requires sufficient data. Fourth, the research focuses are different. Correla-
tion analysis mainly studies dynamic processes, while correlation analysis mainly
focuses on static studies. Therefore, the degree of relevance analysis is more
adaptable and unique in the socioeconomic system.
In addition, regression analysis, analysis of variance, and principal component
analysis in mathematical statistics are all methods for systematic analysis. These
methods have the following shortcomings: (1) a large amount of data is needed, and
it is difficult to find a statistical law with a small amount of data; (2) a sample is
required to follow a typical probability distribution, and a linear relationship between
the data of various factors and the data of the system characteristics is needed. The
factors are unrelated to each other. This requirement is often difficult to meet.
(3) There may be a phenomenon in which the quantitative results are inconsistent
with the qualitative analysis, resulting in the distortion and inversion of the relation-
ships and laws of the system. The gray relational analysis method makes up for the
shortcomings caused by the use of mathematical statistical methods for systematic
analysis. It is equally applicable to large sample sizes and whether there are patterns
in the sample. Moreover, the calculation is very convenient, and there is no discrep-
ancy between the quantitative results and the qualitative analysis results.
At present, the application of gray correlation analysis is very extensive, and it
has almost penetrated into various fields of social and natural sciences, such as
agriculture, fisheries, education, health, politics and law, environmental protection,
military, geography, geology, petroleum, hydrology, meteorology, etc. In the field of
social economy, good results have been achieved.
3.2 Several Calculation Methods of Gray Correlation
In this section, several commonly used methods for calculating the degree of gray
correlation are introduced, including the general calculation method, the absolute
degree of gray correlation, the degree of gray relative correlation, and the degree of
comprehensive gray correlation (Deng 1987, 1990; Liu et al. 2014).
3.2.1 General Calculation Method
Now, assume that the system behavior sequence

38 X. Chen
X 0 = ðx0 ð1Þ, x0 ð2Þ . . ., x0 ðnÞÞ

X 1 = ðx1 ð1Þ, x1 ð2Þ . . ., x1 ðnÞÞ
......
X i = ðxi ð1Þ, xi ð2Þ . . ., xi ðnÞÞ
......
X m = ðxm ð1Þ, xm ð2Þ . . ., xm ðnÞÞ
where X0 is the parent sequence, Xi is the subsequence, and xi (k) is the observation
data of factor xi at time k.
The calculation of the gray correlation degree generally includes the following
steps: (1) transformation of the original data; (2) calculation of the difference
sequence; (3) calculation of the maximum and minimum difference between the
two poles; (4) calculation of the correlation coefficient; and (5) calculation of the
gray correlation (Deng 1987, 1990; Liu et al. 2014). The details are as follows:
Step 1: Raw data transformation
Because the dimension (or unit) of each factor in the system is not necessarily the
same, for example, the labor force is a person, the output value is 10,000 yuan, the
output is tons, etc., sometimes the magnitude of the value is different, such as the per
capita income of several hundred yuan and the grain yield per hectare. The cost is
several thousand kilograms, the output value of some industries reaches tens of
billions, and the output value of some industries is only tens of thousands of yuan.
Such data are often difficult to directly compare, and their geometric curve ratios are
also different. Therefore, it is necessary to eliminate the dimensions (or units) of the
original data and convert them into a comparable data series. See Chap. 2 for the
transformation and processing methods of the original data.
Taking the initial value transformation as an example, we have
X 0i = X i =xi ð1Þ = x0i ð1Þ, x0i ð2Þ, ⋯, x0i ðnÞ , I = 0, 1, 2, . . . , m;
Step 2: Find the difference sequence. Remember
Δi ðkÞ = x00 ðk Þ - x0i ðkÞ

Δi = ðΔi ð1Þ, Δi ð2Þ, ⋯, Δi ðnÞÞ
k = 1, 2, . . . , n; i = 0, 1, 2, . . . , m;
Step 3: Find the maximum difference and minimum difference between the two
poles. Remember
M = max max Δi ðkÞ

i k
Table 3.1 Raw data (Chen 2003, 2023)

Serial number 1 2 3 4 5 6 7
X0 2.64 2.72 2.73 2.72 2.64 2.63 2.54
X1 63.19 59.12 46.48 51.06 48.18 38.6 41.03
X2 43.60 41.10 56.90 58.50 62.20 64.50 67.70
X3 69.79 59.45 51.05 43.16 36.68 29.15 24.84
X4 2.61 2.24 1.55 1.48 1.44 1.30 1.26
X5 1.18 1.05 1.04 0.96 0.94 0.88 0.89
M = min min Δi ðk Þ
i k
Step 4: Find the correlation coefficient.

The parent sequence of the data transformation is {X0(t)}, and the subsequence is
{Xi(t)}. Then, the correlation coefficient L0i(k) between the parent sequence {X0(k)}
and the subsequence {Xi(t)} can be calculated using the following equation:
m þ ξM
γ 0i ðkÞ =
Δi ðkÞ þ ξM
where ξ 2 (0, 1) is the resolution coefficient, k = 1, 2, . . ., n; i = 1, 2, . . ., m.

The correlation coefficient reflects the closeness (closeness) of the two compared
sequences at a certain moment. At Δmin, γio = 1, and at Δmax, the correlation
coefficient is the minimum value. Therefore, the range of the correlation coefficient
is 0 < γ ≤ 1.
Step 5: Calculate the degree of gray correlation.
n
1
r 0i = γ 0i ðk Þ; k = 1, 2, . . . , n; i = 1, 2, . . . , m;
n k=1
If the weights at different moments are inconsistent, the gray relational degree can
be defined as:
n
r0i = W k r 0i ðkÞ
k-1
k = 1, 2, . . . ; ni = 1, 2, . . . , m;
n
where W k = 1:
k=1
Assume that there are parent sequence X0 and subsequences X1, X2, X3, X4 and X5,
and assume that the resolution coefficient is 0.5 and the weights are the same
(Table 3.1).
Step 1: Find the initial value sequence
40 X. Chen
Table 3.2 Data after initialization (Chen 2003, 2023)

X 00 1.00 1.03 1.03 1.03 1.00 1.00 0.96
X 01 1.00 0.94 0.74 0.81 0.76 0.61 0.65
X 02 1.00 0.94 1.31 1.34 1.43 1.48 1.55
X 03 1.00 0.85 0.73 0.62 0.53 0.42 0.36
X 04 1.00 0.86 0.59 0.57 0.55 0.50 0.48
X 05 1.00 0.89 0.88 0.81 0.79 0.74 0.76
Table 3.3 Values after difference series processing (Chen 2003, 2023)
Δ1 0.00 0.09 0.29 0.22 0.24 0.39 0.31
Δ2 0.00 0.09 0.28 0.31 0.43 0.48 0.59
Δ3 0.00 0.18 0.30 0.41 0.47 0.58 0.60
Δ4 0.00 0.17 0.44 0.46 0.45 0.50 0.48
Δ5 0.00 0.14 0.15 0.22 0.21 0.26 0.20
The initial value of each sequence is transformed, and the initial value is selected
as the denominator for transformation. The transformed data is calculated by the
following equation:
X 0i = X i =xi ð1Þ = x0i ð1Þ, x0i ð2Þ, ⋯, x0i ð7Þ

i = 0, 1, 2, . . . , 5:
The transformed data are shown in Table 3.2.

Step 2: Find the difference sequence
By Δi ðkÞ = x00 ðkÞ - x0i ðk Þ , i = 1, 2, . . . , 5, and we have Table 3.3.
Step 3: Find the difference between the two poles
M = max max Δi ðkÞ = 0:60

i k
m = min min Δi ðk Þ = 0
i k
Step 4: Calculate the correlation coefficient

Take ξ = 0.5,
m þ ξM 0:30
γ 0i ðkÞ = = ; i = 1, 2, ...5
Δi ðk Þ þ ξM Δi ðk Þ þ 0:30
Thus, Table 3.4 is obtained.

Step 5: Calculate the degree of gray correlation
Table 3.4 Correlation coefficient (Chen 2003, 2023)

γ1 1.00 0.76 0.50 0.57 0.56 0.44 0.49
γ2 1.00 0.77 0.52 0.49 0.41 0.38 0.34
γ3 1.00 0.63 0.50 0.42 0.39 0.34 0.33
γ4 1.00 0.64 0.41 0.39 0.40 0.37 0.39
γ5 1.00 0.68 0.67 0.58 0.59 0.54 0.60
7
1
r 01 = γ 01 ðkÞ = 0:62
7 k=1
7
1
r 02 = γ 02 ðkÞ = 0:56
7 k=1
7
1
r 03 = γ 03 ðkÞ = 0:52
7 k=1
7
1
r 04 = γ 04 ðkÞ = 0:51
7 k=1
7
1
r 05 = γ 05 ðkÞ = 0:67
7 k=1
3.2.2 Absolute Degree of Gray Correlation
Assuming that the parent sequence {X0} and the subsequence {Xi} have the same
length, they are
X 0 = ðx0 ð1Þ, x0 ð2Þ . . ., x0 ðnÞÞ

X i = ðxi ð1Þ, xi ð2Þ . . ., xi ðnÞÞ
Then, the corresponding starting point annihilation sequence is
X 00 = ðx0 ð1Þ, x0 ð2Þ, . . . , x0 ðnÞÞ

X 0i = ðxi ð1Þ, xi ð2Þ, . . . , xi ðnÞÞ
where X 00 ðk Þ = x0 ðkÞ - x0 ð1Þ

42 X. Chen
X 0i ðk Þ = xi ðkÞ - xi ð1Þ
Then, the formula for calculating the absolute gray correlation between X0 and Xi
is
1 þ js0 j þ jsi j
ε0i =
1 þ js0 j þ jsi j þ jsi - s0 j
where
n-1
1
j s0 j = x00 ðkÞ þ x00 ðnÞ
k=2
2
n-1
1
j si j = x0i ðk Þ þ x0i ðnÞ
k=2
2
n-1
1 0
jsi- s0 j = x0i ðkÞ - x00 ðkÞ þ x ðnÞ - x00 ðnÞ
k=2
2 i
The gray absolute correlation degree ε0i has the following properties (Liu et al.
2014):
1. 0 < ε0i ≤ 1;
2. ε0i is only related to the geometric shapes of X0 and Xi and has nothing to do with
their relative spatial positions; in other words, the translation does not change the
magnitude of the absolute correlation degree;
3. Any two sequences are not absolutely unrelated, that is, ε0i is always nonzero;
4. The greater the degree of geometric similarity between Xi and X0, the greater ε0i;
5. When any observation data in X0 change, ε0i will change accordingly;
6. When the lengths of X0 and Xi change, ε0i also changes;
7. ε0i = εi0.
Assume that there are parent sequence X0 and subsequences X1, X2, X3, X4, and X5
(Table 3.5).
Step 1: Zeroing the starting point
By X 0i ðkÞ = xi ðkÞ - xi ð1Þ Available;
x00 ð1Þ = x0 ð1Þ - x0 ð1Þ = 0

x00 ð2Þ = x0 ð2Þ - x0 ð1Þ = 2:72 - 2:64 = 0:08
Similarly, other zeroing values can be obtained, as shown in Table 3.6.

Step 2: Seeking |s0|, |si| and |si - s0|
Table 3.5 Raw data (Chen 2003, 2023)

X0 2.64 2.72 2.73 2.72 2.64 2.63 2.54
X1 63.19 59.12 46.48 51.06 48.18 38.6 41.03
X2 43.60 41.10 56.90 58.50 62.20 64.50 67.70
X3 69.79 59.45 51.05 43.16 36.68 29.15 24.84
X4 2.61 2.24 1.55 1.45 1.44 1.30 1.26
X5 1.18 1.05 1.04 0.96 0.94 0.88 0.89
Table 3.6 Data with initial values after zeroing (Chen 2003, 2023)
X 00 0.00 0.08 0.09 0.08 0.00 -0.01 -0.10
X 01 0.00 -4.07 -16.71 -12.13 -15.01 -24.59 -22.16
X 02 0.00 -2.50 13.30 14.90 18.60 20.90 24.10
X 03 0.00 -10.34 -18.74 -26.63 -33.10 -40.63 -44.95
X 04 0.00 -0.37 -1.06 -1.13 -1.17 -1.31 -1.35
X 05 0.00 -0.13 -0.14 -0.22 -0.24 -0.30 -0.29
6
1
js0 j = x00 ðkÞ þ x00 ð7Þ = 0:19
k=2
2
6
1
js1 j = x01 ðkÞ þ x01 ð7Þ = 83:59
k=2
2
6
1
js2 j = x02 ðkÞ þ x02 ð7Þ = 77:25
k=2
2
6
1
js3 j = x03 ðkÞ þ x03 ð7Þ = 151:92
k=2
2
6
1
js4 j = x04 ðkÞ þ x04 ð7Þ = 5:72
k=2
2
6
1
js5 j = x05 ðkÞ þ x05 ð7Þ = 1:18
k=2
2
6
1 0
js1- s0 j = x01 ðkÞ - x00 ðkÞ þ x ð7Þ - x00 ð7Þ = 83:78
k=2
2 1
44 X. Chen
6
1 0
k=2
2 2
6
1 0
k=2
2 3
6
1 0
k=2
2 4
6
1 0
k=2
2 5
Step 3: Find the absolute degree of association
1 þ js0 j þ js1 j 1 þ 0:19 þ 83:59

ε01 = = = 0:50
1 þ js0 j þ js1 j þ js1 - s0 j 1 þ 0:19 þ 83:59 þ 83:78
Similarly, we can obtain
ε02 = 0:50
ε03 = 0:50
ε04 = 0:54
ε05 = 0:63
3.2.3 Gray Relative Degree of Relevance
length and the initial value is not equal to zero, then their initial values are
X 0i = X i =xi ð1Þ
X 00 = X 0 =x0 ð1Þ
Then, the corresponding initial value sequence is
X 00 = x00 ð1Þ, x00 ð2Þ, ⋯, x00 ðnÞ

X 0i = x0i ð1Þ, x0i ð2Þ, ⋯, x0i ðnÞ
The gray relative degree of X0 and Xi is
1 þ s00 þ s0i
r 0i =
1 þ s00 þ s0i þ s0i - s00
where
n-1
1
s00 = x00 ðkÞ þ x00 ðnÞ
k=2
2
n-1
1
s0i = x0i ðkÞ þ x0i ðnÞ
k=2
2
n-1
1 0
s0i - s00 = x0i ðkÞ - x00 ðkÞ þ x ðnÞ - x00 ðnÞ
k=2
2 i
The gray relative degree of X0 and Xi is r0i. The gray relative degree r0i has the
following properties (Liu et al. 2014):
1. 0 < r0i ≤ 1;
2. r0i is only related to the rate of change of the sequence X0 and Xi relative to the
starting point and is not related to the size of each observation data.
3. The rate of change of any two sequences is not unrelated, i.e., r0i is always
nonzero;
4. The more consistent the rate of change of X0 and Xi relative to the starting point is,
the greater r0i is;
5. If any observation data in X0 or Xi are changed, r0i will change accordingly; if the
sequence length changes, r0i will also change;
6. r0i = ri0.
Using the parent sequence X0 and the subsequences X1, X2, X3, X4, and X5 in the
above example as the original data, the gray relative degree of correlation between
the parent sequence and the individual subsequences is determined.
Step 1: Initialize the sequence.
From formula X 0i = X i =xi ð1Þ, the initial value sequence is obtained, as shown in
Table 3.7:
Step 2: Seeking s00 , s0i and s0i - s00
46 X. Chen
Table 3.7 Data after initialization (Chen 2003, 2023)

X 00 1.00 1.03 1.03 1.03 1.00 1.00 0.96
X 01 1.00 0.94 0.74 0.81 0.76 0.61 0.65
X 02 1.00 0.94 1.31 1.34 1.43 1.48 1.55
X 03 1.00 0.85 0.73 0.62 0.53 0.42 0.36
X 04 1.00 0.86 0.59 0.57 0.55 0.50 0.48
X 05 1.00 0.89 0.88 0.81 0.79 0.74 0.76
6
1
s00 = x00 ðkÞ þ x00 ð7Þ = 5:57
k=2
2
61
1
s01 = x01 ðkÞ þ x01 ð7Þ = 4:18
k=2
2
s02 = 7:27
s03 = 3:32
s04 = 3:31
s05 = 4:50
6
1 0
s01 - s00 = x01 ðk Þ - x00 ðk Þ þ x ð7Þ - x00 ð7Þ = 3:99
k=2
2 1
s02 - s00 = 7:08
s03 - s00 = 3:13
s04 - s00 = 3:12
s05 - s00 = 4:31
Step 3: Find the relative degree of gray correlation
1 þ s00 þ s01 1 þ 5:57 þ 4:18

r01 = = = 0:57
0 0 0
1 þ s0 þ s1 þ s1 - s00 1 þ 5:57 þ 4:18 þ 3:99
Similarly, we can obtain
r 02 = 0:54
r 03 = 0:59
r 04 = 0:59
r 05 = 0:57
3.2.4 Gray Comprehensive Correlation
length, and the initial values are not equal to zero, ε0i and r0i are the gray absolute
and relative degrees of correlation between {X0} and {Xi}, respectively, θ 2 [0, 1],
then ρ0i = θε0i + (1 - θ)r0i is the gray comprehensive correlation between X0 and Xi.
The gray comprehensive correlation degree not only reflects the degree of
similarity of the polyline but also reflects the closeness of the change rate of X0
and Xi relative to the starting point and is a quantitative indicator that more compre-
hensively describes whether the sequences are close. Generally, take θ = 0.5.
The gray comprehensive correlation degree ρ0i has the following properties (Liu
et al. 2014):
1. 0 < ρ0i ≤ 1;
2. ρ0i is not only related to the size of each observation data of series X0 and Xi but
also related to the rate of change of each data relative to the starting point;
3. ρ0i is always nonzero;
4. When the data in X0 and Xi are changed, ρ0i will also change accordingly;
5. When the sequence length of X0 and Xi changes, ρ0i also changes;
6. When θ takes different values, ρ0i is also different;
7. ρ0i = ρi0.
Using the above example to calculate the gray comprehensive relevance, we take
θ = 0.5; then, the corresponding comprehensive correlation degrees are
ρ01 = θε01 þ ð1- θÞr 01 = 0:5 × 0:5 þ ð1- 0:5Þ × 0:59 = 0:545
Similarly, we obtain
ρ02 = 0:52
ρ03 = 0:55
ρ04 = 0:57
ρ04 = 0:60
48 X. Chen
3.3 Application Examples of Gray Relational Theory

in Fishery Science
In this section, the application of the gray correlation method in fishery science and
its case analysis are mainly described, and good research results have been obtained
in the aspects of industrial structure analysis, resource fishing ground analysis, and
fishery biology research. The main contents include (1) fishery economic industrial
structure adjustment and analysis; (2) fishery resource assessment; (3) fishery
resource sustainable utilization assessment; (4) influencing factor assessment in the
water quality of fisheries; (5) application in the aquaculture industry; and (6) basic
biological evaluation of fish growth models. Detailed analysis is now carried out
based on relevant examples (Chen 2003, 2023).
3.3.1 Application of Gray Correlation in the Analysis

of Fishery Industry Structure
In the paper “Analysis of Fishery Production Structure in China,” Chen and Zhou
(2002b) analyzed the fishery production structure in China in the past 50 years
through the correlation method in gray theory and explored the change process of
fishery production and its contribution to fishery development. Identify the problems
in the development of China’s fisheries and the factors that restrict development and
provide a basis for decision-making for the sustainable development of China’s
fisheries.
The data are from the China Fisheries Statistical Yearbook (1949–1997). The data
items include the total production of aquatic products, the production of seawater,
the production of freshwater, the production of seawater fishing and aquaculture, the
production of freshwater fishing and aquaculture, and the fish and shrimp of seawa-
ter and freshwater. Crabs, shellfish, algae, etc., and some major fishing and farming
species. According to the development of China’s fisheries, the analysis was
conducted in three time periods: 1954–1977, 1978–1984, and 1985–1997. In this
study, the general gray correlation degree calculation method was used, and the
resolution coefficient was set to 0.5. The main analysis results are as follows:
3.3.1.1 Correlation Analysis of the Total Aquatic Production

with the Production of Seawater and Freshwater
3.3.1.1.1 Correlation Analysis of the Total Aquatic Production

and the Production of Seawater and Freshwater Between 1954
and 1977
Using the total production of aquatic products as the mother series, the correlation
between the total production of aquatic products (X0), the production of seawater
(Xs), and the production of freshwater (Xf) from 1954 to 1977 was analyzed
(Table 3.8).
The correlations between the total production of aquatic products and the pro-
duction of seawater and freshwater are as follows:
r 0s54–77 = 0:7122
r0f 54–77 = 0:6177
3.3.1.1.2 Correlation Analysis of the Total Aquatic Production

and the Production of Seawater and Freshwater in 1978–1984
and 1985–1997
Similarly, the correlation coefficients between the total production of aquatic prod-
ucts and the production of seawater and freshwater in 1978–1984 and 1985–1997
can be obtained, and the results are shown in Tables 3.9 and 3.10.
r 0s78–84 = 0:8163
Table 3.8 Correlation coefficients between the total production of aquatic products and the
production of seawater and freshwater from 1954 to 1977 (Chen and Zhou 2002b)
Year 1954 1955 1956 1957 1958 1959 1960 1961
r0s 1.00 0.824 0.858 0.931 0.999 0.972 0.960 0.956
r0f 1.00 0.752 0.796 0.897 0.999 0.953 0.940 0.584
Year 1962 1963 1964 1965 1966 1967 1968 1969
r0s 0.842 0.773 0.771 0.746 0.668 0.636 0.685 0.675
r0f 0.775 0.687 0.685 0.655 0.565 0.531 0.584 0.573
Year 1970 1971 1972 1973 1974 1975 1976 1977
r0s 0.631 0.576 0.505 0.531 0.476 0.471 0.458 0.436
r0f 0.525 0.467 0.397 0.422 0.369 0.365 0.353 0.333
50 X. Chen
production of seawater and freshwater from 1978 to 1984 (Chen and Zhou 2002b)
Year 1978 1979 1980 1981 1982 1983 1984
r0s 1.0000 0.9127 0.8681 0.8142 0.7857 0.7044 0.6293
r0f 1.0000 0.7548 0.6597 0.5635 0.5192 0.4124 0.3333
production of seawater and freshwater between 1985 and 1997 (Chen and Zhou 2002b)
Year 1985 1986 1987 1988 1989 1990 1991
r0s 1.000 0.745 0.680 0.606 0.641 0.653 0.882
r0f 1.000 0.666 0.591 0.534 0.549 0.562 0.835
Year 1992 1993 1994 1995 1996 1997
r0s 0.868 0.798 0.546 0.423 0.438 0.572
r0f 0.818 0.729 0.450 0.333 0.347 0.476
r 0f 78–84 = 0:6061
r 0s85–97 = 0:6814
r0f 85–97 = 0:6072
3.3.1.2 Correlation Analysis of Seawater and Freshwater Production

and Fishing and Aquaculture Production
3.3.1.2.1 Correlation Analysis of Seawater Production and Fishing

and Aquaculture Production
Using seawater production as the parent sequence and fishing and aquaculture
production as the subsequences, the correlation coefficients between seawater pro-
duction and marine fishing and mariculture production in the three time periods of
1954–1977, 1978–1984 and 1985–1997 were obtained (Tables 3.11, 3.12, and
3.13).
The correlations between seawater production and the production of marine
fishing and mariculture are as follows:
r sf54–77 = 0:9769
rsa54–77 = 0:7830
Table 3.11 Correlation coefficients between seawater production and marine fishing and maricul-
ture production from 1954 to 1977 (Chen and Zhou 2002b)
Year 1954 1955 1956 1957 1958 1959 1960 1961
rsf 1.0000 0.9983 0.9711 1.0000 0.9840 0.9917 0.9979 0.9996
rsa 1.0000 0.9747 0.6931 0.9996 0.8050 0.8896 0.9702 0.9935
Year 1962 1963 1964 1965 1966 1967 1968 1969
rsf 0.9958 0.9866 0.9716 0.9846 0.9883 0.9989 0.9841 0.9818
rsa 0.9404 0.8319 0.6972 0.8117 0.8503 0.9841 0.8066 0.7839
Year 1970 1971 1972 1973 1974 1975 1976 1977
rsf 0.9733 0.9553 0.9527 0.9751 0.9710 0.9554 0.9467 0.8814
rsa 0.7102 0.5897 0.5753 0.7250 0.6926 0.5906 0.5444 0.3333
Table 3.12 Correlation coefficients of seawater production and marine fishing and mariculture
production from 1978 to 1984 (Chen and Zhou 2002b)
Year 1978 1979 1980 1981 1982 1983 1984
rsf 1.0000 0.9673 0.9321 0.9041 0.9181 0.8458 0.7777
rsa 1.0000 0.8089 0.6623 0.5740 0.6157 0.4394 0.3333
Table 3.13 Correlation coefficients of seawater production and marine fishing and mariculture
production from 1985 to 1997 (Chen and Zhou 2002b)
Year 1985 1986 1987 1988 1989 1990 1991
rsf 1.0000 0.9951 0.9838 0.9627 0.9579 0.9615 0.9497
rsa 1.0000 0.9765 0.9253 0.8405 0.8232 0.8362 0.7943
Year 1992 1993 1994 1995 1996 1997
rsf 0.9248 0.8912 0.8845 0.8601 0.7098 0.7101
rsa 0.7154 0.6261 0.6101 0.5568 0.3333 0.3337

r sf78–84 = 0:9064
rsa78–84 = 0:6334

r sf85–97 = 0:9070
rsa85–97 = 0:7209
52 X. Chen
3.3.1.2.2 Correlation Analysis of Freshwater Production and Fishing

and Aquaculture Production
Using freshwater production as the parent sequence and freshwater fishing and
aquaculture production as the subsequences, the correlation coefficients between
freshwater production and fishing and aquaculture production in the three time
periods of 1954–1977, 1978–1984, and 1985–1997 were obtained (Tables 3.14,
3.15, and 3.16).
The correlations between freshwater production and freshwater fishing and fresh-
water aquaculture production are as follows:
rff54–77 = 0:6971
r fa54–77 = 0:5226
Table 3.14 Correlation coefficients between freshwater production and the yields of freshwater
catch and freshwater cultivation from 1954 to 1977 (Chen and Zhou 2002b)
Year 1954 1955 1956 1957 1958 1959 1960 1961
rff 1.0000 0.9021 0.9099 0.7086 0.6959 0.6929 0.7784 0.6309
rfa 1.0000 0.8015 0.8157 0.5159 0.5007 0.4971 0.6061 0.4282
Year 1962 1963 1964 1965 1966 1967 1968 1969
rff 0.8706 0.8397 0.8096 0.6943 0.6702 0.6751 0.6690 0.6326
rfa 0.7466 0.6966 0.6507 0.4988 0.4710 0.4765 0.4696 0.4300
Year 1970 1971 1972 1973 1974 1975 1976 1977
rff 0.6170 0.5959 0.5933 0.5867 0.5538 0.5352 0.5408 0.5282
rfa 0.4138 0.3925 0.3899 0.3834 0.3523 0.3354 0.3403 0.3291
Table 3.15 Correlation coefficients between freshwater production and the yields of freshwater
Year 1978 1979 1980 1981 1982 1983 1984
rff 1.0000 0.9063 0.9165 0.7917 0.5368 0.4819 0.3333
rfa 1.0000 0.9613 0.9658 0.9072 0.7488 0.7052 0.5625
Table 3.16 Correlation coefficients of freshwater production and the production of freshwater
Year 1985 1986 1987 1988 1989 1990 1991
rff 1.000 0.8311 0.7289 0.7016 0.7474 0.7223 0.9630
rfa 1.000 0.961 0.9309 0.9217 0.9368 0.9287 0.9924
Year 1992 1993 1994 1995 1996 1997
rff 0.6381 0.5137 0.4070 0.3654 0.4029 0.3333
rfa 0.8983 0.8410 0.7746 0.7425 0.7716 0.7146
r ff78 - 84 = 0:7095
r fa78 - 84 = 0:8358
r ff85–97 = 0:6427
r fa85–97 = 0:8780
3.3.1.3 Correlation Analysis of the Production of Seawater

and Freshwater and the Production of Each Major Species
3.3.1.3.1 Correlation Analysis of Seawater Production and the Production

of Each Major Species
Using the seawater yield as the mother sequence and the yield of each major species
as the subsequence, the correlations between the seawater yield and the marine fish,
shrimp and crabs, shellfish and algae from 1954 to 1977 are as follows:
r sf54–77 = 0:6918
r ssc54–77 = 0:6963
r sc54–77 = 0:6781
r sa54–77 = 0:6740
The correlations between seawater production and fish, shrimp and crabs, and
shellfish and algae from 1978 to 1984 are as follows:
r sf78–84 = 0:8099
r ssc78–84 = 0:8450
r sc78–84 = 0:7750
r sa78–84 = 0:5880
The correlations between seawater production and fish, shrimp and crabs, and
shellfish and algae from 1985 to 1997 are as follows:
r sf85–97 = 0:9345
54 X. Chen
r ssc85–97 = 0:9604
r sc85–97 = 0:9549
r sa85–97 = 0:7170
3.3.1.3.2 Correlation Analysis of Freshwater Production and the Production

of Major Species
Using freshwater production as the parent sequence and the production of each
major freshwater species as the subsequence, the correlations between freshwater
production and freshwater fish, shrimp and shellfish in 1954–1977 are as follows:
rff54–77 = 0:6063
r fsc54–77 = 0:6418
r fc54–77 = 0:7230
The correlations between freshwater production and freshwater fish, shrimp and
crabs, and shellfish from 1978 to 1984 are as follows:
r ff78–84 = 0:6661
r fsc78–84 = 0:6620
r fc78–84 = 0:7269
The correlations between freshwater production and freshwater fish, shrimp,

crabs, and shellfish from 1985 to 1997 are as follows:
r ff85–97 = 0:7779
r fsc85–97 = 0:7745
r fc85–97 = 0:8164
3.3.1.4 Main Conclusions
1. The different yields of the three time periods were analyzed using the gray
correlation method, and the results are shown in Figs. 3.1, 3.2, and 3.3.
2. Figure 3.1 shows that the contribution of seawater production to the total pro-
duction of aquatic products is always greater than that of freshwater production.
However, with the passage of time, the contribution of seawater production to the
fish
0.6918
aquaculture
shrimp and crabs
0.7830
Seawater production OR 0.6963
0.7122 fishing shellfish
0.9769 0.6781
algae
0.6740
Total production
fish
0.6063
aquaculture
0.5226 shrimp and crabs
Freshwater production OR
0.6177 0.6418
fishing
0.6971
shellfish
0.7230
Fig. 3.1 Correlation between fishery production in China from 1954 to 1977 (Chen and Zhou
2002b)
fish
0.8099
aquaculture
shrimp and crabs
0.6334
0.9064 0.7750
algae
0.5880
Total production
fish
0.6661
aquaculture
0.6061 0.6620
fishing
0.7095
shellfish
0.7269
2002b)
total production of aquatic products decreased, and its correlation degree

decreased from 0.8163 in 1978–1984 to 0.6814 in 1985–1977, while the contri-
bution of freshwater production to the total production of aquatic products
decreased. The correlation degree increased slightly from 0.6061 in 1978–1984
to 0.6072 in 1985–1977.
56 X. Chen
fish
0.9345
aquaculture
shrimp and crabs
0.7209
0.9070 0.9549
algae
0.7170
Total production
fish
0.7779
aquaculture
0.6072 0.7745
fishing
0.6427
shellfish
0.8164
2002b)
3. From the perspective of the composition of seawater production, the contribu-

tion of seawater fishing to the total production of marine aquatic products is
always greater than that of mariculture, and the average correlation degrees are
0.9301 and 0.7124, respectively. The contribution of mariculture to the total
production of marine aquatic products increased, and its correlation increased
from 0.6334 in 1978–1984 to 0.7209 in 1985–1977.
For marine fish, shrimp and crabs, shellfish and algae, their contributions to
the total aquatic product production were basically the same before 1977, and
their correlations were all between 0.674 and 0.696, indicating that fishery
resource development and utilization are relatively balanced. However, between
1978 and 1984, the contribution of fish, shrimp, and crabs to the production of
seawater aquatic products increased significantly, and their correlations
increased to 0.8099 and 0.8450, respectively, while the correlations between
algae and the production of seawater aquatic products decreased to 0.5880. After
1984, the increase in the production of seawater aquatic products mainly came
from shrimp, crabs, shellfish and fish, and the correlations were 0.9604, 0.9549,
and 0.9345, respectively.
4. In freshwater fisheries, fishing and aquaculture have undergone structural
changes. Before 1977, freshwater production mainly came from freshwater
fishing, with a correlation degree of 0.6971, while the correlation degree
between freshwater aquaculture and freshwater aquatic product production
was only 0.5226. From 1978 to 1984, both freshwater fishing and freshwater
aquaculture developed to a certain extent. However, the development of fresh-
water aquaculture was even more rapid, and its correlation reached 0.8358. After
1985, the development of freshwater fishing shrank, its contribution to the
production of freshwater aquatic products decreased, and freshwater aquaculture

was further developed. and its correlation reached 0.8780.
Among freshwater fish, shrimp, and shellfish, the contribution of shellfish to
the production of freshwater aquatic products is always highest, while the
contribution of fish, shrimp, and crabs is relatively low. In different time periods,
the contributions of freshwater fish, shrimp, crabs, and shellfish to their yields
increased, and the correlations between 1985 and 1997 were 0.7779, 0.7745,
and 0.8164, respectively.
5. The correlation analysis of the time series shows that it is impossible to increase
the proportion of marine catch production in marine fisheries. In contrast, the
potential of marine aquaculture is relatively large, but attention should be given
to the protection of the aquaculture water environment. In addition, basic
research on the aquaculture capacity and optimal combination of waters has
been carried out. In addition, in freshwater fisheries, the increase in freshwater
fishing yield is also difficult to achieve. Some resources have been overfished,
and freshwater aquaculture has been the main source of the recent increase in the
production of aquatic products.
Through the above analysis, we can clearly understand the structure of
China’s fishery production in the past and present as well as its status, existing
problems and potential, which further indicates that the development of China’s
fisheries must be based on natural resource conditions and the level of fishery
economic development. We should formulate development plans in a targeted
manner, strive to reduce fishing intensity, and focus on the development of some
ecological fishery economies to ensure the sustainable development of China’s
fisheries.
3.3.1.5 Application of Gray Correlation in the Field of Fishery Resource

Assessment
3.3.1.5.1 Selection of Factors Affecting Changes in Fishery Resources
Yan et al. (1996) published “Relational factors for changes in Taihu silverfish
resources and methods for resource forecasting.” The gray correlation analysis
method was used to screen the main factors related to the Taihu silverfish resources.
According to the qualitative analysis of production practices, the nonbiological
factors that affect the changes in the number of whitebait in Lake Taihu are mainly
the water level in spring from March to May, the water level in summer from June to
September, and the fishing intensity (based on the amount of fishing in the lake after
spring and autumn floods). In terms of biology, there is a relationship between
competing bait, predation, and being preyed upon. Therefore, in the quantitative
analysis, the yields of whitebait, lake anchovy, shrimp, culter, and small trash fish
are considered the main sequences. The six items, including the amount of whitebait
in spring and autumn, were used as subsequences to calculate the degree of gray
correlation. The original values are shown in Table 3.17.
58
Table 3.17 The yield of natural fish in Lake Taihu and the values of its main factors (Yan et al. 1996)
Content 1989 1990 1991 1992 1993 1994
Main sequences Whitebait (tons) 1509.06 1479.27 2008.62 1606.92 1763.50 1118.18
Lake anchovy (ton) 7460.48 8142.34 6634.16 4625.41 3486.50 6706.55
Shrimp (tons) 898.22 1122.59 751.35 560.50 523.10 879.40
Culter (ton) 518.30 551.35 731.62 732.66 922.25 301.95
Small trash fish (tons) 2039.29 1884.88 2.070.88 3102.21 4057.87 2827.57
Subsequences Ship tonnage (ton) 31173.5 31462.5 31987.5 43721.0 45695.5 38276.0
Labor (person) 11083.0 11382.0 11429.0 11843.0 12546.0 12386.0
The resource index of whitebait in spring 3.48 2.70 5.05 3.45 2.55 4.3100
The resource index of whitebait in autumn 0.13 0.50 0.90 0.80 0.13 0.0600
Water level in spring (m) 3.02 3.14 3.33 3.13 3.08 2.9800
Water level in autumn (m) 3.52 3.17 3.96 3.03 3.70 2.9325
Note: The unit of the resource index of whitebait is kg/h
X. Chen
Table 3.18 Correlation values of natural fish and their related factors and their order (Yan et al.
1996)
The resource The resource
index of index of Water Water
Ship whitebait in whitebait in level in level in
Fish tonnage Labor spring autumn spring autumn
Whitebait 0.7528(4) 0.8553(2) 0.7484(5) 0.4565(6) 0.8501(3) 0.8978(1)
Lake 0.5741(5) 0.6745(3) 0.6678(4) 0.4896(6) 0.7218(2) 0.7349(1)
anchovy
Shrimp 0.5600(5) 0.6665(1) 0.6332(4) 0.4132(6) 0.6584(2) 0.6541(3)
Culter 0.8328(1) 0.7927(2) 0.7208(5) 0.4945(6) 0.7900(3) 0.7565(4)
Small 0.8664(1) 0.7375(2) 0.6785(5) 0.5146(6) 0.7050(3) 0.6866(4)
trash fish
Note: The numbers in parentheses are the order of the degree of gray correlation
The data in Table 3.17 were used to calculate the general gray correlation degree,
and the resolution coefficient ρ = 0.55 was used to obtain the gray correlation degree
value in Table 3.18. Table 3.18 shows that the water level is the first correlation
factor for whitebait and lake anchovy, and fishing intensity is the first correlation
factor for shrimp, culter, and small trash fish.
3.3.1.5.2 Selection of Factors Affecting the Catch
Wang (1996) published “Gray correlation analysis of Danjiangkou Reservoir catch

and its related factors.” The raw data of release specification for fish, fishery law
enforcement management, the fish release quantity, the amount of fishing effort, the
inflow of water, and the reservoir area are collected (Table 3.19). The study was
divided into two time periods: 1971–1982 (case I) and 1982–1990 (case II).
Assume that the data series of catch, release specification for fish, fishery law
enforcement management, fish release quantity, fishing effort, inflow of water, and
reservoir area after the averaged values of each year are x0, x1, x2, x3, x4, x5, and x6,
respectively, the correlation coefficient corresponding to each period is ξi (k), and the
correlation degree is ri (i = 1, 2..., 6). The results of the averaging of the raw data are
shown in Table 3.20. The correlation coefficient is shown in Table 3.21. Table 3.22
shows the degree of gray correlation and its ranking.
Table 3.22 shows that in case I, the correlation sequence between each factor and
the catch is 0.992 > 0.988 > 0.984 = 0.984 > 0.871 > 0.752, i.e.,
r1 > r6 > r4 = r5 > r2 > r3, r1, r6, r4, r5, r2, and r3 are the gray correlation between
the release specification for fish (cm) and the catch, reservoir area (10,000 hm2) and
the catch, fishing effort (10,000 tons) and the catch, inflow of water (100 million m3)
and the catch, fishery law enforcement management (10,000 yuan) and the catch,
and fish release quantity and the catch, respectively.
The gray correlation study showed the following:
60
Table 3.19 Raw data of gray correlation analysis (Wang 1996)

Release Fishery law enforcement Fish releasing
specification for management (ten thousand Fishing effort quantity (ten thousand Inflow of water Reservoir area Catch
Year fish (cm) yuan) (10,000 tons) kg) (100 million m3) (10,000 hm2) (10,000 kg)
1971 8 2 14 0.135 377.47 5.10 90.42
1972 8 2 13.2 0.135 281.25 5.09 97.93
1978 12.8 15.83 22.5 2.175 228.28 3.74 96.41
1979 12.8 15.83 22.5 2.175 320.41 4.07 107.85
1986 15 39.75 58 10.04 240.11 5.11 227.23
1987 16 37.20 70.02 10.04 375.39 5.40 227.40
1988 20 26.92 68 7.07 380.75 5.00 263.53
1989 20 17.38 59.9 7.07 375.00 4.80 237.00
1990 20.4 14.17 72 3.09 389.00 5.80 278.00
X. Chen
Table 3.20 Initialization results of raw data (initialization using 1971 as the denominator) (Wang
1996)
Situation Year x0 x1 x2 x3 x4 x5 x6
I 1971 1.0 1 1 1 1 1 1
1972 1.0 1 1 1 0.94 0.75 1
1978 1.07 1.60 7.92 16.11 1.61 0.61 0.73
1979 1.19 1.60 7.92 16.11 1.86 0.85 0.80
1986 2.51 2 19.88 74.37 4.14 0.64 1
II 1986 1 1 1 1 1 1 1
1987 1 1 0.94 1 1.21 1.56 1.06
1988 1.15 1.25 0.68 0.70 1.03 1.59 0.98
1989 1.04 1.25 0.45 0.70 1.03 1.56 0.94
1990 1.22 1.28 0.36 0.31 1.24 1.62 1.14
Table 3.21 Correlation coefficients between various factors and catches (Wang 1996)
Situation Year ξ1 ξ2 ξ3 ξ4 ξ5 ξ6
I 1971 1 1 1 1 1 1
1972 0.998 0.998 0.998 0.996 0.991 0.998
1978 0.985 0.840 0.705 0.985 0.987 0.991
1979 0.989 0.842 0.721 0.982 0.991 0.989
1986 0.986 0.674 0.336 0.957 0.951 0.960
II 1986 1 1 1 1 1 1
1987 1 0.883 1 0.684 0.448 0.883
1988 0.820 0.492 0.503 0.791 0.508 0.728
1989 0.684 0.435 0.572 0.978 0.467 0.820
1990 0.883 0.346 0.333 0.958 0.532 0.850
Table 3.22 The degree of gray correlation and ranking of each factor and the catch (Wang 1996)
Situation r1 r2 r3 r4 r5 r6
Gray relational degree I 0.992 0.871 0.752 0.984 0.984 0.988
II 0.877 0.631 0.682 0.882 0.591 0.856
Sorting I 1 5 6 3 3 2
II 2 5 4 1 6 3
1. The release specification for fish had the greatest impact on the catch. The
Danjiangkou Reservoir is a type of reservoir mainly inhabited by Erythroculter.
When the release specification for fish is small, most of the fish species will be
eaten by Erythroculter.
2. The next impact factor on the catch is the reservoir area. In a reservoir with a
small area, fish will be restricted by the density factor, so the population density
may be very high, but the total yield is not high. The larger the area of the
reservoir is, the larger the living space of the fish. The density decreases, the food
62 X. Chen
can increase relatively, the fish grows rapidly, and the individual is large;
therefore, the total yield is high.
3. Fishing effort and inflow of water also have a greater impact on the catch, which
is second only to the reservoir area and has a greater impact on the catch than
fishery law enforcement management and the fish release quantity. The inflow of
water indicates the amount of nutrients, which directly or indirectly restricts the
amount of fish resources and growth.
4. The impact of fishery law enforcement management and the fish release quantity
on the catch is small. The Danjiangkou Reservoir has a relatively large area, so it
is difficult to achieve effective fishery law enforcement management. This is the
reason why r2 is smaller than the other factors. However, r2 = 0.887 > 0.8,
indicating that fishery law enforcement management is still closely related to
catch. Therefore, it is necessary to strengthen fishery law enforcement manage-
ment. In case I, r3 was the smallest, indicating that the release specification for
fish (<16 cm) was too small, and only a one-sided increase in the fish release
quantity without increasing the release specification for fish resulted in a very
poor economic benefit.
In case II, there are 0.882 > 0.877 > 0.856 > 0.682 > 0.631 > 0.591, i.e.,
r4 > r1 > r6 > r3 > r2 > r5. The gray correlation analysis shows the following:
1. Under this situation, fishing effort has replaced the release specification for fish as
the primary factor affecting the catch, indicating that enhanced fishing is eco-
nomical. The release specification for fish dropped to second place, indicating
that the release specification for fish (20.4 cm) at this time basically met the
requirements. If the release specification for fish continues to increase, the effect
on the catch will gradually weaken, and thus, the economic effect will be worse.
2. The effect of the fish release quantity on the catch increased from the sixth to
fourth, indicating that the effect of the fish release quantity under this size
gradually increased, and an increase in the fish release quantity significantly
increased the catch. It is economical to increase the fish release quantity.
3. The inflow of water decreased from the fourth to the last, indicating that the effect
of fish release quantity on the catch in this period replaced the position of the
inflow of water; that is, the fish release quantity (10,000 kg) can affect the catch
compared to the inflow of water (100 million m3).
4. Fishery law enforcement management is still in fifth place, indicating that the role
of the fishery in the two situations has not changed. If the fish release quantity
cannot be increased, the fishing effort is certain, and fishery law enforcement
management should be strengthened to obtain a higher catch.
5. It is understandable that the reservoir area is always in a more important position.
Therefore, illegal occupation of the water surface should be minimized, or the
water surface occupation fee should be appropriately levied.
3.3.1.5.3 Evaluation of the Abundance of Fishery Resources
Ommastrephes bartramii, neon flying squid, is widely distributed in the North

Pacific Ocean and has abundant resources. This resource was first developed and
utilized by Japanese squid fishing survey vessels in 1974. Subsequently, South
Korea and Taiwan Province of China also joined the development ranks and
gradually developed into a fleet of gill net-based fishing fleets. The output of squid
reached 300,000–400,000 tons. In 1993, the Chinese mainland began to develop and
utilize the resources of neon flying squid in the North Pacific Ocean, and the
operating fishing grounds continued to expand eastward. The squid fishing fisheries
in the North Pacific Ocean have become an important part of China’s offshore
fisheries. Therefore, conducting research on the spatial distribution of its resources
and fishing grounds and their interannual change is of great significance to ensure the
rational exploitation and utilization of this squid (Chen et al. 2003).
Chen et al. (2003) used the statistics of China’s squid fishing catch in the North
Pacific Ocean from 1995 to 2001, including the operating location, operating date,
number of vessels operating in the fishing area, total production in the fishing area,
and average daily production. The average daily production for each longitude and
latitude in each year was also calculated. The gray relational degree is used to
evaluate the resource status of neon flying squid in each year. That is, the maximum
value of each longitude (latitude) in each year is selected to form the optimal vector
and used as the parent sequence. The greater the degree of gray correlation is, the
better the resource status, and vice versa.
Based on the history of Chinese squid fishing vessels in the North Pacific Ocean
exploiting and utilizing squid resources, this study was conducted in three sea areas
to the west of 160°E, 160°E to 170°E and to the east of 170°E. The results of this
study are as follows:
1. Comparison of the resource status of squid in the waters west of 160°E
In the sea area west of 160°E, the optimal vectors for the average daily yield of
each longitude (143°E–160°E) are 2.2, 2.848, 3.182, 3.719, 5.803, 2.524, 3.526,
2.406, 2.462, 2.317, 2.667, and 3.806. The gray correlations between the average
daily production of each year at different longitudes and the optimal average daily
production in 1995–2001 were 0.7061, 0.5555, 0.6851, 0.8025, 0.7097, 0.6294,
and 0.5504, respectively. The order of abundance of squid from high to low in
each year was 1998, 1999, 1995, 1997, 2000, 1996, and 2001.
2. Comparison of the resource status of squid in the waters west of 160°E–170°E
In the sea area of 160°E–170°E, the optimal vectors of the average daily yield
of each longitude (161°E–170°E) from 1997 to 2001 were 1.779, 2.013, 2.371,
2.048, 1.873, 1.569, 1.585, 1.453, and 1.549, respectively. The gray correlations
between the average daily production of each year at different longitudes and the
optimal average daily production in 1997–2001 were 0.6620, 0.8467, 0.6318,
0.6040, and 0.4938, respectively. The order of abundance of squid from high to
low in each year is 1998, 1997, 1999, 2000, and 2001.
3. Comparison of the resource situation of squid in the sea east of 170°E
64 X. Chen
In the sea area to the east of 170°E, the optimal vectors of the average daily
yield of each longitude (171°E–173°W) from 1999 to 2001 are 1.178, 0.857,
1.346, 1.706, 1.817, 1.876, 1.224, 1.454, 1.07, and 1.427, 1.375, 1.421, 1.587,
1.642, 1.627, 1.276, and 0.819. The gray correlations between the average daily
production of each year at different longitudes and the optimal average daily
production in 1999–2001 were 0.966, 0.640, and 0.677, respectively. The order
of abundance of squid from high to low in each year is 1999, 2001, and 2000.
4. Comparison of the resource status of squid at different latitudes
In the North Pacific Ocean, the optimal vectors at various latitudes (37°N–45°
N) are 2.05, 4.152, 2.496, 2.225, 2.895, 2.752, 2.405, 2.693, and 2.927. The gray
correlations between the yield of each year and the optimal average daily yield
were 0.854, 0.709, 0.768, 0.861, 0.769, 0.649, and 0.630, respectively. The order
of abundance of squid from high to low in each year is 1998, 1995, 1999, 1997,
1996, 2000, and 2001.
Based on the gray correlation evaluation, we obtained the resource status of squid
in various sea areas of the North Pacific Ocean from 1995 to 2001. The status of
squid resources in the North Pacific Ocean was the best in 1998, while the status of
squid in 2000, 2001, and 1996 was poor. In 1996, it was at an intermediate level.
This is basically consistent with the actual production situation and marine environ-
mental conditions. For example, in 1998, the Kuroshio power was strong, and it was
a warm-water year, while in 1996, the Oyashio was strong and the Kuroshio was
relatively weak, and it was a cold-water year. Therefore, as a warm-water species of
this squid, the strength of the Kuroshio directly affects the amount of resources and
the formation of fishing grounds for neon flying squid.
3.3.2 Application of Gray Correlation in the Evaluation

of Sustainable Use of Fishery Resources
The sustainable use of fishery resources is the core and essential issue of the
sustainable development of the fishery economy. Chen and Zhou (2002a) published
the “Gray Relational Assessment of the Sustainable Utilization of Fishery
Resources” and analyzed the evaluation of the sustainable use of fishery resources
in the East China Sea using the gray relevance analysis method.
The evaluation index system proposed by Chen and Zhou (2002a) includes the
three subsystems of resource environment, society, and economy. The resource
environment subsystem includes trophic level R101, the proportion of high-quality
fish in marine fishing yield R102, the proportion of catch from nonselective fishing
gear in the total marine fishing yield R103, the marine fishing yield per ship R104, the
marine fishing yield per tonnage of motor-driven fishing vessels R105, and the marine
fishing yield per kilowatt of motor-driven fishing vessels R106. In the subsystems of
society, there are six indicators, including marine fishing professional labor R201,
marine fishing part-time labor R202, the proportion of marine fishing labor in fishery
Table 3.23 Index values of Year R101 R102 (%) R103(%) R106 (t/KW)
the resource and environmen-
1978 2.64 63.189 0.436 1.178
tal subsystems after screening
(Chen and Zhou 2002a) 1979 2.72 59.118 0.411 1.050
1980 2.73 46.483 0.569 1.038
1981 2.72 51.056 0.585 0.956
1982 2.64 48.178 0.622 0.935
1983 2.63 38.596 0.645 0.875
1984 2.54 41.034 0.677 0.891
1985 2.56 39.083 0.623 0.869
1986 2.52 37.618 0.676 0.881
1987 2.50 37.917 0.671 0.821
1988 2.40 30.400 0.683 0.727
1989 2.43 36.130 0.699 0.683
1990 2.49 36.125 0.724 0.663
labor R203, the proportion of marine fishing labor in fishery population R204, fishery
population R205, and the per capita share of aquatic products R206. The subsystems of
the economy include 11 indicators, including marine fishing yield R301, the propor-
tion of marine fishing yield in marine fishery yield R302, the proportion of marine
fishing yield in the total fishery output R303, the proportion of the total fishery output
in the agricultural output value R304, the number of motor fishing vessels R305, the
total tonnage of motor-driven fishing vessels R306, the total power of motor-driven
fishing vessels R307, the per capita income of fishermen R308, the per capita income
of fishermen R309, the per capita marine fishing yield of fishing labor R310, and the
per capita marine fishing yield of fishery population R311. The data are from the
“China Fishery Statistics Collection” (1989–1993) (edited by the Fishery Bureau of
the Ministry of Agriculture of the People’s Republic of China, 1996), the China
Fishery Statistics Collection (1994–1998) (edited by the Fishery Bureau of the
Ministry of Agriculture of the People’s Republic of China, 2000), and China
Fisheries Statistics for 40 years (edited by the Fisheries Department of the Ministry
of Agriculture of the People’s Republic of China, 1991).
Considering that there are differences in the dimensions of the original data and
the significant differences in the order of magnitude between the indicators, the
correlation coefficients between the indicators in each subsystem should be initial-
ized. Additionally, 11 indicators for evaluation were obtained through principal
component analysis and independence analysis. They are R101, R102, R103, and
R106; R201, R203, R204, and R206; and R302, R310, and R311. The raw data of each
indicator and their weights are shown in Tables 3.23, 3.24, 3.25, and 3.26.
The optimal value of each indicator was selected to form the parent sequence, and
each year was used as the subsequence. The general gray correlation analysis method
was used to evaluate the sustainable use of fishery resources in the East China Sea
between 1978 and 1990. The correlation coefficient was set to 0.5. The results are
shown in Table 3.27.
66 X. Chen
Table 3.24 The values of Year R201 (people) R203 (%) R204 (%) R206 (people)
each index of the social
1978 341,835 75.146 21.636 1,830,102
subsystem after screening
(Chen and Zhou 2002a) 1979 344,977 68.925 21.266 1,887,286
1980 359,779 61.343 22.191 1,940,048
1981 381,907 61.467 22.757 1,992,514
1982 366,601 52.893 19.992 2,106,660
1983 320,813 44.588 18.063 2,175,966
1984 344,289 50.804 21.316 1,966,558
1985 361,232 50.216 20.325 2,302,175
1986 408,852 52.307 21.830 2,333,331
1987 427,508 52.343 22.719 2,352,121
1988 442,793 51.419 23.305 2,423,091
1989 416,503 50.048 21.748 2,417,011
1990 419,822 48.973 20.757 2,524,825
Table 3.25 Indicator values Year R302 (%) R310 (t/fishing labor) R311 (t/people)
of the economic subsystem
1978 91.433 3.839 0.831
after screening (Chen and
Zhou 2002a) 1979 89.969 3.540 0.753
1980 89.582 3.345 0.744
1981 89.386 3.332 0.730
1982 88.706 3.485 0.726
1983 87.590 3.731 0.678
1984 85.479 3.550 0.814
1985 84.333 3.646 0.731
1986 84.643 3.526 0.779
1987 84.289 3.674 0.858
1988 83.179 3.674 0.837
1989 83.589 4.148 0.894
1990 84.068 5.472 0.910
Table 3.26 The weight of Indicator R1 R2 R3

each subsystem and its index
Weight 0.5245 0.1688 0.3067
(Chen and Zhou 2002a)
Indicator R101 R102 R103 R106
Weight 0.1621 0.1574 0.2537 0.4267
Indicator R201 R203 R204 R206
Weight 0.2857 0.2857 0.2857 0.1429
Indicator R302 R310 R311
Weight 0.3333 0.3333 0.3333
From the evaluation results in Tables 3.27, it can be seen that the level of
sustainable utilization of fishery resources in the East China Sea between 1978 and
1990 basically showed a downward trend. Among them, the level of sustainable use
during 1983 to 1986 was relatively low, the level of sustainable use in 1983 was the
Table 3.27 Evaluation results of the sustainable use of fishery resources in the East China Sea from
1978 to 1990 (Chen and Zhou 2002a)
Year 1978 1979 1980 1981 1982 1983 1984
Evaluation value 0.7473 0.6541 0.5579 0.5401 0.4703 0.4234 0.4347
Year 1985 1986 1987 1988 1989 1990
Evaluation value 0.4330 0.4550 0.4915 0.5165 0.4691 0.5252
lowest, and the level of sustainable use in 1990 was only 75% of that in 1978. This
evaluation result is basically in line with the reality of the exploitation and utilization
of fishery resources in the East China Sea. Since the 1980s, the fishery resources in
the East China Sea have been overexploited and utilized. Especially in 1983, due to
the decline in offshore fishery resources, some resources in the open sea, such as
mackerel and mackerel, were not exploited and utilized. After 1983, the develop-
ment and utilization of some pelagic fish resources in offshore waters promoted the
improvement of the level of sustainable utilization of fishery resources (Chen and
Zhou 2002a).
3.3.3 Evaluation of the Influencing Factors of Fishery Water

Quality
Naked carp in Qinghai Lake is the only aquatic economic animal in Qinghai Lake
and plays a central role in the entire Qinghai Lake ecosystem. Therefore, it is
particularly important to evaluate the aquatic environment quality of the aquatic
germplasm resources of the Qinghai Lake Naked Carp. Wang (2015) used the gray
relational analysis method to evaluate and analyze the fishery environmental quality
based on the water quality data of five monitoring sections in the Qinghai Lake
Naked Carp National Aquatic Germplasm Resource Reserve in 2012, with a view to
provide a scientific basis for protecting and recovering Naked Carp resources in
Qinghai Lake.
A total of five monitoring sections were set up in the study area: Shaliu River,
Quanji River, Buha River, Heima River, and Wharf 151. The evaluation criteria for
the relevant factors of water environmental quality refer to the “Environmental
Quality Standards for Surface Water” (GB3838-2002), and the standards are
shown in Table 3.28. The research data are from the Fishery Environmental Mon-
itoring Station of Qinghai Province.
First, the study series was normalized: for the comparison series, the standard
value corresponding to Class I was set to 1, and the standard value corresponding to
Class V water quality was set to 0; the normalized method of taking the standard
value of dissolved oxygen is contrary to other water quality parameters. The
normalized comparison sequence is shown in Table 3.29. For the series to be
compared, among the 5 values of the same evaluation factor, the highest pollutant
content is taken as 0, the lowest is taken as 1, and the remaining 3 standard values are
68 X. Chen
Table 3.28 Water environmental quality assessment standards (Wang 2015)

Potassium permanganate Hexavalent
Class DO TN index Ammonia Mercury chromium
I 7.5 0.2 2 0.15 0.00005 0.01
II 6 0.5 4 0.5 0.00005 0.05
III 5 1.0 6 1.0 0.0001 0.05
IV 3 1.5 10 1.5 0.001 0.05
V 2 2.0 15 2.0 0.001 0.10
Table 3.29 Comparison series after normalization (Wang 2015)

Potassium permanganate Hexavalent
Class DO TN index Ammonia Mercury chromium
I 1 1 1 1 1 1
II 0.727 0.833 0.846 0.811 1.000 0.556
III 0.545 0.556 0.692 0.541 0.947 0.556
IV 0.182 0.278 0.385 0.270 0.000 0.556
V 0 0 0 0 0 0
Table 3.30 Normalized sequence of five sections to be compared (Wang 2015)

Monitoring Potassium Hexavalent
point DO TN permanganate index Ammonia Mercury chromium
Liusha 0.533 1 0.706 0.726 1 0
River
Quanji 0.267 0.167 1 0.732 0.500 1
River
Heimahe 0 0.208 0.941 0.701 0.750 0.056
River
Buha River 0.667 0.333 0.824 1 0 0.778
Wharf 151 1 0 0 0 0 0.722
Table 3.31 Calculation results of the correlation degree of each monitoring point (Wang 2015)
Class Liusha river Quanji river Heimahe river Buha river 151 Wharf
I 0.580 0.526 0.362 0.506 0.261
II 0.712 0.479 0.492 0.583 0.261
III 0.701 0.497 0.518 0.537 0.314
IV 0.302 0.495 0.453 0.571 0.567
V 0.273 0.297 0.483 0.320 0.694
determined according to the interpolation method. Table 3.30 shows the series of
normalized data to be compared.
The correlation coefficient was set to 0.5, and the gray correlation degree was
calculated. The calculation results are shown in Table 3.31. The higher the correla-
tion degree is, the better the correlation with the comparison series, indicating that
the water quality level is close to a certain water quality standard in the comparison
series. It can be seen from the calculation results in Table 3.31 that the Shaliu River
monitoring point has the highest correlation with the Class II water quality standard,
but the correlation with the Class III water quality standard is also relatively large,
and its correlation degree is 0.701, indicating that the water quality has just reached
Class II. In the same way, the Heima River, Buha River, and 151 Wharf can be
evaluated as Class III, Class II, and Class V. Among them, the correlation between
the monitoring points of the Heima River and the Class II water quality standards is
only second to that of the Class III water quality standards, indicating that the water
environment quality presents a trend of a virtuous cycle. The correlation degree
indicates that the water environment is in an unstable state, and there is a trend of
deterioration to Class III water. The water environment quality of the 151 Wharf
monitoring point is assessed as Class V because this monitoring point has four
monitoring indicators inferior to it. For the same environmental factors at other
monitoring points, when the series of comparisons were normalized, four factors in
the series after normalization were 0, resulting in the evaluation of the water
environment as overprotected. Taking all the monitoring points as the five influenc-
ing factors of the Qinghai Lake naked carp aquatic germplasm resource protection
zone, the arithmetic mean is calculated, and the correlation between the entire water
area and the water quality standards of Classes I, II, III, IV, and V is found. They
were 0.477, 0.505, 0.513, 0.478, and 0.413, respectively. The overall water quality
evaluation of the Qinghai Lake naked carp aquatic germplasm resource protection
area is Class III, and the correlation between this area and the Class II water quality
standard is only second to the Class III water quality standard, indicating the overall
water quality status of the entire region.
3.3.4 Application of Gray Correlation in the Aquaculture

Industry
Chen (1991) published “The application of gray correlation analysis in aquaculture.”

In the study, the feed coefficient column was used as the reference series, and the
animal and plant protein content series were used as the comparison series
(Table 3.32). The gray correlation analysis showed that the correlation degree
between the plant protein content and the feed coefficient was 0.96, and the
Table 3.32 Relationship between plant and animal protein content and diet coefficient (Chen
1991)
Feed group number
Item 1 2 3 4 5 6 7
Plant protein content (%) 24.45 22.55 21.61 20.67 19.70 18.76 17.82
Animal protein content (%) – 2.02 3.06 4.04 5.08 6.12 7.10
Feed coefficient 1.86 1.72 1.68 1.50 1.67 1.85 1.95
70 X. Chen
Table 3.33 Benefits of fish farming in rice fields (Chen 1991)

Survey site number
Item 1 2 3 4 5 6 7
Amount of fish species (tails/mu) 444 269 69 419 263 395 500
Organic fertilizer input (kg/mu) 926 96 28.5 416.5 333.5 329 416.5
Labor input (units/mu) 3.7 4.6 2.9 4.2 10.0 1.3 2.1
Net income from fisheries (yuan/mu) 83.96 39.87 30.19 54.08 86.18 40.00 61.39
correlation degree between the animal protein content and the feed coefficient was
0.58. Therefore, the effect of plant and animal protein content in the grass carp diet
on the diet coefficient of grass carp was significantly stronger than that of the latter.
Studies have shown that the rational selection of plant protein content is a crucial
factor (Chen 1991).
At the same time, Chen (1991) used fishery net income as the reference series and
fish species input, organic fertilizer input, and labor input as the comparative series
(Table 3.33). The gray correlation between labor input and fishery net income was
0.76, 0.62, and 0.77, respectively. The analysis showed that the main factor affecting
the net income of rice-fish farming was labor input, followed by fish species input.
The key to achieving good returns from rice-fish farming is the implementation of
human management measures, as described by many aquaculture experts as “three-
part farming, seven-part management.”
3.3.5 Applications in Fishery Biology

3.3.5.1 Analysis of Factors Affecting Fish Behavior
Gray correlation analysis is also used in the evaluation of factors affecting fish
behavior. He (1989) published “A Study on the Application of Gray System Theory
in the Analysis of Fish Cage Experiments,” which used the gray relational analysis
method to analyze the factors affecting the catch of the cage. In the cage fishery,
there are many factors that affect the yield of cage catches, and the degree of
influence is also different. However, the primary and secondary relationships of
the factors are not clear, and the entire system can be regarded as gray. Four factors,
i.e., cage time, cage space, bait in the cage, and cage darkness, were considered in the
experiment. In the factor analysis, the cage catch yield was used as the reference
sequence, and the four factors were used as the subsequences. Because it is difficult
to measure the freshness of the bait and the darkness in the cage, fuzzy quantification
was used for processing. The shading is represented by x, and the freshness is
represented by y. Then, the values of the two are set as follows:
Table 3.34 Experimental data of various factors and cage catch production after normalization
(He 1989)
Time series 1 2 3 4
Cage yield X0 1.125 1.333 2.000 2.200
Fishing time X1 0.1987 0.1987 1.391 0.9934
(Fishing time X1)a (48)a (48)a (24)a (60)a
Cage space X2 1.074 0.7395 0.6219 0.9561
Cage darkness X3 0.7500 0.8330 1.800 1.667
Freshness of the bait X4 0.8750 1.130 2.333 2.000
a
The data in parentheses represent unstandardized experimental data, and the unit is hour
Table 3.35 Degree of correlation between various factors and cage catch yield (He 1989)
Influencing factors
Contents Bait freshness Darkness in the cage Cage space Fishing time
Gray correlation 0.8645 0.7568 0.6345 0.4997
Ranking 1 2 3 4
x = ½2; 1; 0; 0T
Round truncated cage, folding cage, rectangular cage, round port cage
y = ½3; 1; 1; 0; 0T
Hairtail, mackerel, anchovy, loach, eel

The obtained test data are shown in Table 3.34 after processing according to the
above methods.
After calculation, the gray correlation between each factor and the catch was
obtained (Table 3.35). Table 3.35 shows that the main factor affecting the catch is the
freshness of the bait, indicating that the fish in the island reef area prefer to eat bait
with a strong odor. Fresh bait was placed in the fish cage, and the cage feeding rate
was high, which was confirmed by the production practice. At the same time, the
impact of cage darkness on the yield of cage catches was also very significant,
second only to the bait. This may be caused by the habit of fish in the island reef area.
These fishes live in rocky reefs, caves, or crevices for a long time and are used to
dark areas. It can also be seen from Table 3.35 that the effect of cage space cannot be
ignored. A large cage space is conducive to fish habitation, and the probability of fish
escaping from the cage is relatively small. The large cage space is also conducive to
fish survival. However, oversized cages can lead to problems such as increased costs
and increased operational and transportation difficulties. For the fishing time factor,
the calculation results showed that it had the smallest impact on the catch of cages.
This may be because the samples in the experimental data were all collected over
24 h, and the fish were likely to enter the cage within 24 h, thus causing the fishing
impact of time on the catch of the cages. If the fish have not entered the cage within
72 X. Chen
24 h, the freshness of the bait in the cage will be greatly reduced, and the bait factor
will not play a role, which directly affects the catching effect of the cage. Despite the
extension of the cage time, the effect was not increased.
3.3.5.2 Selection and Comparison of Fish Growth Models
The gray correlation analysis method was used to select and evaluate the fish growth
model and achieved good results. Chen (1991) published “The Application of Gray
Relational Analysis in Fisheries,” which used the gray relational analysis method to
compare three commonly used fish growth models with the actual values and select
the optimal model based on the degree of gray relational analysis.
According to the known growth data of grass carp, the following three growth
models were obtained:
Growth model I: Wt = 12011[1–e-0.3002 (t + 0.3399)] 3
Growth model II: Wt = 36971–38031 e-0.0404t
Growth model III: Wt = 7298/(1+ e3.9890–1.3086t)
According to the three growth models, the theoretical body weight of each age
group was calculated (Table 3.36). The measured body weight was used as the
reference column, and the model-calculated value was the subsequence. The corre-
lation between the calculated value and the measured value was obtained using the
gray correlation analysis method. The gray correlations with models I, II, and III
were 0.85, 0.89, and 0.77, respectively, indicating that model II was the better model
for describing the body weight growth pattern of grass carp (Table 3.36).
3.3.5.3 Gray Correlation Analysis of Fish Morphological Traits

and Body Weight
In the breeding of aquatic animals, the body weight of growth traits is often an
important indicator for the selection of growth traits. In addition to body weight,
growth traits also include many morphological traits that have varying degrees of
correlation with body weight. Indirect selection of body weight can be achieved
through the selection of morphological traits. Therefore, it is necessary to understand
the degree of correlation between various morphological traits and body weight,
Table 3.36 Average body weight of grass carp (Chen 1991) (unit: g)
Age
Method 1 2 3 4 5 6 7 8
Actual measurement 475 1515 3305 4915 6235 6910 8170 9475
Growth model I 436 1543 3047 4638 6119 7400 8455 2989
Growth model II 446 1892 3281 4615 5896 7126 8308 9443
Growth model III 468 1477 3534 5666 6770 7148 7254 7283
which is an important basis for breeding programs. In this study, Liu et al. (2017)
applied the gray relational analysis method to analyze the relationship between
morphological traits and body weight of small yellow croaker and analyzed the
relative importance of different morphological traits to body weight, which provided
important guidance for the development of improved breeding programs for small
yellow croaker.
The experimental fish were small yellow croaker populations aged 4.5 months
and were cultured in Xixuan Fishery Science and Technology Island of Zhejiang
Institute of Marine Fisheries. A total of 123 tails were randomly sampled, with a
body weight of 18.172 ± 5.370 g. The morphological traits were accurately mea-
sured tail by tail using a Vernier caliper, including full length, body length, head
length, trunk length, tail length, tail peduncle length, tail peduncle height, and body
height. The statistics were divided into males and females. Because the dimensions
of body weight and morphological traits are different, it is impossible to directly
compare the traits. Therefore, it is necessary to perform appropriate data conversion.
In the study, the standard deviation method was used to dimensionlessly process the
data of each trait, and then the data obtained after the transformation were used. Gray
correlation analysis was performed. The eight morphological traits and body weight
indicators of small yellow croaker were treated as a gray system, the general gray
correlation method was used for calculation, and the resolution coefficient was set to
0.5. The degree of association between morphological traits and body weight can be
determined according to the degree of gray correlation, thereby determining the
relative importance of the morphological traits to body weight.
According to statistics, among all traits, the coefficient of variation of body
weight was the largest, and the coefficient of variation between morphological traits
was not much different. The minimum, maximum, and mean values of the female
samples of small yellow croaker were greater than the corresponding parameters of
the male samples. The results of the significance test showed that. The body length
and torso length of the female samples were significantly larger than those of the
male samples (P < 0.01), and the head length, tail stalk length, tail stalk height, and
body weight of the male samples were significantly larger than those of the male
samples (P < 0.05). The difference between male and female samples was not
significant (P > 0.05), indicating that there was a difference in growth rate between
male and female individuals of small yellow croaker, but not all morphological traits
showed significant differences.
The calculation results of the gray correlation are shown in Table 3.37. Table 3.37
shows that the gray correlation degree between different morphological traits and
body weight of small yellow croaker is in the range of 0.5266–0.6812 (female) and
0.5288–0.7116 (male). The gray correlations between various morphological traits
and body weight in male samples were greater than those in female samples. In
addition to the tail length trait, the standard deviation of the correlation coefficient of
the male samples was also slightly smaller than that of the female samples. The
results of the gray correlation between morphological traits and body weight in the
female and male samples were as follows: full length > body length > trunk
length > tail length > tail stalk length > body height > tail stalk height > head
74 X. Chen
Table 3.37 Gray correlation between morphological traits and body weight of samples (Liu et al.
2017)
Female Male
Gray correlation Gray correlation
Traits (±SD) Ranking (±SD) Ranking
Full length 0.6812 ± 0.0977 1 0.7116 ± 0.0970 1
Body length 0.6539 ± 0.0956 2 0.6728 ± 0.0905 2
Head length 0.5266 ± 0.0693 8 0.5284 ± 0.0676 8
Body length 0.5658 ± 0.0796 3 0.5729 ± 0.0786 4
Tail length 0.5648 ± 0.0775 4 0.5732 ± 0.0779 3
Caudal peduncle length 0.5424 ± 0.0736 5 0.5472 ± 0.0727 5
Tail stalk height 0.5343 ± 0.0737 7 0.5385 ± 0.0730 6
Body height 0.5345 ± 0.0714 6 0.5378 ± 0.0700 7
length (female); Length > tail length > trunk length > tail stalk length > tail stalk
height > body height > head length (male). Therefore, there was a certain difference
in the degree of association between the morphological traits and body weight of the
female and male samples of small yellow croaker. Therefore, if the data of male and
female samples are analyzed separately, the results will be more accurate and
reliable. A comprehensive comparison of morphological traits showed that the
gray correlation between full length and body weight was the highest (0.6812 for
females and 0.7116 for males), followed by body length (0.653 9 for females and
0.672 8 for males). The head length was the lowest (0.5266 for females and 0.5284
for males), and there were certain differences in the ranking of other traits between
female and male samples.
In this study, the gray correlation analysis method was used to analyze the
correlation between the eight morphological traits of small yellow croaker and
body weight. The results showed that the correlation between the full length and
the body weight of the female and male samples was the largest, followed by the
body length. The head length was the smallest. According to the principle of gray
correlation analysis, morphological traits with high correlation are closely related to
body weight and vice versa. The effect of head length is relatively small. The
correlation between several other morphological traits and body weight showed
certain differences between male and female samples. This indicates that when
analyzing the growth data of small yellow croaker, it is necessary to analyze the
female and male samples separately.
References
Chen CQ (1991) Fishery application of gray relational analysis. J Hydroecol 5:25–27. (In Chinese)
Chen XJ (2003) Application of gray system theory in fishery science. China Agricultural Press.
(In Chinese)
(In Chinese)
Chen XJ, Zhou YQ (2002a) Assessment of sustainable use of fisheries resources based on the
methods of gray relative relationship. J Fish China 26(4):331–336. (In Chinese)
Chen XJ, Zhou YQ (2002b) Gray relationship analysis of Chinese fishery yield construction.
Chinese Fish Econ 2:30–33. (In Chinese)
Chen XJ, Xu LX, Tian SQ (2003) Spatial and temporal analysis of Ommastrephe bartrami
resources and its fishing ground in North Pacific Ocean. J Fish China 27(4):334–342.
(In Chinese)
Deng JL (1987) Basic method of the gray system. Huazhong Technology College Press.
(In Chinese)
Deng JL (1990) A course in gray system theory. Huazhong Technology University Press.
(In Chinese)
He ZJ (1989) Application of gray system theory to pot fishing experimenting and analyzation. J
Zhejiang Coll Fish 8(2):123–126. (In Chinese)
Liu SF, Yang YJ, Wu LF et al (2014) Gray system theory and its application. Science Press, Beijing.
(In Chinese)
Liu F, Lou B, Chen RY et al (2017) Analysis of gray relationship between morphological traits and
body weight in the small yellow croaker (Pseudosciaena polyactis). J Shang Ocean Univ 26(1):
Wang SB (1996) The gray system relevant analysis of the fish catch and its relevant factors of
Danjiangkou reservoir. Syst Sci Comprehen Stud Agric 12(1):4–7. (In Chinese)
Wang W (2015) Application of gray correlation analysis on water quality assessment of the
Gymnocypris przewalskii in aquatic germplasm reserve. Heilongjiang Agric Sci 3:98–100.
(In Chinese)
Yan XM, Hu SK, Shi XK (1996) A study on the factors affecting icefish resource and the
forecasting of the resources. J Fish China 20(4):307–313. (In Chinese)
Chapter 4
Gray Cluster Analysis
Xinjun Chen
Abstract Clustering is the process of dividing a collection of physical or abstract

objects into classes composed of similar objects. In other words, according to a
certain standard (such as distance), a data set is divided into different classes or
clusters, so that the similarity of data objects within the same cluster is as large as
possible, at the same time no longer within the same cluster of data objects as large as
possible differences. The cluster generated by clustering is a set of data objects,
which are similar to each other and different from other objects in the same cluster.
There are many classification problems in natural science and social science. Cluster
analysis, also known as group analysis, is a statistical analysis method for the study
of classification problems (samples or indicators). Cluster analysis originated from
taxonomy, but cluster is not equal to classification. The difference between cluster-
ing and classification is that the classification is unknown, clustering is an
unsupervised learning task, do not know the real sample markers, only make the
similarity of samples together. Classification is a supervised learning task, which
uses known sample markers to train the learner to predict the classification of
unknown samples. Clustering can be used as a separate process to find the inherent
distribution structure of data, and it can also be used as a precursor process of other
learning tasks. The content of clustering analysis is very rich, including systematic
clustering, ordered samples clustering, dynamic clustering, fuzzy clustering, graph
theory clustering, gray clustering, and so on. Clustering is also an important concept
in data mining. Gray clustering is a method to aggregate some observation indexes or
objects into several definable categories according to the correlation matrix or the
whitening weight function of gray number, it can be divided into gray constellation
clustering, gray relational clustering, and gray whitening function clustering. In this
chapter, the concept and types of gray clustering are introduced, and the principles
and calculation methods of gray constellation clustering, gray relational clustering,
gray variable weight clustering, and gray fixed-weight clustering are expounded, at
the same time, the application of gray clustering in fishery science is analyzed.
X. Chen (✉)
78 X. Chen
Keywords Gray clustering · Gray constellation clustering · Gray relational

clustering · Gray variable weight clustering · Gray fixed-weight clustering
4.1 Concept of Gray Clustering
Gray clustering is a method that aggregates some observation indicators or obser-

vation objects into several definable categories according to the correlation matrix or
the whitening weight function of gray numbers. A cluster can be viewed as a
collection of objects belonging to the same type of observation. In practical prob-
lems, each observation object often has many characteristic indicators, which makes
it difficult to perform accurate classification.
According to different clustering methods, gray clustering can be divided into
gray constellation clustering, gray relational clustering, and gray-type whitening
function clustering. Gray constellation clustering quantitatively determines the rela-
tionship between samples based on their own attributes, uses the principle of
similarity, and performs natural clustering based on this relationship. Gray relational
clustering is mainly used for the merging of similar factors to simplify complex
systems. Through gray relational clustering, we can analyze whether there are
several factors in many factors that are closely related so that the comprehensive
average index of these factors or one of them can be used to represent these factors in
the study, and at the same time, the information is not available. Therefore, before
conducting a large-scale survey, the collection of unnecessary variables (factors) can
be reduced through the gray correlation clustering of typical sampling data to save
costs and funds. Gray class whitening weight function clustering is mainly used to
check whether the observed objects belong to different preset categories so that they
can be treated differently. In terms of computational workload, gray-type whitening
functions are more complex than gray correlation clustering and constellation
clustering (Chen 2003, 2023).
4.2 Gray Constellation Clustering

4.2.1 Principles and Methods
Constellation clustering is a relatively simple and easy-to-implement clustering

method in gray clustering. The basic principle is as follows: each sample point is
placed in an upper semicircle according to a certain number of relationships, a
sample point is represented by a “star point,” and the sample points of the same
type form a “constellation,” which is then sketched. The boundary between different
constellations is drawn so that classification can be carried out. In essence, it trans-
forms a large amount of information (or index value) in a sample into dimensionless
by means of transformation (range transformation) of the original data and becomes
a simple problem of spatial coordinate comparison.
4 Gray Cluster Analysis 79
In general, constellation clustering has the following steps:

1. Range transformation is performed on the original index values, and the
transformed values fall within the closed interval of [0°, 180°]. The calculation
formula is
X ij - X j min
aij = × 180 °
X j max - X j min
where aij is the transformed data, expressed as an angle; Xij is the raw data; Xjmax
is the maximum value of the jth variable; Xjmin is the minimum value of the jth
variable; ( j = 1, 2, 3 . . ., P is the number of indicators); (i = 1, 2, 3 . . ., N is the
sample number).
2. For each indicator, a weight Wj is given according to its degree of influence on the
system change, so that
p
wj = 1
j=1
0 < wj < 1:
where wj is the weight of the jth index

3. Calculate the Cartesian coordinates of each indicator.
Using the transformation relationship between polar coordinates and rectan-
gular coordinates, the values of Xi and Yi of each index of each point are first
obtained, and then the values of Xi and Yi of each index of each point are added
together to obtain the coordinates of each sample point. Value. Its transformation
formula is
p
Xi = W j cos aij
j=1
p
Yi = W j sin aij
j=1
where Xi is the abscissa of the ith sample point; Yi is the ordinate of the ith sample
point.
4. Draw a constellation diagram
Draw an upper semicircle with a radius of 1, using the center of the circle as
the origin of coordinates and the base of the upper semicircle as the X-axis.
Similar and close sample points are clustered together to form a “constellation.”
5. Calculate the comprehensive index value
The mathematical expression of the comprehensive index value is
80 X. Chen
p
Zi = aij W j 0 < Z i < 1
j=1
where Zi is the comprehensive index value.

6. Cluster analysis
According to the comprehensive index value and the clustering situation of the
constellation diagram, the classification result is determined.
4.2.2 Example Analysis
It is assumed that there are seven samples (Table 4.1), and each sample has six
indicators: X0, X1, X2, X3, X4, and X5. The seven samples were classified using the
constellation clustering method.
Step 1: Calculate the extreme value of each variable
X 0max = 2:73 X 0min = 2:54

X 1max = 63:19 X 1min = 38:60
X 2max = 67:70 X 2min = 41:10
X 3max = 69:78 X 3min = 24:84
X 4max = 2:61 X 4min = 1:26
X 5max = 1:18 X 5min = 0:88
Step 2: Range transformation

X ij - X j min
Based on the formula aij = X j max - X j min × 180 ° , the transformation data are
obtained:
X 01 - X 0 min 2:64 - 2:54

a01 = × 180 ° = × 180 = 94:74∘
X 0 max - X 0 min 2:73 - 2:54
By analogy, the data in Table 4.2 can be obtained.
Table 4.1 Original data table (Chen 2003, 2023)

Sample 1 2 3 4 5 6 7
X0 2.64 2.72 2.73 2.72 2.64 2.63 2.54
X1 63.19 59.12 46.48 51.06 48.18 38.6 41.03
X2 43.6 41.1 56.9 58.5 62.2 64.5 67.7
X3 69.78 59.44 51.05 43.15 36.68 29.15 24.84
X4 2.61 2.24 1.55 1.48 1.44 1.30 1.26
X5 1.18 1.05 1.04 0.96 0.94 0.88 0.89
Table 4.2 Range transformation data (Chen 2003, 2023)

Sample 1 2 3 4 5 6 7
α0 94.74 170.53 180.00 170.53 94.74 85.26 0.00
α1 180.00 150.21 57.68 91.21 70.13 0.00 17.79
α2 16.92 0.00 106.92 117.74 142.78 158.35 180.00
α3 180.00 138.58 104.98 73.34 47.42 17.26 0.00
α4 180.00 130.67 38.67 29.33 24.00 5.33 0.00
α5 180.00 102.00 96.00 48.00 36.00 0.00 6.00
Step 3: Take weights and perform Cartesian coordinate transformation on each

indicator.
It is assumed that the weights of all indicators are equal, i.e., 1/6.
p p
Based on the formula X i = W j cos aij , Y i = W j sin aij , the following data
j=1 j=1
are obtained:
6
1
X1 = W j cos a1j = × ðcos 94:74 þ cos 180 þ . . . þ cos 180Þ = - 0:52
j=0
6
6
1
Y1 = W j sin a1j = × ðsin 94:74 þ sin 180 þ . . . þ sin 180Þ = 0:21
j=0
6
The rest are similar.
X 2 = - 0:41, Y 2 = 0:51
X 3 = - 0:06, Y 3 = 0:73
X 4 = 0:06, Y 4 = 0:71
X 5 = 0:31, Y 5 = 0:71
X 6 = 0:52, Y 6 = 0:29
X 7 = 0:66, Y 2 = 0:07
Step 4: Draw a constellation diagram (Fig. 4.1)

Step 5: Calculate the comprehensive index value
p
From formula Z, i = aij W j , the following is obtained:
j=1
5
1
Z1 = a1j W j = × ð94:74 þ 180 þ . . . þ 180Þ = 138:61
j=0
6
82 X. Chen
0.8
0.7 Sample 1
0.6 Sample 2
0.5 Sample 3
Sample 4
0.4
Y Sample 5
0.3 Sample 6
0.2 Sample 7
0.1
0
-1 -0.5 0 0.5 1
X
Fig. 4.1 Constellation cluster map (Chen 2003, 2023)
Z 2 = 115:33
Z 3 = 97:37
Z 4 = 88:36
Z 5 = 69:18
Z 6 = 44:37
Z 7 = 33:96
Step 6: Cluster analysis

According to the comprehensive index value and the clustering situation of the
constellation diagram, we believe that it can be divided into three categories: sample
1 and sample 2 are one category; sample 3, sample 4, and sample 5 are one category;
and sample 6 and sample 7 are one category. The comprehensive evaluation values
were 126.97, 84.97, and 39.17, respectively, indicating that the differences between
the three categories were extremely significant.
4.3 Gray Relational Clustering
4.3.1 The Basic Method of Gray Relational Clustering
Gray correlation clustering actually uses the basic principle of gray correlation to
calculate the degree of correlation between samples and then classifies the types of
samples according to the degree of gray correlation. The calculation principle and
method are as follows:
Now, there are m samples, each sample has n indicators, and the following
sequence is obtained:
X 1 = ðx1 ð1Þ, x1 ð2Þ, . . . , x1 ðnÞÞ

X 2 = ðx2 ð1Þ, x2 ð2Þ, . . . , x2 ðnÞÞ
.................................:
X m = ðxm ð1Þ, xm ð2Þ, . . . , xm ðnÞÞ
For all i ≤ j, i, j = 1, 2, . . ., m, calculate the gray correlation degree εij between Xi

and Xj, thus obtaining the upper triangular matrix A.
ε11 ε12 ⋯ ε1m
ε22 ⋯ ε2m
A= , where εii = 1; i = 1, 2, . . ., m;
⋱ ⋮
εmm
If we take the critical value r 2 [0, 1], it is generally required that r > 0.5. When
εij ≥ r, Xi and Xj can be regarded as the same type of features.
r can be determined according to the needs of the actual problem. If r is closer to
1, then the classification is finer, and there are fewer variables in each group. In
contrast, the classification is more, and there are more variables in each group.
4.3.2 Case Analysis
We continue to analyze the examples in the previous section and use the gray
absolute correlation calculation method (Chap. 3) for cluster analysis. In this exam-
ple, there are a total of 7 samples, and each sample has 6 indicators. To save the cost
of future surveys and the collection of sample data, we need to classify the indicators
to streamline the indicators.
Step 1: Zeroing the starting point
Using X 0i ðkÞ = xi ðk Þ - xi ð1Þ, the initialized data in Table 4.3 can be obtained:
Step 2: Seeking |s0|, |si| and |si - sj|
js0 j = 0:19
js1 j = 83:59
js2 j = 77:25
js3 j = 151:92
84 X. Chen
Table 4.3 Initialized data (Chen 2003, 2023)

Serial Sample Sample Sample Sample Sample Sample Sample
number 1 2 3 4 5 6 7
X00 0.00 0.08 0.09 0.08 0.00 -0.01 -0.10
X01 0.00 -4.07 -16.71 -12.13 -15.01 -24.59 -22.16
X02 0.00 -2.50 13.30 14.90 18.60 20.90 24.10
X03 0.00 -10.34 -18.74 -26.63 -33.10 -40.63 -44.95
X04 0.00 -0.37 -1.06 -1.13 -1.17 -1.31 -1.35
X05 0.00 -0.13 -0.14 -0.22 -0.24 -0.30 -0.29
js4 j = 5:72
js5 j = 1:18
js1- s0 j = 83:78
js2- s0 j = 77:06
js3- s0 j = 152:11
js4- s0 j = 5:91
js5- s0 j = 1:37
Step 3: Find the degree of gray absolute correlation
ε01 = 0:50
ε02 = 0:50
ε03 = 0:50
ε04 = 0:54
ε05 = 0:63
ε12 = 0:50
ε13 = 0:78
ε14 = 0:54
ε15 = 0:51
ε23 = 0:49
ε24 = 0:47
ε25 = 0:47
ε34 = 0:52
ε35 = 0:51;
ε45 = 0:64
Then, the absolute matrix A can be obtained.
1 0:50 0:50 0:50 0:54 0:63

1 0:50 0:78 0:54 0:51
1 0:49 0:47 0:47
A=
1 0:52 0:51
1 0:64
1
If we assume that the critical value of the absolute correlation degree is 0.60, then
we can check in turn X5 and X0, X3 and X1, and X5 and X4. Taking the indicator with
the smallest label as the representative of each category, X5 and X4 can be combined
into X0 to form a category so that the clustering results of the six indicators are
fX 5 , X 4 , X 0 g, fX 3 , X 1 g, fX 2 g
In other words, in the future collection of sample data, we only need to collect the
data of three indicators, X0, X1, and X2.
4.4 Gray Variable Weight Clustering

4.4.1 Concept and Method of Gray Variable Weight
Clustering
Suppose there are n clustering objects, m clustering indicators, and s different gray
classes. According to the ith (i = 1, 2, . . ., n) object with respect to the j ( j = 1, 2, . . .,
m) index, the sample value xij (i = 1, 2, . . ., n; j = 1, 2, . . ., m) classifies the ith object
into the k (k 2 {1, 2, . . ., s}) gray class. Among them, it is called gray clustering.
Assume now that the whitening weight function of j index k subclasses f kj ðÞ is the
typical whitening weight function shown in Fig. 4.2, xkj ð1Þ, xkj ð2Þ, xkj ð3Þ, xkj ð4Þ are the
turning point of f kj ðÞ. The typical whitening weight function is denoted as (Liu et al.
2014):
f kj xkj ð1Þ, xkj ð2Þ, xkj ð3Þ, xkj ð4Þ

86 X. Chen
Fig. 4.2 f jk
0
x kj (1) x kj (2) x kj (3) x kj ( 4) x
Fig. 4.3 f jk
0 x kj (3) x kj ( 4) x
Fig. 4.4
f jk
0 x kj (1) x kj (2) x kj ( 4) x
If whitening weight function f kj ðÞ has no first turning point xkj ð1Þ and the second
turning point xkj ð2Þ, as shown in Fig. 4.3, then f kj ðÞ is called the lower limit measure
of the whitening weight function, denoted as f jk - , - , xkj ð3Þ, xkj ð4Þ .
If the whitening weight function f kj ðÞ coincides between the second turning point
of xkj ð2Þ and the third turning point xkj ð3Þ, as shown in Fig. 4.4, then f kj ðÞ is called a
Fig. 4.5
f jk
0 x k (1) x k ( 2) x
j j
moderate measure of the whitening weight function, denoted as

f jk xjk ð1Þ, xjk ð2Þ, - , xkj ð4Þ .
If whitening weight function f kj ðÞ has no third turning point xkj ð3Þ and the fourth
turning point xkj ð4Þ, as shown in Fig. 4.5, then f kj ðÞ is called the upper limit measure
whitening weight function, denoted as f jk xkj ð1Þ, xjk ð2Þ, - , - .
Through the above analysis, we can obtain the whitening weight function under
different conditions.
1. For the typical whitening function shown in Fig. 4.2,
0 2 xjk ð1Þ, xkj ð4Þ
x=
x - xkj ð1Þ
x 2 xkj ð1Þ, xkj ð2Þ
xj ð2Þ - xkj ð1Þ
k
f kj ðxÞ =
1 x 2 xkj ð2Þ, xkj ð3Þ
xkj ð4Þ - x
xkj ð4Þ - xkj ð3Þ
2. For the lower bound measure whitening weight function shown in Fig. 4.3,
0 2 0, xkj ð4Þ
x=
f kj ðxÞ = 1 x 2 0, xkj ð3Þ

xkj ð4Þ - x
3. For the whitening weight function of the moderate measure shown in Fig. 4.4,
there is
88 X. Chen
0 2 xkj ð1Þ, xkj ð4Þ

x=
x - xkj ð1Þ
f kj ðxÞ = x 2 xkj ð1Þ, xkj ð2Þ
xkj ð4Þ - x
xj ð4Þ - xkj ð2Þ
k
4. For the upper bound measure whitening weight function shown in Fig. 4.5, we
have
0 x xkj ð1Þ
x - xkj ð1Þ
f kj ðxÞ = x 2 xkj ð1Þ, xkj ð2Þ
1 x ≥ xkj ð2Þ
For the whitening weight function of the jth index k subcategory shown in
Fig. 4.2, let λkj = 12 xjk ð2Þ þ xjk ð3Þ .
Fig. 4.3, let λkj = xkj ð3Þ.
Fig. 4.4 and Fig. 4.5, let λkj = xkj ð2Þ;
λkj
We call λkj the critical value of subcategory k of index j and call ηkj = m the
λkj
j=1
weight of index j with respect to the k subcategories.

Now let xij be the specimen of object i with respect to index j, f kj ðÞ is the
whitening weight function of j index k subclass, ηkj is the weight of the j index
m
with respect to the k subcategory, then σ ki = f kj xij × ηkj is called the gray variable
j=1
weight clustering coefficient for the object i belonging to k gray class.
m m m
Let σ i = σ 1i , σ 2i , ⋯, σ si = f 1j xij × η1j , f 2j xij , ⋯, f sj xij × ηsj is the
j=1 j=1 j=1
clustering coefficient vector of object i.

Now, set σ ki = max σ ki ; then, it is said that object i belongs to gray class k*.
1≤k≤s
Gray variable weight clustering is applicable to situations in which the meaning
and dimension of the indicators are the same. When the significance and dimension
of the clustering indicators are different and the sample values of different indicators
are very different in number, it is not appropriate to use gray variable weight
clustering.
There are three coastal fishing areas, and the three clustering indicators are the output
value of marine fishing, the output value of aquaculture, and the output value of
aquatic product processing. The sample data are shown in matrix A:
x11 x12 x13 80 20 100

A = xij = x21 x22 x23 = 40 30 30
x31 x32 x33 10 90 60
Clustering was performed based on high, medium, and low output values.
It is now assumed that the whitening weight functions for the indicators of marine
fishing output, aquaculture output, and aquatic product processing output are
f 11 ½0, 80,- , - , f 21 ½0, 40,- , 80, f 31 ½- , - , 10, 20

f 12 ½0, 90,- , - , f 22 ½0, 45,- , 90, f 32 ½- , - , 15, 30
f 13 ½0, 100,- , - , f 23 ½0, 50,- , 100, f 33 ½- , - , 20, 40
From the above whitening weight functions, they are
0 , x<0
x<0 x
0 , , 0 ≤ x ≤ 40
x 40
f 11 ðxÞ = , 0 ≤ x < 80 f 21 ðxÞ =
80 80 - x ,
1 , 40 < x ≤ 80
x > 80 40
0 , x > 80
, x<0
0 x<0
0 ,
1 , 0 ≤ x ≤ 10 x
f 31 ðxÞ = 20 - x f 12 ðxÞ = , 0 ≤ x < 90
90
10 , 10 < x ≤ 20 1 ,
0 x > 90
, x > 20
90 X. Chen
0 , x<0 , x<0
0
x
45 , 0 ≤ x ≤ 45 1 , 0 ≤ x ≤ 15
f 22 ðxÞ = f 32 ðxÞ = 30 - x
90 - x , 45 < x ≤ 90
45 15 , 15 < x ≤ 30
0 , 0 ,
x > 90 x > 30
0 , x<0
x<0
0 , x
50 , 0 ≤ x ≤ 50
x
f 13 ðxÞ = , 0 ≤ x < 100 f 23 ðxÞ =
100 100 - x , 50 < x ≤ 100
1 , 50
x > 100
0 , x > 100
, x<0
0
1 , 0 ≤ x ≤ 20
f 33 ðxÞ = 40 - x
20 , 20 < x ≤ 40
0 , x > 40
Thus, we have
λ11 = 80, λ12 = 90, λ13 = 100

λ21 = 40, λ22 = 45, λ23 = 50
λ31 = 10, λ32 = 15, λ33 = 20
From the formula ηkj = λkj=

3
λkj obtained:
j=1
80 80
η11 = =
80 þ 90 þ 100 270
90 90
η12 = =
80 þ 90 þ 100 270
100 100
η13 = =
80 þ 90 þ 100 270
40
η21 =
135
45
η22 =
135
50
η23 =
135
10
η31 =
45
15
η22 =
45
20
η23 =
45
m
Then, when i = 1, from the formula σ ik = f jk xij × ηjk we have
j=1
m
δ11 = f kj xij × ηkj = f 11 ðx11 Þ × η11 þ f 12 ðx12 Þ × η12 þ f 13 ðx13 Þ × η13
j=1
80 90 100
þ f 12 ð20Þ ×
= f 11 ð80Þ × þ f 13 ð100Þ ×
270 270 270
80 20 90 100
=1× þ × þ 1× = 0:74
270 90 270 270
Similarly, we can obtain:
σ 21 = 0:15
σ 31 = 0:22
σ 12 = 0:37
σ 22 = 0:74
σ 32 = 0:22
σ 13 = 0:59
σ 23 = 0:15
σ 33 = 0:22

Then, the following formula σ ki = max σ ki is obtained:
1≤k≤s
max σ k1 = max δ11 , δ21 , δ31 = maxf0:74, 0:15, 0:22g = δ11

1≤k≤3
92 X. Chen

1≤k≤3

1≤k≤3
The above results indicate that the second fishing area is a moderately developed
area of the fishery economy, and the first and third fishing areas are highly developed
areas.
4.5 Gray Fixed-Weight Clustering

4.5.1 General Method of Gray Fixed-Weight Clustering
When the clustering indicators have different meanings and dimensions and there is
a large disparity in the number, if we adopt gray variable weight clustering, the effect
of some indicators in the clustering may be very weak. There are two methods to
solve this problem: one is to use the raw data processing methods in Chap. 2 (such as
the initial value or average) for dimensionless processing and then perform cluster-
ing. This method treats all clustering indicators equally and cannot reflect the
difference in the role of different indicators in the clustering process. Another
method is to assign weights to each clustering index in advance. There are many
methods to assign weights, and the analytic hierarchy process (AHP) is generally
used. This method is a clustering method that assigns weights in advance, so it is
called gray fixed-weight clustering.
Gray fixed-weight clustering can be performed according to the following steps:
Step 1: Give the whitening weight function of the jth index k subcategory f kj ðÞ
(J = 1, 2, . . ., m; k = 1, 2, . . ., s).
Step 2: Determine the clustering weight ηj ( j = 1, 2..., m) of each index according
to the qualitative analysis conclusion.
Step 3: Use the whitening weight function obtained in Steps 1 and 2 f kj ðÞ (J = 1,
2, . . ., m; k = 1, 2, . . ., s), clustering weight ηj ( j = 1, 2, . . ., m), and the sample value
xij of object i with respect to j index (i = 1, 2, . . ., n; j = 1, 2, . . ., m) to calculate the
m
fixed-weight clustering coefficient σ ki = f jk xij × ηj , I = 1, 2, . . ., s.
j=1

Step 4: If σ ki = max σ ki , then it is concluded that object i belongs to gray class
1≤k≤s
k *.
We also illustrate the examples in the previous section.

Step 1: Determine the whitening weight function as in the previous section.

Step 2: Determine the cluster weights of fishing, aquaculture, and processing
indicators.
η1 = 0:50, η2 = 0:30, η3 = 0:20

m
Step 3: From the formula σ ki = f jk xij × ηj , calculate
j=1
3
σ 11 = f 1j x1j × ηj = f 11 ðx11 Þ × η1 þ f 12 ðx12 Þ × η2 þ f 13 ðx13 Þ × η3
j=1
= f 11 ð80Þ × 0:5 þ f 12 ð20Þ × 0:3 þ f 13 ð100Þ × 0:2

20
= 1 × 0:5 þ × 0:3 þ 1 × 0:2 = 0:756
90
Similarly, we can obtain:
σ 31 = 0:20
σ 12 = 0:41
σ 22 = 0:82
σ 32 = 0:10
σ 13 = 0:48
σ 23 = 0:29
σ 33 = 0:50

Then, the following formula σ ki = max σ ki is obtained:
1≤k≤s

1≤k≤3

1≤k≤3

1≤k≤3
The above results indicate that the second fishing area belongs to an area with a
moderately developed fishery economy, the first one belongs to a highly developed
area, and the third fishing area belongs to an underdeveloped area of fisheries.
94 X. Chen
4.6 Analysis of the Application of Gray Clustering

in Fishery Science
Gray clustering has been widely used in fishery science, mainly in the aspects of
fishery regional economy, fish population division, fish nutritional value evaluation,
and fishery water environment evaluation. The application of gray clustering in
fishery science is specifically analyzed based on the reports in the relevant literature.
4.6.1 Application of Gray Clustering in the Fishery Regional

Economy
In the study of the fishery regional economy, it is very important to formulate future
development strategies based on the natural resource conditions of each region and
the development level of the fishery economy. Therefore, it is very important to use
the gray clustering method to scientifically classify the fishery economy in China’s
coastal areas to formulate different types of development plans and implement
classification guidance to ensure the sustainable development of the fishery economy
in coastal areas. Meaning.
Chen and Zhang (2001) published a preliminary study on the types of fishery
economic regions in China’s coastal provinces and cities. The statistical data of
coastal fishery development in 11 provinces in China in 1997 were selected, namely
Tianjin, Hebei, Liaoning, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guang-
dong, Guangxi, and Hainan. Based on the characteristics and development of the
aquaculture industry, a comprehensive technical and economic evaluation index for
fisheries was established, and 17 indicators were used. There are the total production
of aquatic products (tons), marine fishing production (tons), marine aquaculture
production (tons), freshwater fishing yield (ton), freshwater aquaculture yield
(ton), offshore fishery yield (ton), total output value (10,000 yuan), aquatic product
processing yield (ton), per capita aquatic product yield (ton/person), per capita total
fishery output value (yuan/person), per capita net income of fishermen (yuan/per-
son), per capita net income of fishermen (yuan/person), per capita investment in
fixed assets (yuan/person), average per unit yield of seawater and freshwater aqua-
culture (kg/ha), the water surface utilization rate of mariculture (%), and the water
surface utilization rate of freshwater aquaculture (%). Gray constellation clustering
was used to preliminarily classify the types of fishery economic regions in 11 coastal
provinces in China.
In the calculation, the weight of each indicator is set to 1/17, and the values of the
abscissa (Xi) and the ordinate (Yi) of each province are calculated, as shown in
Table 4.4. The constellation clustering diagram is plotted in Fig. 4.6. The calculated
comprehensive index value (Z) is shown in Tables 4.5.
According to the comprehensive index values of the 11 coastal provinces and the
clustering distribution map (Fig. 4.6), fishery economic development can be divided
Table 4.4 Cluster analysis results (Chen and Zhang 2001)

Sample point Tianjin Hebei Liaoning Shanghai Jiangsu Zhejiang
X value 0.5263 0.5896 0.2052 0.4064 0.3354 0.0829
Y value 0.3396 0.4568 0.6318 0.3548 0.4102 0.6794
Sample point Fujian Shandong Guangdong Guangxi Hainan
X value 0.0276 -0.3990 0.0224 0.6367 0.7256
Y value 0.5765 0.5855 0.5708 0.4812 0.3004
1 Tianjin
0.8 Hebei
0.6 Liaoning
0.4 Shanghai
0.2 Jiangsu
Y
0 Zhejiang
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
-0.2 Fujian
-0.4 Shandong
-0.6 Guangdong
-0.8 Guangxi
-1 Hainan
X
Fig. 4.6 Gray clustering results of comprehensive index values of various provinces and cities
(Chen and Zhang 2001)
Table 4.5 Comprehensive indicator values of the fishery economy in various provinces and cities
(Chen and Zhang 2001)
Province Tianjin Hebei Liaoning Shanghai Jiangsu Zhejiang
Comprehensive value 0.5263 0.5896 0.2052 0.4064 0.3354 0.0829
Province Fujian Shandong Guangdong Guangxi Hainan
Comprehensive value 0.0276 -0.3990 0.0224 0.6367 0.7256
into four regions. Among the four provinces, category III includes five provinces and
municipalities, namely Jiangsu, Shanghai, Guangxi, Hebei, and Tianjin, and cate-
gory IV includes Hainan Province (Table 4.6).
4.6.2 Application of Gray Clustering in the Evaluation

of the Nutritional Value of Fish
Xie and Pan (1992) published “A Comprehensive Evaluation of the Nutritional

Value of Several Famous and High-quality Fishes.” The authors used the protein
96
Table 4.6 Results of the division of fishery economic zones (Chen and Zhang 2001)
Type Province Z value Type Province Z value Type Province Z value Type Province Z value
Class I Shandong 0.684 Class II Guangdong 0.514 Class III Jiangsu 0.356 Class IV Hainan 0.163
Class II Fujian 0.496 Class III Shanghai 0.308
Class II Zhejiang 0.465 Class III Hebei 0.253
Class II Liaoning 0.441 Class III Tianjin 0.253
Class III Guangxi 0.241
X. Chen
Table 4.7 Evaluation criteria Evaluation parameters Low Medium High

for the nutritional value of
Protein content (%) ≤10 15 ≥20
freshwater fish in China (Xie
and Pan 1992) Fat content (%) ≤1 3 ≥10
Essential Amino Acid Index (%) ≤20 50 ≥80
Flavored amino acid content (%) ≤10 20 ≥30
Table 4.8 Nutritional composition of six fish species (Xie and Pan 1992)
Red Nile Channel Freshwater Southern Grass
Item Eel tilapia catfish pomfret catfish carp
Protein content (%) 16.51 16.94 16.31 18.75 15.28 17.53
Fat content (%) 18.31 0.95 2.99 6.68 1.48 1.8
Essential Amino Acid 55.43 70.87 83.79 61.54 63.33 42.76
Index (%)
Flavored amino acid 19.78 29.34 32.62 25.22 19.57 16.77
content (%)
content, fat content, essential amino acid index of protein, and flavored amino acid
content, which can represent the nutritional value of fish. The nutritional value of eel,
red Nile tilapia, channel catfish, freshwater white pomfret, southern catfish, and
grass carp was evaluated using gray clustering to compare their nutritional values.
The proposed evaluation criteria are shown in Tables 4.7. The specific calculation
steps are as follows:
4.6.2.1 Determination of Clustering Factors
According to the items listed in Table 4.7, the cluster analysis uses four factors,
namely x = {x1, x2, x3, x4} = {protein content, fat content, Essential Amino Acid
Index, flavored amino acid content}. The level of nutrition value was divided into
three levels, with low, medium, and high denoted as I, II, and III, respectively.
4.6.2.2 Dimensionless Calculation
For comparison, the influence of the dimensions in Tables 4.7 and 4.8 needs to be
eliminated. Therefore, dimensionless calculation is needed. The calculation results
are shown in Tables 4.9.
M j = max xij i = 1, 2, ⋯9
dij = xij =M j j = 1, 2, ⋯4
98
Table 4.9 Nutritional components and standards of the six fish species after nondimensionalization (Xie and Pan 1992)
Channel Freshwater Southern Grass
Protein content (%) Eel Red Nile tilapia Catfish pomfret catfish carp I II III
Fat content (%) 0.826 0.847 0.816 0.938 0.764 0.877 0.6 0.75 1
Essential Amino Acid Index(%) 1 0.052 0.163 0.365 0.081 0.098 0.055 0.164 0.546
Flavored amino acid content (%) 0.662 0.846 1 0.735 0.756 0.51 0.239 0.597 0.955
Protein content (%) 0.606 0.899 1 0.773 0.599 0.514 0.307 0.613 0.919
X. Chen
4.6.2.3 Determine the Whitening Functions of Various Types
According to Table 4.9, the threshold λjk of the whitening function is determined,
and the maximum value of fji is 1.
4.6.2.4 Find the Clustering Weight of Each Item
For example, for the protein content, the low trophic level weight is
1=λ11 1=0:6
η11 = =
1=λ11 þ 1=λ21 þ 1=λ31 þ 1=λ41 1=0:6 þ 1=0:055 þ 1=0:239 þ 1=0:613
= 0:065
The rights of other categories can also be obtained.
4.6.2.5 Calculate the Clustering Coefficient
Let the coefficient of the ith sample be σ ik
4
σ ik = f jk dij ηjk
j=1
For example, for eels, the clustering coefficient of the low trophic level is σ 11
σ 11 = f 11 ðd 11 Þ η11 þ f 21 ðd12 Þ η21 þ f 31 ðd 13 Þ η31 þ f 41 ðd14 Þ η41 = 0 þ 0 þ 0:606

× 0:118 × ð1- 0:977ÞÞ þ 0 = 0:002
where fjk is obtained by interpolation according to the whitening function. Similarly,

the clustering coefficients of eels with respect to other nutritional levels can be
obtained.
σ 12 = 0:246
σ 13 = 0:445
4.6.2.6 Construct a Clustering Vector and Classify It According

to the Maximum Principle
For eel, its clustering vector σ 1

100 X. Chen
Table 4.10 Clustering coefficients and clustering results of each sample (Xie and Pan 1992)
Red Nile Channel Freshwater Southern Grass
Grade Eel tilapia catfish pomfret catfish carp
I 0.002 0.034 0.088 0 0.044 0.08
II 0.246 0.113 0.091 0.254 0.253 0.189
III 0.445 0.374 0.473 0.361 0.079 0.089
Clustering III III III III II II
σ 1 = ð0:002, 0:246, 0:445Þ
σ 13 = 0.445 is the maximum, so the eels belong to the high trophic level.
Clustering coefficient of various fishes, the results of σ jk and clustering are shown
in Table 4.10.
The clustering results showed that, except for the mesotrophic level of southern
catfish and carp, all the other fish were of high trophic level. From high to low:
Channel catfish (σ 3 = 0.473) > Eel (σ 1 = 0.445) > Red Nile tilapia
(σ 2 = 0.374) > Freshwater Pomfret (σ 4 = 0.361) > Southern catfish
(σ 5 = 0.253) > silver carp (σ 6 = 0.189).
The clustering results also showed that the nutritional value of the five famous
and high-quality fish was higher than that of the carp, and the nutritional value of the
middle-class southern catfish was also higher than that of the carp. Among the five
famous fish species, the channel catfish was the highest (σ 3 = 0.473), and the eel was
slightly lower than that (σ 1 = 0.445). Red Nile tilapia and freshwater pomfret are
relatively close to each other (σ 2 = 0.374, σ 4 = 0.361).
In addition, according to the survey, the current sales price of aquatic products in
China is basically as follows: eel > channel catfish > red Nile tilapia > freshwater
white pomfret > southern catfish > grass carp, indicating that the market value of
fish mainly depends on the nutritional value. However, the price of eel is higher than
that of channel catfish, which may be due to the following factors: Japan has a large
demand for eels, and it is impossible to artificially breed eels; different measurement
conditions may also cause inconsistency; the size of each type of fish and different
feeding methods and different measurement seasons may cause some errors.
In addition to the above components, the nutritional value of fish also includes a
variety of vitamins and inorganic salts. Therefore, in future evaluations of the
nutritional value of fish, more factors should be selected to meet the actual situation.
4.6.3 Application of Gray Clustering Analysis

in the Environmental Assessment of Fishery Waters
Liu and Xiong (2000) published “The Application of Gray Clustering Analysis in
the Scientific Management of Reservoir Water Quality.” This paper selected TP, TN,
and COD as the evaluation factors to evaluate the fishery water quality of Shihe
Table 4.11 Water quality evaluation criteria (Liu and Xiong 2000)
II (Poor IV
Evaluation I (Very poor trophic III (Medium (Eutrophic V (Extremely
factor trophic level) level) trophic level) level) eutrophic level)
TP/μg.L-1 <1 4 23 110 >600
TN/mg.L-1 <0.02 0.06 0.31 1.20 >4.60
COD/mg.L-1 <0.09 0.36 1.80 7.10 >27.10
Table 4.12 Nondimensional water quality evaluation standards (Liu and Xiong 2000)
II (Poor IV
TP/μg.L-1 <0.009 0.036 0.209 1 6
TN/mg.L-1 <0.017 0.05 0.258 1 3.833
COD/mg.L-1 <0.013 0.051 0.254 1 3.817
Reservoir and to grasp the water quality status. The fishery water quality evaluation
criteria used in this study are shown in Tables 4.11. The specific calculation steps are
as follows:
4.6.3.1 Dimensionless Processing
First, the gray class (evaluation level) and the original whitening number (measured
data of each factor) are dimensionlessly processed to facilitate comparison and
eliminate the effect of dimension. The dimensionless processing formula of the
gray class is λij = aij/aj, i = 1, 2, 3; j = 1, 2, 3, 4, 5. In the formula, λij represents
the standard value of the whitening number (measured data) after dimensionless
treatment; ai represents the reference standard of the ith factor. Here, the standard
value of grade IV of each factor is used as the corresponding ai value. The
nondimensional water quality standards are shown in Table 4.12.
4.6.3.2 Calculating the Clustering Weight

n
Use the formula: wij = λij = λiij The weight of each factor with respect to each
i=1
level is obtained (Table 4.13).
4.6.3.3 Calculation of the Clustering Coefficient
The clustering coefficient of the kth clustering object at the jth level is
102 X. Chen
Table 4.13 The weight of each factor with respect to each level (Liu and Xiong 2000)
II (Poor IV
TP/μg.L-1 0.231 0.263 0.290 0.333 0.440
TN/mg.L-1 0.456 0.365 0.358 0.333 0.281
COD/mg.L-1 0.333 0.372 0.352 0.333 0.280
Table 4.14 Clustering coefficients and results (Liu and Xiong 2000)
I (Very poor II (Poor III (Medium IV (Eutrophic V (Extremely
trophic level) trophic level) trophic level) level) eutrophic level)
0 0 0.770 0.223 0
n
σ ij = f ij ðd ki Þwij
i=1
where k = 1 and j = 1, 2, 3. The method of obtaining fij(dki)is described in Liu and

Xiong (2000). The calculation results are shown in Table 4.14.
4.6.3.4 Evaluation Results
The horizontal clustering quantity of Shihe Reservoir is σ = (0, 0, 0.770, 0.223, 0),
where σ 13 = 0.770. it is a medium trophic level.
4.6.4 Application of Gray Clustering in the Classification

of Fish Populations
Chen et al. (2002) used the gray variable weight clustering method to preliminarily
classify the population structure of Ommastrephes bartramii in the Northwest
Pacific. In this study, a total of 120 squid samples were randomly collected in the
waters west of 165°E in the northwestern Pacific Ocean from October to November
1999. The mantle length (ML), fin length (Q1), fin width (Q2), eye diameter (Y),
length of the right first arm (WN1), length of the right second arm (WN2), length of
the right third tentacle (WN3), length of the right fourth tentacle (WN4), and length
of the right tentacle spike (SL) were determined. The cluster analysis method of gray
variable weights was used to study the population structure of Ommastrephes
bartramii in the northwestern Pacific Ocean.
Table 4.15 Clustering results of each sample (Chen et al. 2002)

Category Sample Total
Class I 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 99
25, 26, 29, 30, 31, 32, 33, 35, 38, 41, 42, 45, 46, 47, 48, 49, 50, 52, 53, 54,
55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 66, 68, 69, 70, 71, 73, 74, 75, 76, 77,
78, 79, 80, 81, 82, 83, 84, 85, 86, 88, 89, 91, 92, 93, 94, 95, 96, 97, 98, 99,
100, 101, 102, 103, 104, 105, 106, 107, 108, 110, 112, 115, 116, 117,
118, 120
Class II 24, 27, 28, 34, 36, 37, 39, 40, 43, 44, 51, 65, 67, 72, 87, 90, 109, 111, 113, 21
114, 119
According to the 9 collected morphological characteristic indexes, the ratios of

other metrics to mantle length were obtained, i.e., Q1/ML, Q2/ML, Y/ML, W1/ML,
W2/ML, W3/ML, W4/ML, and S/ML, and a matrix was constructed with 8 indicators
as columns and 120 samples as rows.
d1,1 ð0:425Þ d1,2 ð0:766Þ ⋯ d 1,8 ð0:492Þ

d2,1 ð0:428Þ d2,2 ð0:736Þ ⋯ d 2,8 ð0:553Þ
d ij =
⋮ ⋮ ⋮ ⋮
d120,1 ð0:444Þ d 120,2 ð0:791Þ ⋯ d120,8 ð0:528Þ
Three indicators of high, medium, and low levels were used to set the gray-like
whitening function. The clustering coefficient was obtained, and the clustering
coefficient vector was constructed. The clustering results are shown in Table 4.15.
The results of the study are as follows:
1. The gray variable weight clustering method was used to analyze the population
structure of Ommastrephes bartramii in the northwestern Pacific Ocean
(Table 4.15).
2. The calculation showed that there was a significant difference in the average
values of the eight morphological characteristics of the two populations
(Table 4.16). The variation in the coefficient of variation was between 2.54%
and 9.15%. The most significant indicator was the ratio of the length of each arm
to the mantle length, while the coefficient of difference in the ratio of fin length
and fin width to mantle length was relatively small.
3. The conclusions reached are basically the same as those of previous studies. The
use of morphological analysis methods requires quantitative biological measure-
ment, and sampling and measurement are simple and easy to implement. At the
same time, using the gray clustering analysis method, its mathematical processing
method is relatively simple and scientific. However, the morphological feature
value is easily affected by environmental factors, which reduces the stability of
the feature itself and affects the credibility of the identification results (Chen et al.
2002).
104 X. Chen
Table 4.16 Average morphological characteristics of the two groups (Chen et al. 2002)
Class I Indicator SQ1/SML SQ2/SML S Y/SML SW1/SML
Mean value 0.4372 0.7717 0.0785 0.4929
Class II Indicator SQ1/SML SQ2/SML S Y/SML SW1/SML
Mean value 0.4223 0.7521 0.0723 0.4545
Coefficient of difference 3.41% 2.54% 7.90% 7.80%
Class I Indicator SW2/SML SW3/SML SW4/SML SSL/SML
Mean value 0.5847 0.6256 0.5846 0.6016
Class II Indicator SW2/SML SW3/SML SW4/SML S SL/SML
Mean value 0.5449 0.5801 0.5311 0.5561
Coefficient of difference 3.41% 6.81% 7.27% 9.15%
References
(In Chinese)
(In Chinese)
Chen XJ, Zhang XG (2001) Preliminary discussion on regional types of Chinese fisheries economy
in coastal provinces. J Shanghai Fish Univ 10(2):183–186. (In Chinese)
Chen XJ, Tian SQ, Ye XC (2002) Study on population structure of flying squid in Northwestern
Pacific based on gray system theory. J Shanghai Fish Univ 11(4):335–341. (In Chinese)
Liu DQ, Xiong BX (2000) Application of gray cluster analysis in scientific management of
reservoir water quality. J Hydroecol 20(3):41–47. (In Chinese)
(In Chinese)
Xie J, Pan HJ (1992) Gray system comprehensive evaluation of nutritional value of several famous
fishes. Zhujiang Aquatic (19):67–72. (In Chinese)
Chapter 5
Basic Principles of Gray Dynamic Modeling
Xinjun Chen
Abstract Gray dynamic modeling is one of the important components of gray

system theory. Gray system is a kind of system which contains not only known
information, but also unknown information or non-known information. One of the
important contents of studying the gray system is how to abstract and build a model
from an ambiguous system with insufficient overall information, which can make the
factors of the gray system from unclear to clear, from little knowledge to more
knowledge provides the basis for research. In the theory of gray system, a model that
describes the continuous changing process of things in a gray system is established
by generating and transforming the original data sequence that represents the
behavior characteristics of the gray system, called the gray model. Gray models
are the long-term fuzzy description of the development of things by establishing the
gray differential prediction model with a little and incomplete information. The basic
idea of its modeling is to use the original data to form the original sequence (0) and to
generate the sequence (1) by the method of accumulation. The generated sequence
(1) can weaken the randomness of the original data and make it present more obvious
characteristic law. The differential equation model is creat by use of the generated
sequence (1), i.e. GM model. The GM (1,1) model represents a first-order,
one-variable differential equation model. The gray model usually adopts the “five-
step modeling” method, which includes system qualitative analysis, factor analysis,
preliminary quantification, dynamic quantification and optimization, it has the
advantages of not needing a large number of samples and not requiring regular
distribution of samples. Generally speaking, social system, economic system, eco-
logical system, and fishery system are all gray systems. For example, there are many
factors that lead to the decline of fishing yield, but not many are known, so the gray
forecasting method can be used for the prediction of the gray system of fishing yield.
In this chapter, we will mainly introduce the technical route of gray modeling,
X. Chen (✉)
106 X. Chen
including the principle of gray dynamic modeling and the modeling process and
steps of common GM (1, 1) and GM (1, n) models.
Keywords Gray dynamic modeling · GM (1,1) · GM (1, N ) · Gray system
5.1 Technical Route of Gray Modeling
5.1.1 Principle of Gray Dynamic Modeling
Gray predictive modeling is based on the concept of gray modules. Gray system
theory believes that all random quantities are gray quantities and gray processes that
vary within a certain range and within a certain period of time. The processing of the
gray quantity is not to seek its statistical law and probability distribution but to find
the law from the irregular raw data, that is, to process the data in a certain way to
make it into more regular time series data. Then, the model is established. This is
because in an objective system, no matter how complex it is, there is always
correlation, overall function, and order within the system. Therefore, as the data
that characterize the system’s behavior, it always contains a certain pattern. The
sequence data generated by processing in a certain way are called “modules.” Its
geometric meaning is the general term formed by the continuous curve and its bottom
(i.e., the abscissa) of the generated sequence data on the two-dimensional plane of time
and data. We call the module composed of known data columns the white module, and
the module that is extrapolated from the white module to the future, that is, the module
composed of the predicted value, is called the gray module.
Under normal circumstances, for a given raw data column
ð1Þ ð2Þ ð3Þ ðN Þ

X ð0Þ = xð0Þ , xð0Þ , xð0Þ , . . . , xð0Þ
It cannot be directly used for modeling because these data are mostly random and
irregular. If the original data column is generated by one accumulation, a new data
column can be obtained.
ð1Þ ð2Þ ð3Þ ðN Þ

X ð1Þ = xð1Þ , xð1Þ , xð1Þ , ⋯, xð1Þ
i
ðiÞ ðiÞ
where xð1Þ = xð0Þ
k=1
The newly generated data sequence is a monotonic growth curve, which obvi-
ously enhances the regularity of the original number sequence, while the randomness
is weakened. For a nonnegative data series, the greater the number of accumulations,
the more obvious the weakening of randomness, and the stronger the regularity, so it
is easier to approximate the exponential function. The processed data weaken the
randomness of the original data series, thereby finding the regularity of its variation
and providing intermediate information for the establishment of dynamic models.
5 Basic Principles of Gray Dynamic Modeling 107
For biology, social economics, and environmental ecosystems, the application of

differential equations facilitates the description of the dynamic characteristics of
their internal physical or chemical processes. Gray system theory can be used to
establish a differential equation model because gray system theory treats a random
quantity as a gray quantity that changes within a certain range and treats a stochastic
process as a gray process that changes in a certain range and a certain time zone.
Second, gray system theory generates irregular raw data and turns it into a more
regularly generated sequence of numbers for modeling.
Therefore, gray GM modeling is actually a generated data model, while the
original data model is generally used for modeling. In addition, gray system theory
adjusts, corrects, and improves the accuracy of the model through different gener-
ations of gray numbers, different selections of data, and the supplementation of
different levels of residual models.
5.1.2 Common GM (n, h) Model
The GM (n, h) model refers to a differential equation with n-order h variables.

Different GM models with n and h have different meanings and uses. The GM model
can be roughly classified into the following two categories.
5.1.2.1 GM (n, 1) Model
The GM (n, 1) model is commonly used, that is, the GM model with only one variable.
The requirement for the data column is the time series of the “comprehensive effect.”
Because the larger n is, the more complex the calculation, the accuracy may not be
higher. Therefore, in general, the value of n is below the third order. The most
commonly used n = 1-order model is simple in calculation and has wide applicability.
ð 1Þ
The differential equation of the GM (1, 1) model is dxdt þ axð1Þ = u.
Coefficient vector a = ½a, μT .
The corresponding time function is xð1Þ ðt þ 1Þ = xð0Þ ð1Þ - ua e - at þ ua.
After derivation and reduction, we can obtain xð0Þ ðt þ 1Þ = - a xð0Þ ð1Þ - ua e - at.
The above two equations are the basic calculation formulas for the gray prediction
of the GM (1, 1) model.
GM (2, 1) is a second-order model with two characteristic roots. Its dynamic
process can reflect different situations; that is, it may be monotonic, nonmonotonic,
or oscillating.
The differential equation of the GM (2, 1) model is
d 2 xð1Þ dxð1Þ
þ a 1 þ a2 xð1Þ = u
dt 2 dt
Coefficient vector a = ða1 , a2 , uÞT The time response function is

108 X. Chen
u
xð1Þ ðt Þ = C1 eλ1 t þ C2 eλ2 t þ
a2
where λ1,λ2 are the two characteristic roots, and the main dynamic characteristics of
the system can be analyzed according to the following different situations (Deng
1990; Liu et al. 2014).
1. If λ1 = λ2, then the dynamic process is monotonic.
2. If λ1 ≠ λ2 is a real number, the dynamic process may be nonmonotonic.
3. If λ1.λ2 is the conjugate complex root, then the dynamic process is cyclically
oscillating.
5.1.2.2 GM (1, h) Model
The GM (1, 1) and GM (2, 1) described above are generally used for prediction. As a
state analysis model, the GM (1, h) model is commonly used, which can reflect the
effect of h - 1 variables on the first derivative of the dependent variable. Since
h > 1, it is called the first-order linear dynamic model of h sequences. The modeling
steps are as follows (Deng 1990; Chen 2003; Liu et al. 2014):
Let h variables X1, X2, . . ., Xh form the original sequence of numbers xi(0) = {xi(0)
(1), xi(0) (2), . . ., xi(0) (N )} (i = 1, 2, . . ., h). A new sequence of numbers is obtained
by accumulating Xi(0) once:
X i ð1Þ = xi ð1Þ ð1Þ, xi ð1Þ ð2Þ, . . . , xi ð1Þ ðnÞ ðI= 1, 2, . . . , hÞ
Establishment of the differential equation:
dx1 ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ

þ ax1 = b1 x2 þ b2 x3 þ ⋯ þ bh - 1 xh
dt
Coefficient vector a = ðb1 , b2 , . . . , bh - 1ÞT is solved using the least squares

method, i.e.,
-1 T
a = BT B B YN
where B is the cumulative matrix and YN is the vector of the constant terms:
1 ð1Þ
- x ð1Þ þ xð1Þ ð2Þ ð 1Þ
x2 ð2Þ ⋯
ð 1Þ
xh ð2Þ
2
1 ð1Þ
B= - x ð2Þ þ xð1Þ ð3Þ ð 1Þ
x2 ð3Þ ⋯
ð 1Þ
xh ð3Þ
2 ⋯ ⋯ ⋯ ⋯
1 ð1Þ
- x ðn - 1Þ þ xð1Þ ðnÞ ð 1Þ
x2 ðnÞ ⋯
ð 1Þ
xh ðnÞ
2
T
Y N = x1 ð0Þ ð2Þ, x1 ð0Þ ð3Þ, . . . , x1 ð0Þ ðnÞ
Then, the solution of the differential equation can be obtained:
h h
ð1Þ ð0Þ bi - 1 ð 1 Þ bi - 1 ð 1 Þ
x 1 ð t þ ! Þ = x 1 ð 1Þ - x ðt þ !Þ e - at þ x ðt þ !Þ
i=2
a i i=2
a i
5.2 GM (1, 1) Model
The GM (1, 1) model is actually the prediction of the gray series, which carries out
the quantitative prediction of the time series data, such as population prediction,
labor prediction, output prediction, output value prediction, and various trend pre-
dictions. Its future development is predicted. This type of prediction is not only
widely used but also has universal significance. To this end, we provide a more
detailed introduction.
5.2.1 The Basic Steps of Establishing the GM (1, 1) Model
Step 1: The data sequence X(0) = {x(0) (1), x(0) (2), . . ., x(0) (N )} is accumulated and
generated once to obtain
X ð1Þ = xð1Þ ð1Þ, xð1Þ ð2Þ, . . . , xð1Þ ðN Þ
t
where xð1Þ ðt Þ = xð0Þ ðk Þ
k=1
Step 2: Construct the accumulation matrix B and the constant term vector YN, i.e.,
1
1 ð1Þ
- x ð1Þ þ xð1Þ ð2Þ
2 1
1 ðÞ
- x ð2Þ þ xð1Þ ð3Þ
B= 2
⋮
⋮
1 ð1Þ
- x ðN - Þ þ xð1Þ ðN Þ
2
1
110 X. Chen
T
Y N = x1 ð0Þ ð2Þ, x1 ð0Þ ð3Þ, . . . , x1 ð0Þ ðN Þ
Step 3: Solve the gray parameters using the least squares method a
a -1 T
a= = BT B B YN
u
Step 4: Substitute the gray parameters into the time function:
u - at u
xð1Þ ðt þ 1Þ = xð0Þ ð1Þ - e þ
a a
ð1Þ
Step 5: Taking the derivative of X , we get the following
u - at
xð0Þ ðt þ 1Þ = - a xð0Þ ð1Þ - e
a
or
xð0Þ ðt þ 1Þ = xð1Þ ðt þ 1Þ - xð1Þ ðt Þ
Step 6: Calculating the difference ε(0) (t) between x(0)(t) and xð0Þ ðt Þ, and the
relative error e (t)
εð0Þ ðt Þ = xð0Þ ðt Þ - xð0Þ ðt Þ

e ðt Þ = εð0Þ ðt Þ=xð0Þ ðt Þ
Step 7: Testing the model accuracy and application of the model for prediction.
To analyze the reliability of the model, the accuracy of the model must be tested.
Currently, the most common diagnostic method is to perform a posterior error test on
the model. That is, the deviation s1 of the observation data is first calculated:
m 2
s21 = xð0Þ ðt Þ - xð0Þ ðt Þ
t=1
Residual deviation s2:
m-1 2
1
s22 = qð0Þ ðt Þ - qð0Þ ðt Þ
m-1 t=1
Then, calculate the posterior ratio: c = ss12 and the probability of small error:
p = qð0Þ ðt Þ - qð0Þ < 0:6745s1
The model was diagnosed based on the posterior ratio c and the small error
probability p. When p > 0.95 and c < 0.35, the model can be considered reliable and
can be used for prediction. At this time, the system behavior can be predicted based
on the model.
The above seven steps are the entire modeling and prediction analysis process.
When the residual of the established model is large and the accuracy is not ideal, to
improve the accuracy, residual GM (1, 1) model modeling analysis should be
performed to correct the prediction model.
5.2.2 Different Forms of the GM (1, 1) Model
Based on the results of Professor Deng’s research on the GM (1, 1) model, a variety
of different forms of the GM (1, 1) model are proposed, and the main ones are listed
as follows: (Deng 1990; Chen 2003; Liu et al. 2014):
xð0Þ ð2Þ = β - αxð0Þ ð1Þ
1.
x ðk Þ = ð1 - αÞxð0Þ ðk - 1Þ; k = 3, 4, ⋯n
ð0Þ

2.
x ðk Þ = β - αxð0Þ ðk - 1Þ; k = 3, 4, ⋯n
ð0Þ

ð1Þ
3. x ðk - 1Þ - 0:5b ð0Þ
xð0Þ ðkÞ = ð1Þ x ðk - 1Þ; k = 3, 4, ⋯n
x ðk - 2Þ þ 0:5b
k-2
b - axð0Þ ð1Þ
4. xð0Þ ðk Þ = 1 - 0:5a
1þ0:5a 1þ0:5a ;k= 2, 3, . . ., n
ð0Þ ð0Þ klnð1 - αÞ
5. x ðk Þ = 1
ð 1 - αÞ 3
x ð 3Þ e ,k>3
-α(k - 2)
6. x(0)(k) = (β - αx (1))e (0)
7. xð0Þ ðk Þ = ð1- ea Þ xð0Þ ð1Þ - b

a e - aðk - 1Þ
ð0Þ
8. x ðk Þ = ð- aÞ xð0Þ ð1Þ - b
a e - aðk - 1Þ
5.2.3 Interpretation and Analysis of the Development

Coefficient-a Value
The parameter -a in the GM (1, 1) model is usually referred to as the development

coefficient, and the parameter b is referred to as the gray effect (Deng 1990; Liu et al.
2014).
-a reflects xð0Þ and xð1Þ the development trend. In general, the system action
quantity should be exogenous or predetermined, while GM (1, 1) is a single-
sequence modeling, and only the behavior sequence of the system (called the output
112 X. Chen
sequence) is used, and there is no external action sequence (or called the input
sequence). In GM (1, 1), the gray effect is the data mined from the background value,
which reflects the relationship of data changes, and its exact connotation is gray. The
existence of gray action quantity is a concrete manifestation of connotative exten-
sion. Its existence is the watershed that distinguishes gray modeling from general
input–output modeling and is also an important sign to distinguish the gray system
view from the gray box view.
In addition, Liu et al. (2014) conducted an in-depth study on the development
coefficient -a in the GM (1, 1) model by analyzing the simulation error and
prediction error of GM (1, 1), and the magnitude and value of the development
coefficient -a were compared. The prediction accuracy of the system and its possible
application are discussed, and the following conclusions are drawn:
1. When -a ≤ 0.3, GM (1, 1) can be used for medium- and long-term prediction;
2. When 0.3 < -a ≤ 0.5, it can be used for short-term prediction, and medium- and
long-term prediction can be used with caution;
3. When 0.5 < -a ≤ 0.8, caution should be taken when using GM (1,1) for short-
term prediction;
4. When 0.8 < -a ≤ 1, the residual correction GM (1, 1) model should be used;
5. When -a > 1, the GM (1, 1) model should not be used.
5.3 GM (1, n) Model of Gray Sequence
The GM (1, 1) model and the GM (2, 1) model are both single-sequence linear
dynamic models, while the GM (1, n) model is a multivariate (multivariate) first-
order linear dynamic model. It is mainly used for system dynamic analysis (Deng
1990; Chen 2003; Liu et al. 2014).
For the original data with n sequences and a sequence length of m, we can use the
following data matrix to describe:
ð 0Þ ð 0Þ ð 0Þ
x1 ð1Þ x2 ð1Þ ⋯ xn ð1Þ
ð0Þ x1 ð2Þ x2 ð2Þ ⋯ xn ð2Þ
XN = ⋯ ⋯ ⋯ ⋯
x1 ðmÞ x2 ðmÞ ⋯ xn ðmÞ
Step 1: Calculate the data matrix generated by one accumulation

2 2 2
ð0Þ ð0Þ ð0Þ
x1 ð i Þ x2 ð i Þ ⋯ xN ðiÞ
i=1 i=1 i=1
3 3 3
ð0Þ ð0Þ ð0Þ
ð1Þ
XN = x1 ðiÞÞ x2 ð i Þ ⋯ xN ðiÞ
i=1 i=1 i=1
⋯ ⋯ ⋯ ⋯
M M M
ð0Þ ð0Þ ð0Þ
x1 ð i Þ x2 ð i Þ ⋯ xN ðiÞ
i=1 i=1 i=1
Step 2: Construct the matrix B, Y and the adjacent mean value to generate the
sequence Zi
1 ð1Þ ð1Þ
- x ð2Þ þ x1 ð1Þ ð 1Þ
x2 ð2Þ ⋯
ð1Þ
xn ð2Þ
2 1
1 ð1Þ ð1Þ
B= - x ð3Þ þ x1 ð2Þ ð 1Þ
x2 ð3Þ ⋯
ð1Þ
xn ð3Þ
2 1 ⋯ ⋯ ⋯ ⋯
1 ð1Þ ð1Þ
- x ðmÞ þ x1 ðm - 1Þ ð 1Þ
x2 ðmÞ ⋯
ð 1Þ
xn ðmÞ
2 1
T
Y = x1 ð0Þ ð2Þ, x1 ð0Þ ð3Þ . . ., x1 ð0Þ ðmÞ
1 ð1Þ ð1Þ
-
x ð 2Þ þ x i ð 1Þ
2 i
1 ð1Þ ð1Þ
Z= - x ð 3Þ þ x i ð 2Þ
2 i
⋯
1 ð1Þ ð1Þ
- x ð m Þ þ x i ð m - 1Þ
2 i
Step 3: Solve the gray parameters using the least squares method a
a
-1 T
a= b1
⋮
= BT B B Y
bn - 1
Step 4: Establish the gray differential equation GM (1, n)
n
ð1Þ ð1Þ
x01 ðt Þ þ az1 ðt Þ = bi x t ð t Þ
i=2
ð1Þ
where -a is the system development coefficient; bi xi ðt Þ is the driving term; bi is the
driving coefficient
Step 5: Substitute the gray parameters into the time function
114 X. Chen
n n
bi - 1 ð1Þ bi - 1 ð1Þ
xð1Þ ðt þ 1Þ = xð0Þ ð1Þ - xi ðt þ 1Þ e - at þ x i ð t þ 1Þ
i=2
a i=2
a
Step 6: Substitute the gray parameters into the time function to obtain the
calculated values of the generated data series xð1Þ ðt Þ. Then, taking the derivative of
xð1Þ , we obtain xð0Þ ðt Þ. The difference ε(0)(t) between x(0)(t) and xð0Þ ðt Þ, and the
relative error e(t) are calculated.
Step 7: While establishing the model, the system also performs Laplace transfor-
mation on the parameters of the GM (1, n) model and gives the dynamic link transfer
function wi (s) of the ith influencing factor to its action object under the zero initial
condition.
ð1Þ
x1 ð s Þ bi =a
wi ð s Þ = = ði= 2, 3, . . . , nÞ
ð1Þ
xi ð s Þ 1 þ 1=as
References
(In Chinese)
Deng JL (1990) A course in gray system theory. Central China Science and Technology University
Press
(In Chinese)
Chapter 6
Gray Prediction
Xinjun Chen, Minyang Xie, and Jintao Wang
Abstract Prediction is a kind of activity that makes use of the knowledge and
means that people have already mastered to predict and judge the future develop-
ment of things. Specifically, people use various qualitative and quantitative analysis
methods according to the objective process and certain laws of the development and
change of things in the past, and according to the state of movement and change of
things, a scientific projection of the possible future trends and possible levels of
things. As a kind of human cognitive activity, prediction has existed in human social
practice for a long time and has been developing with the development of produc-
tivity and Relations of production. Therefore, the forecast is actually by means of the
past to predict and understand the future trend of development. Usually, prediction
can be divided into qualitative prediction and quantitative prediction, quantitative
prediction is through the analysis of data for the prediction, often need to establish a
prediction model. Gray prediction is to discover and grasp the law of system
development by processing the original data and establishing the gray model and
to make a scientific quantitative prediction of the future state of the system. The gray
prediction model does not use the original data sequence, but the generated data
sequence. Its core system is gray model, that is, the method of getting approximate
exponential law from the original data by accumulative generation (or other
processing generation) and then modeling. The advantage of gray prediction is
that it can solve the problems of less historical data, integrity of sequence and low
reliability without enough sample space of data, it can generate irregular original
data to get regular generating sequence. The disadvantage is that it applies only to
short-and medium-term forecasts and only to those approximating exponential
growth. There should be sufficient quantitative analysis as to what model should
be chosen for a particular problem. The choice of models, however, is not inflexible.
A model must go through a variety of tests to determine whether it is reasonable,
only through the test of the model can be used as a prediction. In this chapter, we will
mainly introduce the test method of the gray prediction model, the sequence
X. Chen (✉) · M. Xie · J. Wang

e-mail: [email protected]; [email protected]
116 X. Chen et al.
prediction, the gray catastrophe prediction, and the application example of the gray
prediction in fishery science.
Keywords Gray prediction · Gray catastrophe prediction · Fishery science · Gray

prediction test
6.1 Test Method of the Gray Prediction Model

6.1.1 Absolute Correlation Test Method
The original sequence is now available
X ð0Þ = xð0Þ ð1Þ, xð0Þ ð2Þ, . . . , xð0Þ ðnÞ
The corresponding relative error sequence is
ε ð 1Þ ε ð 2Þ ε ð nÞ
Δ= , , . . . , ð0Þ = fΔk gn1
xð0Þ ð1Þ xð0Þ ð2Þ x ð nÞ
ð0Þ
In the above formula, X(0) is the original sequence, X is the corresponding
simulation sequence, and ε is the absolute correlation of the corresponding simula-
ð0Þ
tion sequence X(0) with X . If for a given ε0 > 0, there is ε > ε0, then the model is
said to be a qualified model.
6.1.2 Mean Square Error Ratio and Small Error Probability

Test Method
ð0Þ
Let X(0) is the original sequence, X is the corresponding simulation sequence, and
ε(0) is the residual sequence, the mean and variance of X(0) are
n
1
x= xð0Þ ðkÞ
n k=1
n 2
1
S21 = xð 0 Þ ð k Þ - x
n k=1
The mean and variance of the residuals are

6 Gray Prediction 117
Table 6.1 Accuracy test grade reference table (Liu et al. 1999)
Relative Gray correlation Mean square error Probability of small
Grade error α degree ε0 ratio C0 error p0
Level 0.01 0.90 0.35 0.95
1
Level 0.05 0.80 0.5 0.80
2
Level 0.10 0.70 0.65 0.70
3
Level 0.20 0.60 0.80 0.60
4
n
1
ε= εð k Þ
n k=1
n
1
S22 = ðεðkÞ - εÞ2
n k=1
1. If in which C = SS21 is called the variance ratio. For a given C0 > 0, when C < C0,
the model is called a qualified model with a mean square error ratio.
2. If p = PðjεðkÞ - εj < 0:6745S1 Þ is called the probability of a small error. For a
given p0 > 0, when p > p0, the model is called a small error probability qualified
model.
Through the above analysis, three methods for testing the model are given. These
three methods are all judged by the accuracy of the model by examining the
residuals, in which the smaller the average relative error Δ is, the better, and the
larger the gray correlation degree ε is. Meanwhile, the smaller the mean square error
ratio C is, the better, and the larger the probability of a small error p is, the better.
Given a set of values of α, ε0, C0, and p0, a level of simulation accuracy of the test
model is determined. The commonly used accuracy grades are shown in Table 6.1.
Under normal circumstances, the most commonly used index is the relative error
test.
6.2 Sequence Prediction
Sequence prediction predicts the future behavior of system variables. The commonly
used sequence prediction model is the GM (1, 1) model. According to the actual
situation, other gray models can also be considered. On the basis of qualitative
analysis, an appropriate sequence operator is defined, and then a GM (1, 1) model is
established. After passing the accuracy test, it can be used for prediction. The entire
modeling method can be referred to as the GM (1, 1) model in Chap. 5.
118 X. Chen et al.
6.3 Prediction of Gray Catastrophe
6.3.1 Gray Catastrophe Prediction
Gray catastrophe prediction is essentially the prediction of outliers. What kind of

value is considered an outlier is often determined by the supervisor’s experience and
historical values. The task of gray catastrophe is to give the time when the next or
several outliers appear so that people can prepare in advance and take preventive
countermeasures.
Now let X be the original sequence
X ξ = ðx½qð1Þ, x½qð2Þ, . . . , x½qðmÞÞ
is a catastrophic sequence, then it is called a catastrophic sequence.
Qð0Þ = ðqð1Þ, qð2Þ, . . . , qðmÞÞ
is the catastrophe date sequence.

Catastrophe prediction searches for the regularity of the catastrophe date series
and predicts the dates of several future catastrophes. The catastrophe prediction of
the gray system is achieved by establishing a GM (1, 1) model for the catastrophe
date series.
Let Q(0) = (q(1), q(2), . . ., q(m)) be the catastrophe date sequence, and its
one-time cumulative sequence is
The sequence of generating the mean value of Q(1) is called Z(1); then, q(k) + az(1)
(k) = b is called the catastrophic GM (1, 1) model.
Now let X = (x(1), x(2), . . ., x(n)) be the original sequence, and let n be the date.
Given a certain outlier ξ, the corresponding catastrophe date series
Qð0Þ = ðqð1Þ, qð2Þ, . . . , qðmÞÞ,
where q(m) (≤n) is the date of the most recent catastrophe; then, qðm þ 1Þ is the
prediction date of the next catastrophe. For any k > 0, we call it qðm þ kÞ the
predicted date of the kth catastrophe in the future.
6.3.2 Prediction of Gray Seasonal Catastrophe
Let Ω = [a, b] be the total time zone. If ωi = [ai, bi] ⊂ [a, b] (i = 1, 2, . . ., s) satisfies
s
Ω = [ ωi; Ωi \ ωj = ∅, and any j ≠ i, then ω (i = 1, 2, . . ., s) is called the season in
i=1
Ω, which is also called the time period or time-sharing zone.
Let ωi ⊂ Ω be a season, and let the original series
X = ðxð1Þ, xð2Þ, . . . , xðnÞÞ
A given outlier ξ is called the corresponding catastrophe sequence.
X ξ = ðx½qð1Þ, x½qð2Þ, . . . , x½qðmÞÞ
Correspondingly, we call
as the seasonal catastrophe date sequence.

Seasonal catastrophe prediction can be carried out according to the following
steps:
Step 1: Given the original sequence X = (x(1), x(2), . . ., x(n));
Step 2: Study the variation range of the original series data and determine the
seasonal ωi = [ai, bi];
Step 3: Let y(k) = x(k)-ai, and convert the original sequence to Y = (y(1), y(2), . . .,
y(n)) to improve the data resolution;
Step 4: Given the outlier ξ, find the seasonal catastrophe sequence
Y ξ = ðy½qð1Þ, qð2Þ, . . . , qðmÞÞ
and seasonal catastrophe date series
Qð0Þ = ðqð1Þ, qð2Þ, . . . , qðmÞÞ;
Step 5: Establish a catastrophe GM (1, 1) model, q(k) + az(1) (k) = b;

Step 6: Test the simulation accuracy and make predictions.
120 X. Chen et al.
6.4 Application of Gray Prediction in Fishery Science
Gray series prediction has been widely used in fishery science. At present, the
application of gray prediction in fisheries is mainly in the following aspects: fishery
yield, fishery population, fisheries forecasting (including resource abundance and
catch), fishery disease, etc.
6.4.1 Gray Prediction of Fishery Yield
6.4.1.1 Gray Prediction of Marine Catch in the Indian Ocean
Lu et al. (2022) published “Construction of a prediction model for marine catches in

the Indian Ocean based on the gray system theory model.” Based on the catch data of
the Indian Ocean from 2000 to 2016, the gray system theory method was used to
analyze the main catch categories that affected the total catch, and a variety of GM
models were established and compared. The data from 2018 were verified, and the
obtained optimal GM model was used to predict the total catch in the Indian Ocean
from 2019 to 2025. The results showed that the main categories that affected the total
catch in the Indian Ocean were demersal fish, crustaceans, pelagic fish, other marine
fishes, and cephalopods. The models were GM (1, 5) and GM (1, 6), with average
relative errors of 1.83% and 1.90%, respectively, and the gray correlation degree was
above 0.9. The average relative errors of the forecasts in 2017 and 2018 were 3.78%
and 3.42%, respectively. The predicted total catches in the Indian Ocean in
2019–2020 and 2021–2025 are 11.86–12.9 million tons and 12.27–13.24 million
tons, respectively. The main increase in the catch may come from pelagic fish and
cephalopods and demersal fish. The study suggests that the growth rate of the total
catch in the Indian Ocean during the Fourteenth Five-Year Plan period is limited,
and the total increase is within 800,000 t, which is basically at the stage of full
development. It is recommended that the scale of fishery development be strictly
controlled in the future to ensure the sustainable development and sustainable use of
fishery resources.
The catch data of marine fishing in the Indian Ocean are from the Food and
Agriculture Organization (FAO), whose website is http://www.fao.org/fishery/
statistics/global-capture-production/query/en, and the time range is 2000–2018. The
catch data were downloaded according to the FAOSTAT format, which mainly
included aquatic plants, crustaceans, cephalopods, mollusks (except cephalopods),
pelagic fishes, demersal fishes, migratory fishes, and other marine fishes. In addition,
other marine animals. The data range covers the entire Indian Ocean, including
coastal fishing countries and high seas fishing countries.
Let the total catch in the Indian Ocean be the mother sequence X1, that is, X1 = {x1
(1), x0(2), . . ., x0(n)}, and let the catches cephalopods, crustaceans, demersal fishes,
migratory fishes, and pelagic fishes and the other major species be the subsequences
Table 6.2 The gray correlation coefficients of each category of catch quantum series and the total
catch mother series (Lu et al. 2022)
Catch category Gray relational degree
Demersal marine fish (X2) 0.91
Crustaceans (X3) 0.89
Pelagic marine fish (X4) 0.88
Marine fish NEI (X5) 0.74
Cephalopods (X6) 0.70
Mollusks (except cephalopods) mollusks excl. Cephalopods (X7) 0.56
Freshwater and diadromous fish (X8) 0.54
Aquatic animals NEI (X9) 0.52
Aquatic plants (X10) 0.49
Xi, i.e., Xi = {xi (1), xi(2),. . ., xi(n)}, i = 1, 2,. . ., m. Gray correlation analysis is
performed to obtain the main categories that affect the total catch. The averaging
method was used for the initial value, and the resolution coefficient was set to 0.5.
The GM (1, N ) model was used to predict the total catch in the Indian Ocean. The
top 5 subsequences with the largest gray correlation degree are selected, and the five
GM (1, N ) prediction models are established according to the degree of gray
correlation:
GM (1, 2) model: including the total catch and the maximum correlation value of
the catch of the corresponding major species;
GM (1, 3) model: including the total catch and the top 2 correlations. The catch of
the corresponding major species;
GM (1, 4) model: including the total catch and the top 3 correlation values. The
catch of the corresponding major species;
GM (1, 5) model: including the total catch and the top 4 correlation values of the
GM (1, 6) model: including the total catch and the top 5 correlation values of the
Five GM (1, N) prediction models were established using the total catches from
2000 to 2016 and the catches of the first five categories, and the average relative error
and the gray correlation between the predicted value and the actual value (similarly,
averaging was performed, and the resolution coefficient was set to 0.5), and the gray
correlation degree with the smallest and largest relative errors was used as the
optimal model. The data from 2017 to 2018 were validated. At the same time, the
GM (1, 1) model was used to predict the catch of each category in 2019–2025, and
then the optimal GM (1, N ) model was used to predict the total catch in the Indian
Ocean from 2019 to 2025.
The gray correlation analysis shows (Table 6.2) that the greatest gray correlation
between the total catch (X1) and the catch of each major species in the Indian Ocean
from 2000 to 2016 is the demersal fish (X2), and its value is the smallest for the
aquatic plants (X10), which is 0.52. The top five with the highest degree of gray
122 X. Chen et al.
Table 6.3 Relevant parameters of the gray GM (1, N ) model (Lu et al. 2022)
Gray correlation
coefficient between
Gray the predicted value
prediction and the original
model Response function value
GM (1, 2) X1(t + 1) = (9087339.511 - 4.973X2) e(-1.604t) + 4.973X2 0.79
GM (1, 3) X1(t + 1) = (9087339.511 - 13.874X2 + 23.564X3) e(- 0.70
0.984t)
+ 13.874X2 - 23.564X3
GM (1, 4) X1(t + 1) = (9087339.511 - 0.65
15.644X2 + 17.632X3 + 2.109X4) e(-0.786t) + 15.644X2 -
17.632X3 - 2.109X4
GM (1, 5) X1(t + 1) = (9087339.511 - 1.268X2 - 0.362X3 - 0.92
1.256X4 - 0.882X5) e(-
2.061t)
+ 1.268X2 + 0.362X3 + 1.256X4 + 0.882X5
GM (1, 6) X1(t + 1) = (9087339.511 - 1.057X2 - 1.345X3 - 0.92
1.083X4 - 0.915X5 - 1.071X6) e(-
1.992t)
+ 1.057X2 + 1.345X3 + 1.084X4 + 0.915X5 + 1.071X6
correlation are demersal fishes (X2), crustaceans (X3), pelagic fishes (X4), other
marine fishes (X5), and cephalopods (X6).
Statistical analysis showed that the total marine catch in the Indian Ocean from
2000 to 2018 showed a steady growth trend and reached the highest historical
production in 2017, which was 12.44 million tons. Among them, pelagic fishes,
other marine fishes, demersal fishes, crustaceans, and cephalopods were the main
species, and the average catches from 2017 to 2018 were 5,201,000, 2,554,000,
2,541,000, 930,000, and 466,000 tons, respectively. They accounted for 41.98%,
20.61%, 20.51%, 7.51%, and 3.76% of the total catch, respectively.
According to the order of gray correlation value, demersal fish X2, crustaceans X3,
pelagic fish X4, other marine fish X5, and cephalopods X6 were selected as the factors
affecting the total catch X1. Five GM (1, N ) models were established in Table 6.3.
The relative errors of the GM (1, 5) and GM (1, 6) models are 1.83% and 1.90%,
respectively (Table 6.4). The gray correlation degree between the predicted value
and the original value series is 0.92, so the GM (1, 5) and GM (1, 6) models are the
optimal prediction models.
According to the optimal models GM (1, 5) and GM (1, 6), the data of 2017 and
2018 were verified, and the average relative errors were 3.78 and 3.43, respectively
(Table 6.5), indicating that the accuracy of the model was relatively good.
Using the catch data of demersal fish X2, crustaceans X3, pelagic fish X4, other
marine fish X5, and cephalopods X6 from 2000 to 2016, the GM (1, 1) model was
established (Table 6.6). The 2017–2018 data were used for the test. The catch
prediction models for each category basically met the accuracy test requirements.
The catches of demersal fishes X2, crustaceans X3, pelagic fishes X4, other marine
fishes X5, and cephalopods X6 in 2019–2025 are shown in Tables 6.7. According to
the GM (1, 5) and GM (1, 6) models, the total catch in the Indian Ocean from 2019 to
2025 can be calculated (Tables 6.8). The total catch will increase between 12.27 and
Table 6.4 Relative error of each GM (1, N ) model (Lu et al. 2022)
Year GM (1, 2) GM (1, 3) GM (1, 4) GM (1, 5) GM (1, 6)
2001 7.29 14.92 13.67 13.77 13.46
2002 17.44 38.07 41.20 8.49 9.16
2003 2.87 18.36 24.80 2.11 2.79
2004 3.54 5.32 10.99 0.47 0.90
2005 7.17 0.97 0.22 0.00 0.04
2006 7.99 4.95 6.87 0.63 0.80
2007 1.84 0.76 4.28 1.20 1.18
2008 3.03 0.09 5.40 0.08 0.19
2009 2.80 5.36 8.59 0.07 0.05
2010 1.25 17.55 15.76 0.32 0.06
2011 0.80 14.38 9.86 1.05 0.43
2012 7.01 1.04 6.68 0.18 0.44
2013 8.52 8.31 11.80 0.01 0.32
2014 0.75 11.12 21.16 0.36 0.30
2015 3.31 4.90 1.38 0.45 0.16
2016 11.00 28.66 26.33 0.08 0.13
Mean value 5.41 10.92 13.06 1.83 1.90
Table 6.5 Comparison of the predicted and actual values of GM (1, 5) and GM (1, 6) from 2017 to
2018 (Lu et al. 2022)
2017 2018
Actual Predicted Relative Actual Predicted Relative
Model value value error value value error
GM 12,446,838 11,956,591 3.94 12,332,944 11,901,884 3.50
(1, 5)
GM 12,446,838 11,996,265 3.62 12,332,944 11,919,801 3.35
(1, 6)
Table 6.6 Prediction models of GM (1, 1) catches (Lu et al. 2022)

Small error Mean square error
Catch category Response function probability P ratio C
Demersal fish X2(t + 1) = 9246.22 e(0.019t) - 0.944 0.387
X2 9046.86
Crustaceans X3 X3(t + 1) = 3174.69 e(0.021t) - 0.944 0.385
3101.67
Pelagic fish X4 X4(t + 1) = 11018.39 e(0.028t) - 1.000 0.234
10702.66
Other marine X5(t + 1) = -18194.06 e(- 0.650 0.640
fish X5 0.017t)
+ 18447.27
Cephalopod X6 X6(t + 1) = 268.28 e(0.058t) - 0.944 0.487
243.56
124 X. Chen et al.
Table 6.7 Prediction results of catches of each category from 2019 to 2015 (unit: 104 t) (Lu et al.
2022)
Year Demersal fish Crustaceans Pelagic fish Other marine fish Cephalopods
2019 257.70 96.88 529.63 226.77 44.83
2020 262.76 98.91 544.89 222.92 47.49
2021 267.92 100.99 560.60 219.14 50.30
2022 273.18 103.11 576.76 215.42 53.28
2023 278.55 105.27 593.38 211.77 56.44
2024 284.02 107.48 610.48 208.18 59.79
2025 289.60 109.74 628.08 204.65 63.33
13.24 million tons in 2021–2025. The main increase in the catch may come from
pelagic fish, cephalopods, bottom fish, etc.
6.4.2 Gray Prediction of Shrimp Production in the Bohai Sea
Guo (1992) published “Gray Prediction of Shrimp Yield in the Bohai Sea.” In this
study, the author used the general multiple linear regression method and the gray
system prediction models GM (0, h) and GM (1, h) to model shrimp production in
the Bohai Sea and compared their accuracy.
In this study, X1 represents the relative yield of Penaeus chinensis in the Bohai
Sea, and X2, X3, and X4 represent the relative numbers of juvenile shrimp in Bohai
Bay, Laizhou Bay, and Liaodong Bay, respectively. The relative yields of prawns
and the relative numbers of juveniles are listed in Table 6.9.
We call the data in Tables 6.9 the original sequence, which is denoted as
ð0Þ
X k ðiÞ , k = 1, 2, 3, 4; i = 1, 2, . . . 15
The general multivariate regression equation established using the data in
Table 6.9 is
ð0Þ ð0Þ ð0Þ ð0Þ

X 1 ðiÞ = 0:933X 2 ðiÞ þ 0:459X 3 ðiÞ þ 0:366X 4 ðiÞ - 1:694
Second, the gray static multivariate model GM (0, 4) is established:
ð1Þ ð1Þ ð1Þ ð1Þ

X 1 ðiÞ = 0:987X 2 ðiÞ þ 0:493X 3 ðiÞ þ 0:297X 4 ðiÞ - 32:509
The gray dynamic multivariate model GM (1, 4) is
ð1Þ
dX 1 ð1Þ ð1Þ ð1Þ ð1Þ
þ 1:731X 1 = 1:778X 2 þ 0:763X 3 þ 0:568X 4
dt
The corresponding time-corresponding equation is

6
Gray Prediction
Table 6.8 Forecast results of the total catch in the Indian Ocean from 2019 to 2025 Unit: t (Lu et al. 2022)
Year 2019 2020 2021 2022 2023 2024 2025
GM (1, 5) 11,865,349 12,065,390 12,272,166 12,485,569 12,705,897 12,933,212 13,167,845
GM (1, 6) 11,882,322 12,089,118 12,303,560 12,525,639 12,755,711 12,994,002 13,240,790
125
126 X. Chen et al.
Table 6.9 Relevant data of Penaeus chinensis in the Bohai Sea (Guo 1992)
Relative number of juvenile
shrimp Forecast value
Bohai Laizhou Liaodong Relative yield of GM GM
Year Bay Bay Bay prawns Regression (0, h) (1, h)
1969 119 48 1 100.0 133.5 108.9 100.0
1972 20 157 16 100.5 100.6 101.9 114
1973 66 243 100 236.8 216.9 214.6 221.4
1974 13.9 165 251 313.4 301.7 293.0 305
1975 100 314 114 254.0 288.9 287.3 281.1
1976 64 37 39 87.4 90.6 93.0 95.6
1977 158 123 44 212.7 222.3 229.6 231.1
1978 163 223 42 320.0 276.3 283.3 279.5
1979 305 191 133 404.9 426.3 434.6 441.1
1980 176 276 2 313.2 300.0 310.3 303.1
1981 119 117 46 205.6 184.2 188.8 188.9
1982 20 61 23 58.2 55.6 55.6 55.0
1983 91 72 25 147.1 128.1 132.7 133.4
1984 48 21 27 53.5 63.4 65.7 67.4
1985 24 323 33 174.3 192.8 192.7 177.8
ð1Þ ðÞ ð1Þ ð1Þ

X 1 ðiÞ = 100 - 1:027X 2 ðiÞ - 0:441X 3 ðiÞ - 0:325X 4 ðiÞ e - 1:72tði - 1Þ
ð1Þ ð1Þ ð1Þ
þ1:027X 2 ðiÞ þ 0:441X 3 ðiÞ þ 0:328X 4 ðiÞ
The above three models were used to predict the relative yields of prawns in
Table 6.8, and the results are listed in Table 6.8. The absolute and relative deviations
of the predicted values of the three models relative to the original series values are
listed in Tables 6.10.
The analysis shows that, under the condition that the relative number of juvenile
shrimp in each bay is known, the multivariate gray model established by the gray
system method has higher prediction accuracy and higher reliability than the tradi-
tional regression model. It can also be seen from the prediction results that GM (0, h)
and GM (1, h) have similar prediction accuracies and can be used to predict shrimp
in each bay.
6.4.3 Gray Prediction of Fishery Human Resources
Chen and Zhou (2001) published “Analysis and Prediction of the Human Resources
Structure of China’s Marine Fisheries.” Based on the statistical data of China’s
marine fishery human resources between 1990 and 1998, the study analyzed China’s
1990–1998 period using gray correlation and gray prediction methods. The
Table 6.10 Error analysis of the three models (Guo 1992)

Regression model GM (0, h) GM (1, h)
Absolute Relative Absolute Relative Absolute Relative
Year error error error error error error
1969 33.5 33.5 8.9 8.9 0 0
1972 0.1 0.1 1.4 1.4 13.6 13.6
1973 19.9 8.4 22.2 9.4 13.4 6.5
1974 11.7 3.7 20.4 6.5 8.3 2.7
1975 34.9 13.7 33.3 13.1 27.1 10.7
1976 3.2 3.7 5.6 6.4 8.2 9.3
1977 10.1 4.7 16.9 8.0 18.4 8.6
1978 43.7 13.7 36.7 11.5 40.5 12.7
1979 21.4 5.3 29.7 7.3 36.2 8.9
1980 13.2 4.2 2.9 0.9 10.1 3.2
1981 21.4 10.4 16.8 8.2 16.7 8.1
1982 2.6 4.5 1.6 2.7 3.2 5.6
1983 19.0 12.9 14.4 9.8 13.7 9.3
1984 9.9 18.5 12.2 22.9 13.9 26.0
1985 18.5 10.6 18.3 10.5 3.4 2.0
Table 6.11 Statistics of China’s marine fishery labor from 1990 to 1998. Unit: person (Chen and
Zhou 2001)
Fishery labor Fishing labor Farming labor Service labor Part-time labor
Year X0 X1 X2 X3 X4
1990 2,080,537 960,800 257,119 173,758 688,860
1991 2,167,621 971,668 276,880 216,354 702,719
1992 2,240,263 1,023,730 296,592 204,911 715,030
1993 2,329,479 1,046,095 355,943 211,100 716,341
1994 2,386,469 1,052,384 365,933 239,101 729,051
1995 2,514,682 1,099,454 398,715 244,888 771,625
1996 2,526,353 1,167,362 319,486 232,816 806,689
1997 2,681,563 1,193,838 459,177 269,779 758,769
1998 2,711,360 1,185,079 488,706 268,956 768,619
composition of the marine fishery labor and its changes and the gray forecast of the
development trend of China’s marine fishery labor from 2000 to 2005. The original
data are shown in Table 6.11.
The raw data in Tables 6.10 were subjected to initial transformation to calculate
the relative gray correlation between fishery labor and fishing labor, farming labor,
service labor, and part-time labor, and the values are γ 01 = 0.9029, γ 02 = 0.6477,
γ 03 = 0.6719, and γ 04 = 0.8057, respectively.
At the same time, each of the original sequences in Tables 6.10 was treated with
the zero value of the starting point, and the absolute gray correlations ε between the
128 X. Chen et al.
fishery labor and fishing labor, farming labor, service labor, and part-time labor were
calculated as follows: ε01 = 0.68110, ε02 = 0.65178, ε03 = 0.58665, ε04 = 0.58046.
If θ = 0.5, then the gray comprehensive degree is R = 0.5γ + 0.5ε; then, we can
obtain:
R01 = 0:7920, R02 = 0:6497, R03 = 0:6293, R04 = 0:6931
6.4.3.1 Gray Prediction of Fishery Labor
The accumulative number is generated once for the original data sequence X0, and
the accumulative sequence is obtained as:
X 0 ð1Þ = ð2080537, 4248158, 6488421, 8817900, 11204369, 13719051, 16245404,

18926967, 21638327Þ:
According to the cumulative sequence X0(1), the adjacent mean sequence is

generated to obtain the adjacent mean sequence.
Z 0 ð1Þ = ð3164347:5, 5368290, 7653160:5, 10011135, 12461710, 14982228,

17586186, 20282647Þ
Construct matrices B and Y, respectively, and obtain the parameter sequence
- 0:03274
a = ða, bÞT = BT B - 1 BT Y =
2070175
Then, the GM (1, 1) model of marine fishery labor is
dxð1Þ
- 0:03274xð1Þ = 2070175
dt
The corresponding time response is
b - ak b
xð1Þ ðk þ 1Þ = xð0Þ ð1Þ - e þ = 65303597 e - 0:03274k - 63223060
a a
The simulation value was calculated using the time response equation
Table 6.12 The error test table (Chen and Zhou 2001)
Serial number Actual data Simulation data Residual Relative error %
2 2,167,621 2,173,693 -6072 0.28
3 2,240,263 2,246,047 -5784 0.26
4 2,329,479 2,320,809 8670 0.37
5 2,386,469 2,398,059 -11,590 0.49
6 2,514,682 2,477,881 36,801 1.46
7 2,526,353 2,560,359 -34,006 1.35
8 2,681,563 2,645,583 35,980 1.34
9 2,711,360 2,733,644 -22,284 0.82
xð1Þ = ð2080537, 4254230, 6500277, 8821086, 11219145, 13697025, 16257385,

18902968, 21636612Þ
Solving the simulation value of X(0) xð0Þ , we have
xð0Þ = ð2080537, 2173693, 2246047, 2320809, 2398059, 2477881, 2560359,

2645583, 2733644Þ
The original actual data and the simulated data were compared, and the residuals
and relative errors were obtained. The results are shown in Table 6.12.
9
Average relative error Δ = 1
8 Δk = 0:80% < 0:01.
k=2
The analysis of the average relative error indicates that the accuracy of the model
reaches the first level, which meets the requirements of prediction.
The development trend of marine fishery labor from 2000 to 2005 was predicted,
and the simulation value xð1Þ was obtained:
xð1Þ = xð1Þ ð2000Þ, xð1Þ ð2001Þ, . . . , xð1Þ ð2004Þ, xð1Þ ð2005Þ
= xð1Þ ð11Þ, xð1Þ ð12Þ, . . . , xð1Þ ð15Þ, xð1Þ ð16Þ
= ð27379905, 30395712, . . . , 40058915, 43496755Þ
The predicted value xð0Þ of X(0) from 2000 to 2005 is obtained by reduction.
xð0Þ = xð0Þ ð2000Þ, xð0Þ ð2001Þ, . . . , xð0Þ ð2005Þ
= ð2918657, 3015807, 3116191, 3219917, 3327095, 3437840Þ

130 X. Chen et al.
6.4.3.2 Gray Prediction of Fishing Labor
Using the sequence of marine fishing labor force X1, the GM (1, 1) model is obtained
through the above calculation:
dxð1Þ
- 0:02994xð1Þ = 938014
dt
b - ak b
xð1Þ ðk þ 1Þ = xð0Þ ð1Þ - e þ = 32291627e - 0:02994k - 31330827
a a
The simulation data were obtained, and the error test was performed. The test
results are shown in Table 6.13.
9
Then, the average relative error is Δ = 1
8 Δk = 1:31% < 0:05, and the error
k=2
accuracy reaches the second level, which meets the requirements of prediction.
The development trend of marine fishing labor from 2000 to 2005 was predicted,
and the predicted value xð0Þ of X(0) from 2000 to 2005 was recovered:
= ð1284890, 1323940, 1364176, 1405636, 1448356, 1492374Þ
6.4.3.3 Gray Prediction of the Farming Labor
Using the sequence of the farming labor X2, the GM (1, 1) model is obtained through
the above calculation:
Serial number Actual data Simulation data Residual Relative error%
2 971,668 981,397 -9729 0.10
3 1,023,730 1,011,223 12,507 1.22
4 1,046,095 1,041,956 4139 0.40
5 1,052,384 1,073,623 -21,239 2.02
6 1,099,454 1,106,252 -6798 0.62
7 1,167,362 1,139,873 27,489 2.35
8 1,193,838 1,174,516 19,322 1.62
9 1,185,079 1,210,211 -25,132 2.12
2 276,880 283,232 -6352 2.29
3 296,592 304,477 -7885 2.66
4 355,943 327,315 28,628 8.04
5 365,933 351,867 14,066 3.84
6 398,715 378,260 20,455 5.13
7 319,486 406,632 -87,146 27.28
8 459,177 437,133 22,044 4.80
9 488,706 469,921 18,785 3.84
dxð1Þ
- 0:07233xð1Þ = 254516
dt
b - ak b
a a
The simulated data were obtained and tested for errors. The test results are shown
in Table 6.14.
9
Then, the average relative error Δ = 1
8 Δk = 7:24% < 0:10, and the error
k=2
accuracy reaches the third level, which can basically meet the requirements of
prediction.
The development trend of farming labor from 2000 to 2005 was predicted, and
the predicted value xð0Þ of X(0) from 2000 to 2005 was restored:
= ð543061, 583795, 627585, 674659, 725264, 779665Þ
6.4.3.4 Gray Prediction of Service Labor
Using the sequence of service labor X3, the GM (1, 1) model is obtained through the
above calculation:
dxð1Þ
- 0:03892xð1Þ = 194347
dt

132 X. Chen et al.
2 216,354 205,074 11,280 5.21
3 204,911 213,212 -8301 4.05
4 211,100 221,673 -10,573 5.01
5 239,101 230,470 8631 3.61
6 244,888 239,616 5272 2.15
7 232,816 249,125 -16,309 7.01
8 269,779 259,011 10,768 3.99
9 268,956 269,290 -334 0.12
b - ak b
a a
The simulated data were obtained and tested for errors. The test results are shown
in Table 6.15.
9
8 Δk = 3:89% < 0:05, and the error
k=2
accuracy reaches the second level, which meets the requirements of prediction.
The development trend of service labor from 2000 to 2005 was predicted, and the
predicted value xð0Þ of X(0) from 2000 to 2005 was restored.
= ð291087, 302639, 314649, 327135, 340117, 353614Þ
6.4.3.5 Gray Prediction of Part-Time Labor
The GM (1, 1) model is obtained through the above calculation using the sequence of
the part-time marine labor force X4:
dxð1Þ
- 0:01570xð1Þ = 689422
dt
b - ak b
a a
The model was used to obtain the simulated data, and the error test was
performed. The test results are shown in Table 6.16.
2 702,719 705,759 -3040 0.43
3 715,030 716,924 -1894 0.26
4 716,341 728,267 -11,926 1.66
5 729,051 739,788 -10,737 1.47
6 771,625 751,492 20,133 2.61
7 806,689 763,381 43,308 5.37
8 758,769 775,458 -16,689 2.20
9 768,619 787,727 -19,108 2.49
9
8 Δk = 2:06% < 0:05, and the error
k=2
accuracy reaches the second level, which meets the prediction requirements.
The development trend of part-time labor from 2000 to 2005 was predicted, and
the predicted value xð0Þ of X(0) from 2000 to 2005 was restored:
= ð812848, 825708, 838771, 852041, 865521, 879214Þ
The studies have shown that the order of gray comprehensive correlation from
high to low is R01 > R04 > R02 > R03. The factors that have the greatest impact on
marine fishery labor are fishing labor, followed by part-time labor, farming labor,
and farming labor. The study suggests that labor marine fishing is the largest factor
affecting labor in marine fisheries. This indicates that in the current human resource
structure of marine fisheries in China, the number of people engaged in fishing is
large, while the number of people engaged in farming is small. The overexploitation
of offshore fishery resources will result in excessive fishing capacity. This structure
of human resources is extremely detrimental to the sustainable development of
China’s offshore fishery resources.
Through the establishment of the gray model GM (1, 1), the development trends
of marine fisheries, marine fishing, marine farming, service labor, and part-time
labor in 2000–2005 were predicted. By 2000–2005, the number of laborers engaged
in marine fisheries in China will reach 2.919–3.438 million, the number of laborers
in marine fishing will be 1.285–1.492 million, and the number of laborers in marine
farming will be 543,000–779,000. The number of people working part-time in the
marine industry is between 813,000 and 879,000.
134 X. Chen et al.
6.4.4 Application of the Gray System in Fishery Forecasting
Fishery forecasting is the main content of fishery oceanography. Fishery forecasting

refers to the forecast of various factors of aquatic living resource conditions in a
certain period of time and within a certain range of marine waters in the future, such
as fishing season, fishing grounds, quantity and quality of fish stocks, and possible
catches. The basis of fishery forecasting is the relationship between fish behavior and
biological conditions and environmental conditions and their patterns, as well as
various fishing and sea condition data obtained from various real-time fisheries
surveys, such as catches, resource conditions, and marine environments. The main
task of fishery forecasting is to predict the fishing grounds, fishing season, and
possible catch. In recent years, the gray system has been used for fishery forecasting
and has achieved good results.
6.4.4.1 Gray Prediction of the Peak Fishing Season of Ommastrephes

bartramii
Understanding the fishing season characteristics of Ommastrephes bartramii in the

northern Pacific Ocean and predicting the peak fishing season is one of the important
parts of fishery forecasting, which is conducive to the scientific production of fishery
enterprises, the rationalization of the operation time, the saving of costs, and the
improvement of fishing efficiency. Gray waveform prediction is a prediction method
for periodic fluctuation series in the gray system. Based on the GM (1, 1) model
group, it can perform medium- and long-term predictions for time series with large
fluctuation amplitudes and small amounts of data. The squid, Ommastrephes
bartramii, in the northern Pacific Ocean is an annual species, its fishing season is
the annual cycle, and the annual variation is drastic. It is difficult to use the
traditional statistical analysis method to carry out better prediction of the changes
in its fishing season. To this end, Xie and Chen (2021a) used fishery production
statistical data from 2013 to 2017 to establish a series of fishing season dates and
used the gray waveform prediction model to predict the peak fishing season of
Ommastrephes bartramii.
6.4.4.1.1 Data and Study Methods
The statistical data of fishery production are from China squid jigging vessels. The
time period is from 2013 to 2017, and the spatial range is 35°–45°N and 140°–179°
E. The fishing data include date, longitude, latitude, and daily yield. The spatial
resolution is 1° × 1°. The catch per unit fishing effort (CPUE) was used to charac-
terize the abundance index of Ommastrephes bartramii.
Because the annual yield is affected by the marine climate environment and
fluctuates greatly, the quartile of the CPUEday sequence (Q1 - Q3) is calculated
based on the CPUEday sequence of the current year in each year. The third quantile
Q3 of the sequence (CPUEday value greater than 75%) was defined as the high-yield
CPUEday value. If there is a high-yield CPUEday value for more than 3 consecutive
days, the first day of these 3 days is called the start of the peak fishing season. If there
is no high-yield CPUEday value for more than 3 consecutive days, the first day of
these 3 days is called the end of the peak fishing season.
The gray waveform prediction model was used to predict the fishing season of
Ommastrephes bartramii in the North Pacific Ocean. The study method is as
follows:
1. Data selection: According to the division results of the peak fishing season, the
sequence number of the beginning of the peak fishing season is found from the
fishing season time to form the peak fishing season date series. Suppose the date
sequence of the fishing season X = (x (1), x (2)..., x (n)), then the n-segment
polyline graph of the sequence X is called xn = x (n) + (t - n) [x (n + 1) - x (n)],
i.e., the sequence X = {xn = x (n) + (t - n) [x (n + 1) - x (n)]| n = 1, 2..., m - 1}.
2. Selection of contour lines. Suppose σ max = max fxðnÞg and
1≤n≤m
σ min = min fxðnÞg, we select s + 1 thresholds ξ0, ξ1, . . ., ξs between σ max and
1≤n≤m
σ min, so that it satisfies 8ξ 2 [σ min, σ max], and
i
ξ0 = σ min , . . . , ξi = ðσ max - σ min Þ þ σ min , . . . , ξs = σ max ;
s
Then, X = ξi (i = 1, 2 ..., s) is the s + 1 contour line with equal intervals of the

line graph X.
3. Selection of contour time sequence. The contour line X = ξi intersects with the
line graph of the peak fishing season’s prosperity. If there is a contour on the K
ξ - xðk Þ
broken line of X, the coordinates are k þ xðkþ1 Þ - xðk Þ, ξ . Let qðk Þ =
ξ - xðk Þ
k þ xðkþ1 Þ - xðk Þ (k = 1, 2 ..., m); then, Q i = (q (1), q (2) ..., q (m)) (k = 0, 1 ...,
s) is the time sequence of the contour.
4. Establish the GM (1, 1) model for the time sequence of the contour. For each
contour line, the contour time sequence Q i = (q (1), q (2) ..., q (m)) (k = 0, 1..., s)
is used to establish a GM (1, 1) model to obtain the predicted values. See Chap. 5
for the calculation and testing of the GM (1, 1) model.
6.4.4.1.2 Research Results
Analysis of Fishing Season Characteristics

In 2013–2017, the earliest fishing season of Ommastrephes bartramii in the northern
Pacific Ocean was May 12 (2017), the latest was December 31 (2014–2016), and the
peak fishing season was June–November (Fig. 6.1). The range of total fishing
operation days was 194–224 days, with an average of 213 days; the daily unit
136 X. Chen et al.
Fig. 6.1 Distribution of daily catch per unit of fishing effort (A–E) for Ommastrephes bartramii in
the northern Pacific Ocean during 2013–2017 (Xie and Chen 2021a)
Fig. 6.1 (continued)

138 X. Chen et al.
Fig. 6.1 (continued)
fishing effort (CPUEday) ranged from 0.13 to 8.05 t/day, with an average of 1.82 t/
day. From the perspective of the main fishing season of each year, the average unit
fishing effort of each month has a consistent trend. The average CPUEday in June and
July was the smallest, accounting for 8.87% and 9.15% of the main fishing season,
respectively. The percentages of average CPUE day in other months in the main
fishing season were August (17.81%), October (13.82%), and November (17.4%)
(Fig. 6.2). To facilitate the division and prediction of the next peak fishing season,
the main fishing seasons were arranged from small to large according to the numbers
1 to 183 to form a date sequence, i.e., number 1 (June 1), number 2 (June 2),...,183
(November 31).
Analysis of Peak Fishing Season

According to the annual CPUEday sequence, the high-yield CPUEday values from
2013 to 2017 were 2.07, 1.91, 2.51, 2.86, and 3.20 t/day, respectively. According to
the results of the characteristics of the peak fishing seasons in each year (Table 6.17),
the first fishing season generally occurs in August, and it is basically in the second
10 days of August, and the subsequent peak fishing seasons occur in September,
October, and November. Except for only one peak fishing season in 2016, there were
at least four peak fishing seasons in all other years, and the highest peak fishing
Jun Jul Aug Sep Oct Nov Annual average CPUE

4.5
4.0
3.5
CPUEday (t·vessel-1·d-1)
3.0
2.5
2.0
1.5
1.0
0.5
0.0
2013 2014 2015 2016 2017
Year
Fig. 6.2 Daily catch per unit of fishing effort (A–E) for Ommastrephes bartramii in the northern
Pacific Ocean from June to November 2013–2017 (Xie and Chen 2021a)
season occurred in 2013, with seven peak fishing seasons. The number of days in
each peak fishing season is different. The minimum number of days is 3 (the first
fishing season in 2013 and 2015), the maximum is 51 days (the first fishing season in
2016), and the average number of days is 10 days. The average CPUEday in each
fishing season was above 1.99 t/day, and the highest CPUEday was 8.05 t/day (the
fourth fishing season in 2017). In general (except for 2017), the average CPUEday
tends to increase with increasing time in each peak season of each year.
According to the date series of the peak fishing season from 2013 to 2015,
10 contour lines X (X0 – X9) are divided. Since at least four data sets are required
for GM (1, 1) modeling, GM (1, 1) modeling is performed on only 7 sets of contour
sequences (X2 - X8). From the fitting results of the model (Table 6.18), the average
relative error is within 11.58%, and the fitting effect of the GM (1, 1) model with
contour line X2 is the best, which is 1.74%. From the perspective of the relevant
parameters of the model, the probability of small error P is 1.00 (>0.95); the
variance ratio of the contour line X7 and the contour line X8 model is C < 0.50,
and the variance ratios of other models are all consistent with C < 0.35. The
accuracy of the model is grade I and grade II. From the perspective of the develop-
ment coefficient a of the model, all models can be used for medium- and long-term
prediction (-a ≤ 0.3).
According to the results of the GM (1, 1) prediction model in the peak fishing
season (Table 6.19), the fitting effect was good in the peak fishing season, except for
the relatively large error in the fourth fishing season in 2014 (relative error of
49.01%). The relative error of the fitting during the peak fishing season was within
140 X. Chen et al.
Table 6.17 Characteristics of peak fishing season of Ommastrephes bartramii in the northern
Pacific Ocean (Xie and Chen 2021a)
Peak fishing season (month-day/date Numbers of Average Highest
Year sequence) days CPUEday CPUEday
2013 1st fishing season (22 Aug-24 3 2.57 3.04
Aug/83–85)
2nd fishing season (3 Sep-8 6 2.58 3.13
Sep/95–100)
3rd fishing season (12 Sep-17 6 2.76 4.61
Sep/104–109)
4th fishing season (21 Sep-24 4 2.69 3.57
Sep/113–116)
5th fishing season (4 Oct-12 9 2.6 3.41
Oct/126–134)
6th fishing season (3 Nov-6 4 2.61 3.21
Nov/156–159)
Nov/178–183)
Aug/65–71)
2nd fishing season (17 Aug-20 4 2.14 2.63
Aug/78–81)
Sep/93–100)
4th fishing season (16 Sep-2 17 2.43 3.54
Oct/108–124)
Nov/170–174)
Aug/62–64)
2nd fishing season (12 Aug-20 9 3.11 3.71
Aug/73–81)
Sep/96–105)
Nov/139–145)
Nov/154–168)
Oct/84–134)
2017 1st fishing season (19 Aug-26Aug/ 8 5.23 6.65
80–87)
2nd fishing season (6 Sep-15 10 5.17 6.34
Sep/98–107)
Sep/111–114)
Nov/155–178)
6
Table 6.18 Fitting results and relevant parameters of GM (1, 1) models with different contours (Xie and Chen 2021a)
The fitting contour sequence results by GM (1, 1) Average relative Variance ratio Development Small error
Contours X model error C coefficient a probability P
Gray Prediction
X2 = 87.78 Contour time 1.40 7.80 9.65 12.76 1.74 0.04 -0.25 1.00
sequence
GM (1, 1) model 1.40 7.62 9.78 12.57
X3 = 100.67 Contour time 2.63 7.68 10.51 12.64 2.72 0.08 -0.24 1.00
sequence
GM (1, 1) model 2.63 7.91 10.03 12.72
X4 = 113.56 Contour time 4.04 7.57 11.09 12.52 5.77 0.19 -0.23 1.00
sequence
GM (1, 1) model 4.04 8.09 10.17 12.80
X5 = 126.44 Contour time 5.01 7.46 11.30 12.40 7.21 0.26 -0.23 1.00
sequence
GM (1, 1) model 5.01 8.11 10.18 12.77
X6 = 139.33 Contour time 5.44 7.34 11.51 12.28 8.66 0.33 -0.22 1.00
sequence
GM (1, 1) model 5.44 8.13 10.18 12.74
X7 = 152.22 Contour time 5.87 7.23 11.71 12.16 10.12 0.39 -0.22 1.00
sequence
GM (1, 1) model 5.87 8.15 10.18 12.71
X8 = 165.11 Contour time 6.41 7.11 11.92 12.05 11.58 0.47 -0.22 1.00
sequence
GM (1, 1) model 6.41 8.17 10.18 12.68
141
142 X. Chen et al.
Table 6.19 Relative errors of the GM (1, 1) prediction model during the peak fishing season (Xie
and Chen 2021a)
Peak fishing season Actual value Predicted value Relative error
1st fishing season in 2014 65.00 71.07 9.34
2nd fishing season in 2014 78.00 77.90 0.12
3rd fishing season in 2014 93.00 97.54 4.88
4th fishing season in 2014 108.00 160.93 49.01
5th fishing season in 2014 170.00 170.51 0.30
2nd fishing season in 2015 93.00 98.75 6.18
3rd fishing season in 2015 108.00 135.50 25.46
Fitted value Actual value

183
Time sequence of peak fishing season
157
131
105
79
53
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Number
Fig. 6.3 Comparison of the predicted values and the actual values of Ommastrephes bartramii
based on the main fishing season forecasting model (Xie and Chen 2021a)
9.34%, and the average relative error was 12.73%. The average relative error of the
verification during the peak fishing season in 2015 was 15.82%.
In this study, we used the gray waveform prediction method and established the
GM (1, 1) model to predict the peak fishing season of Ommatrephes bartramii. In
terms of the relationship between the predicted values and the actual values
(Fig. 6.3), the variation trend of the CPUE is basically the same. From the perspec-
tive of the parameters of the prediction model, the model has good accuracy
(Table 6.18) and can be used for medium- and long-term prediction. However, this
study only considered the changes in fishery production data and did not include the
climatic and marine environmental factors that affect the changes in the abundance
of Ommatrephes bartramii.
6.4.4.2 Establishment of a Gray Prediction Model for the Abundance

Index of Ommastrephes bartramii
The prediction of the abundance index is also an important part of fishery forecast-
ing. Constructing a reasonable gray forecasting model is the basis of scientific
forecasting. To this end, Xie and Chen (2021b) first established a GM (1, 1) model
group for the abundance index sequences of Ommastrephes bartramii with different
time-series lengths and selected the CPUE sequence with the smallest relative error
and variance as the mother sequence. Second, the gray correlations between the
mother sequence and Pacific interdecadal oscillation index (PDO), mean sea surface
temperature (SGSST) of the spawning field, mean sea surface temperature of the
fattening field (FGSST), mean chlorophyll concentration of the spawning field
(SGC), and mean chlorophyll concentration of the fattening field (FGC) are used
to evaluate the effect of environmental factors on the abundance index of
Ommastrephes bartramii. Based on the evaluation results, six gray prediction
models with different orders, including GM (0, N ) and GM (1, N ), were established.
The model with the smallest error was selected as the best model for predicting the
abundance index of Ommastrephes bartramii, which can provide a basis for the
scientific production of squid fishing vessels in the North Pacific Ocean.
6.4.4.2.1 Research Data and Methods
The statistical data of fishery production are from the Chinese squid jigging vessels.
The time period is from 1998 to 2016, and the spatial range is 35°–45°N and 140°–
179°E. The statistical contents include date, longitude, latitude, and daily yield. The
spatial resolution is 1° × 1°.
The climate index PDO was obtained from the website of the Joint Institute for
Atmospheric and Oceanic Research (JISAO) at the University of Washington (http://
research.jisao.washington.edu/pdo/PDO.latest). The environmental data, including
sea surface temperature (SST) and chlorophyll concentration (Chl a), were obtained
from the Oceanwatch website of National Oceanic and Atmospheric Administration
(NOAA) (http://oceanwatch.pifsc.noaa.gov/erddap/index.html). The time range was
from January to December of 1998–2016. The data of the spawning field of
Ommastrephes bartramii are from January to May, and the range is 20°–30°N and
130°–170°E; the data of feeding ground are 35°–50°N and 150°–175°E from July to
November. The temporal resolution is monthly, and the spatial resolution is 1° × 1°.
The average SST and the average Chl a of the spawning grounds and feeding
grounds in each month were calculated using the averaging method.
The GM (1, 1) model was established for the CPUE series of different time
lengths, and the average relative error of the model established by the CPUE series of
each year was calculated. The CPUE series with relatively small errors and variances
were selected as the mother series for subsequent modeling.
144 X. Chen et al.
The environmental and climatic factors during the spawning and feeding periods
were analyzed using the gray correlation method. Using the CPUE of the current
year as the mother sequence, the SST (abbreviated as SGSST and FGSST, respec-
tively) and Chl a concentration (abbreviated as SGC and FGC, respectively) in the
spawning ground and the feeding ground and the Pacific interdecadal oscillation
index (PDO) were used as the subsequences. The correlation between the mother
sequence and each subsequence was calculated, and the one with the largest gray
correlation among the monthly indicators was used as a factor in the abundance
index prediction model. The calculation method of gray correlation is shown in
Chap. 3, and the resolution coefficient is set to 0.5.
The abundance index of Ommastrephes bartramii in the northwestern Pacific
Ocean was predicted using the discrete GM (0, N ) and GM (1, N ) models. The
numbers 0 and 1 represent the order of the model, and N = i + 1 (i is the number of
factors). The specific calculation method of the model is shown in Chap. 5. The
following six models were designed:
Model 1: GM (0, 6) model that includes all factors, including SGSST, FGSST, SGC,
FGC, and PDO;
Model 2: GM (0, 5) model without SGSST;
Model 3: GM (0, 5) model without FGSST;
Model 4: GM (0, 5) model without SGC;
Model 5: GM (0, 5) model without FGC;
Model 6: GM (0, 5) model without PDO;
Model 7: GM (1, 6) model that includes all factors, including SGSST, FGSST, SGC,
FGC, and PDO;
Model 8: GM (1, 5) model without SGSST;
Model 9: GM (1, 5) model without FGSST;
Model 10: GM (0, 5) model without SGC;
Model 11: GM (1, 5) model without FGC;
Model 12: GM (1, 5) model without PDO.
The average relative error between the predicted value and the actual value was
calculated by comparing the model-fitted CPUE with the actual CPUE value. The
data of the last year of the sample were used for model validation. The optimal model
was selected based on the fitting accuracy and prediction accuracy of the model.
6.4.4.2.2 Selection of Model Time Series
According to Fig. 6.4, as the time length of the CPUE sequence increases, the
average relative error of the GM (1, 1) model basically exhibits an increasing
trend, and the variance gradually decreases. The average relative error of the GM
(1, 1) model of the 8-year CPUE series is the smallest (6.28%), so the series with the
smallest relative error in the 8-year CPUE series (1998–2005) is selected as the
mother sequence of the model establishing GM (0, N ) and GM. (1, N ) to improve the
accuracy of model prediction by adding environmental factors.
Fig. 6.4 Relative error of

the GM (1, 1) prediction
model of the CPUE
sequence with different time
lengths (Xie and Chen
2021b)
Table 6.20 Gray correlation coefficients between the subsequences of each environmental factor
and the mother sequence of the current year’s CPUE (Xie and Chen 2021b)
Mean sea Average sea
surface surface Average Average
temperature in temperature in chlorophyll chlorophyll Pacific
the spawning the feeding concentration concentration Interdecadal
ground ground in the spawning in the feeding Oscillation
Month (SGSST) (FGSST) ground (SGC) ground (FGC) Index (PDO)
Jan. 0.754 - 0.751 - 0.946
Feb. 0.755 - 0.613 - 0.956
Mar. 0.754 - 0.708 - 0.965
Apr. 0.747 - 0.747 - 0.965
May 0.740 - 0.694 - 0.942
Jun. - - - - 0.909
Jul. - 0.727 - 0.651 0.559
Aug. - 0.767 - 0.694 0.837
Sep. - 0.732 - 0.620 0.964
Oct. - 0.794 - 0.677 0.968
Nov. - 0.786 - 0.643 0.917
Dec. - - - - 0.898
Average 0.750 0.761 0.702 0.657 0.902
value
6.4.4.2.3 Selection of Influencing Factors
According to the results of gray correlation analysis (Table 6.20), the effect of the
Pacific Interdecadal Oscillation Index (PDO) on CPUE is the largest, and its average
degree of gray correlation is much greater than that of the other four environmental
146 X. Chen et al.
factors. According to the average value of the correlation, the importance of each
factor in descending order is PDO, mean sea surface temperature (FGSST) in the
feeding ground, mean sea surface temperature in the spawning ground (SGSST),
mean chlorophyll concentration in the spawning ground (SGC), and mean chloro-
phyll concentration (FGC) in the feeding ground.
The months with the greatest impact on the abundance index were different for
each environmental factor: FGSST and PDO in October, SGSST in February, SGC
in January, and FGC and CPUE in August had the largest gray correlation degree.
Therefore, the above four environmental factors are considered the key factors in
establishing the squid abundance index of the prediction model.
6.4.4.2.4 Construction and Comparison of Gray Models
From the perspective of the average relative error of the model (Table 6.21 and
Table 6.22), the GM (0, N ) prediction model is higher than the GM (1, N ) prediction
model. The average error in descending order is as follows: (1) GM (0, N ) model:
Table 6.21 The relative errors of the GM (0, N ) prediction model for the abundance index of
Ommastrephes bartramii in the North Pacific Ocean (Xie and Chen 2021b)
Year Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
1999 3.89 3.78 3.93 3.74 3.16 8.39
2000 5.39 5.22 5.31 3.37 7.69 5.73
2001 0.38 0.01 0.25 3.24 3.42 8.85
2002 3.90 3.64 5.02 8.76 2.07 0.72
2003 0.95 0.76 2.97 3.21 2.75 8.46
2004 1.66 3.39 0.05 2.29 1.29 2.85
2005 1.37 1.78 1.01 2.49 1.09 0.18
Average relative error 2.51 2.65 2.65 3.87 3.07 5.26
Validation 9.00 7.89 9.23 1.18 18.80 7.02
Table 6.22 Relative errors of the GM (1, N ) prediction model for the abundance index of
Ommastrephes bartramii (Xie and Chen 2021b)
Year Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
1999 19.34 16.14 19.10 10.32 28.23 0.37
2000 4.09 0.83 4.14 2.70 4.11 18.15
2001 5.51 3.81 5.65 3.35 10.71 0.78
2002 11.14 12.91 11.01 14.40 14.13 62.42
2003 9.08 7.10 9.45 8.80 10.07 9.02
2004 0.16 4.46 0.70 0.97 4.15 3.89
2005 0.07 4.81 0.44 4.53 4.16 33.22
Average relative error 7.06 7.15 7.21 6.44 10.79 18.27
Validation 28.39 16.42 28.46 1.20 45.79 138.54
20
GM(0,N)model GM(1,N)model GM(1,1)model Mean value of relative error
18
16
14
Relative error (%)
12
10
0
Model 1Model 2Model 3Model 4Model 5Model 6 Model 1Model 2Model 3Model 4Model 5Model 6 Original
Model
Model types
Fig. 6.5 The average relative error of all model types for different orders of prediction models (Xie
and Chen 2021b)
Table 6.23 Parameter values of the four GM (0, N ) models (Xie and Chen 2021b)
a SGSST FGSST FGC PDO
Parameters 1.71 0.05 0.30 -51.13 -0.19
model 1 > model 3 > model 2 > model 5 > model 4 > model 6; (2) GM (1, N )
model: model 4 > model 1 > Model 2 > Model 3 > Model 5 > Model 6 (Fig. 6.5).
Based on the results of model validation in 2006 (Table 6.23), whether it is the
GM (0, N ) model or the GM (1, N ) model, the prediction accuracy of Model 4 is
much higher than that of the other models, with a relative error of 1.18%. The
relative errors of the other GM (0, N ) models are as follows: model 6 (relative error
7.02%), model 2 (relative error 7.89%), model 1 (relative error 9.00%), and model
3 (relative error of 9.00%). The relative errors of the other GM (1, N ) models are as
follows: model 2 (relative error 16.42%), model 1 (relative error 28.39%), model
3 (relative error 28.46%), model 5 (relative error of 45.79%), and model 6 (relative
error of 138.54%). Because the fitting error of each model in the GM (0, N ) model is
not large and the validation result of model 4 is much smaller than that of the other
models, model 4 without the SGC factor is selected as the best model for predicting
the abundance index of Ommastrephes bartramii (Fig. 6.6).
In this study, based on gray system theory and methods, the environmental and
climatic factors of spawning grounds and feeding grounds were used as indicators to
predict the abundance index of Ommastrephes bartramii. From the results of the
model (Fig. 6.5), the fitting accuracy of almost all models (except for GM (1, N )
model 6) was greater than that of the GM (1, 1) model, and model 4 (GM (0, N )) did
not contain the SGC factor and had the best prediction effect. From the perspective
of the relationship between the fitted value and the actual value (Fig. 6.6), the
148 X. Chen et al.
2.80
Actual value Fitted value
2.60
2.40
CPUE (t·vessel-1·y-1)
2.20
2.00
1.80
1.60
1.40
1.20
1998 1999 2000 2001 2002 2003 2004 2005 2006
Year
Fig. 6.6 Comparison of the predicted values and the true values of the abundance index of
Ommastrephes bartramii based on the GM (0, N ) model in the North Pacific Ocean (Xie and
Chen 2021b)
variation trend of CPUE is basically the same, and the variation amplitude of the
fitted value predicted by the model is small. The value of -1.71 (Table 6.23) satisfies
the conditions of the medium- and long-term forecast model (-a < 0.3), indicating
that the abundance index of Ommastrephes bartramii in the northern Pacific Ocean
is indeed affected by marine climate factors and environmental factors.
In this study, the preselection of the early-stage data of the prediction model and
the selection of the later-stage model were optimized, and good results were
obtained. The gray system model has the advantage of allowing a small sample
size and does not require a priori information. However, it can be seen from the
results (Fig. 6.4) that the selection of the sample size has a certain range of
application, and a sample size that is too small or too large will affect it. The
accuracy of the final prediction model. In addition, the prediction results of GM
models with different orders are somewhat different, and the results show that the
prediction results of the 0-order GM (0, N ) model are better than those of the first-
order GM (1, N ) model (Fig. 6.5). This is not universal, and the fitting accuracy
results for CPUE sequences with different characteristics may be different. In
summary, in the construction of the gray prediction model, selecting the appropriate
original data series, identifying the key affecting factors, and comparing and screen-
ing a variety of different types of models can more accurately predict the changes in
the abundance index of Ommastrephes bartramii in the northern Pacific Ocean,
which will provide technical support for fishery production.
6.4.4.3 Resource Forecasting of the Pacific Stock of Scomber

australasicus Based on the Gray System
Australian mackerel, Scomber australasicus, is widely found in the Pacific Ocean,

and the major fishing countries include Japan, South Korea, and China. As an
important economic species in various countries, it is particularly important to
grasp the changes in the amount of Australian mackerel resources and to scientifi-
cally predict the amount of resources in fishery production and scientific manage-
ment. To this end, Zhang and Chen (2019) used the gray correlation and gray
prediction model to analyze the effect of environmental factors on the resources of
Australian mackerel based on the resource assessment data provided by the Central
Fisheries Research Institute of Japan. The gray forecast of Australian mackerel
resources can provide technical support for the sustainable development and scien-
tific management of Australian mackerel.
6.4.4.3.1 Materials and Methods
The resource data of Australian mackerel are from the Resource assessment report of
the Australian Mackerel in 2015. The time period is from 1995 to 2014, and the data
are the actual resource and catch data of Australian mackerel in this study. The
resource data of 1995–2012 were used for modeling, and the resource data of 2013
and 2014 were used for verification and comparison.
The marine environmental data include surface temperature (SST), Kuroshio tidal
level difference, and Pacific interdecadal oscillation (PDO). In this study, according
to the distribution area in the resource assessment report of Australian mackerel, the
area (140°E–160°E, 35°N–50°N) was used as the feeding ground. Two fields (130°
E–132°E, 30°N–32°N and 138°E-141°E, 34°N–35°N) were selected as spawning
grounds. In this study, SPSS19 software was used to analyze the correlation between
the monthly average temperature and the amount of resources in each region. The
monthly temperature with the highest linear correlation coefficient was selected as
the temperature (SST1) on the feeding ground and the temperature (SST2, SST3) on
the spawning ground. The spatial resolution of the surface temperature data is
1° × 1°, and the temporal resolution is monthly. The data are from the website
http://iridl.ldeo.columbia.edu/.
Studies have shown that the Kuroshio has a significant impact on pelagic fishery
resources. The strength of the Kuroshio Current is expressed by the tidal level
difference of the Kuroshio Current, and the annual average tidal range data are
selected from the website. The data are from the website http://www.data.jma.go.jp/.
The Pacific interdecadal oscillation PDO (PDO) is a climate change mode on an
interdecadal time scale. The PDO can directly cause interdecadal variability in the
climate in the Pacific Ocean and its surrounding areas and has an important modu-
lating effect on interannual variabilities, such as El Niño-Southern Oscillation
150 X. Chen et al.
(ENSO). Therefore, the annual average PDO is selected in this study. The data are
from the website http://www.research.jisao.washington.edu/pdo/PDO.latest.
The data were averaged and then subjected to general correlation analysis. See
Chap. 3 for the calculation method. The resolution coefficient is 0.1. In this study,
one GM (1, 1) model that does not consider environmental factors and four GM
(1, 2) models that consider environmental factors are established, which are the GM
(1, 2) model based on SST1, the GM (1, 2) based on SST2, the GM (1, 2) model
based on SST3, the GM (1, 2) model based on the tidal level difference, and the GM
(1, 5) model based on SST1, SST2, SST3 and the tidal level difference. The
modeling process of GM (1, 1) and GM (1, N) can be found in Chap. 5.
6.4.4.3.2 Correlation Analysis of Monthly Temperature Factors

and Resources of Australian Mackerel
The correlation between the SSTs of the spawning ground and the feeding ground
and the resources of Australian mackerel in each month (Table 6.24) shows that the
correlation between the SST and the amount of resources in the feeding ground in
August was the highest at 0.42, and the correlation between SST2 in Jan, SST3 in
May and the resource amount of Australian mackerel were the highest at 0.6 and
0.52, respectively. According to the results of previous studies, Australian mackerel
spawns in winter and spring and enters the feeding grounds from late June to early
September. Therefore, SST1 in August was selected to characterize the temperature
characteristics of the feeding ground, and SST2 in Jan and SST3 in May on the
spawning ground represent the temperature characteristics on the spawning ground.
6.4.4.3.3 Analysis of Environmental Factors Affecting the Resources of

Australian Mackerel Based on Gray Correlation
The averaging transformation is performed on each original series, and the gray
correlation degree is calculated. The gray correlation degree of each factor can be
obtained as follows:
LðSST1Þ = 0:8791
LðSST2Þ = 0:8709
LðSST3Þ = 0:8703
Lðtidal level differenceÞ = 0:8597
LðPDOÞ = 0:2312
The analysis shows that L(SST1) > L(SST2) > L(SST3) > L(Tide level
difference) > L(PDO). If L > 0.6 is selected as the environmental factor for the
6
Gray Prediction
Table 6.24 The correlation coefficient between monthly SST and Australian mackerel resources (Zhang and Chen 2019)
Month Jan. Feb. March April May June July Aug. Sep. Oct. Nov. Dec.
SST1 0.19 0.34 0.01 0.17 0.21 0.22 0.21 0.42 0.28 0.09 0.16 0.19
SST2 0.60 0.26 0.01 0.26 0.48 0.23 0.02 0.11 0.07 0.02 0.28 0.58
SST3 0.50 0.37 0.26 0.24 0.52 0.43 0.07 0.08 0.18 0.30 0.15 0.27
151
152 X. Chen et al.
establishment of the GM model, SST1, SST2, SST3, and the tide level difference are
used as environmental factors for establishing the model.
6.4.4.3.4 Establishment of a Gray Forecast Model for Australian Mackerel

Resources
Model 1: The GM (1, 1) model that does not consider environmental factors. The
GM (1, 1) model calculation was performed using the resource amount data, the gray
parameters a = -0.01280, b = 250.2298, and the response function B-
(t + 1) = 19861.4365exp(0.01280 t) - 19544.7203, with an average error of
18.65%.
Model 2: The GM (1, 2) model based on the temperature of the feeding field
SST1. The calculation shows that the gray parameters a = 0.09926 and b1 = 4.6495,
and the response function is B(t + 1) = (309.0000 - 46.8400 * SST1)EXP(-
0.09926 t) + 46.8400 * SST1, with an average error of 28.53%.
Model 3: The GM (1, 2) model based on the spawning field temperature SST2.
The calculation shows that the gray parameters a = 0.08862 and b1 = 3.9551, and
the response function is B(t + 1) = (309.0000 - 44.6291 * SST2)EXP(-
Model 4: GM (1, 2) model based on spawning field temperature SST3. The
calculation shows that the gray parameters a = 0.08999 and b1 = 3.9453, and the
response function is B(t + 1) = (309.0000 - 43.8408 * SST3)EXP(-
Model 5: The GM (1, 2) model based on the tidal level difference (T ). The
calculation shows that the gray parameters are a = 0.1269 and b1 = 7.0036, and the
response function is B(t + 1) = (309.0000 - 55.2094 * T)EXP(-
0.1269 t) + 55.2094 * T, with an average error of 33.79%.
Model 6: The GM (1, 5) model based on the feeding ground SST1, the spawning
ground SST2, the spawning ground SST3, and tidal level difference (T ). The
calculation shows that the gray parameters a = 0.2618, b1 = 58.7562, b2 = -
47.6616, b3 = -3.4524, and b4 = 8.7473, and the response function is B-
(t + 1) = (309.0000 - 224.4268 * SST1 + 182.04945) * SST2 + 13.18674 * SST3 -
33.41153 * T ) EXP(-0.26181 t) + 224.42684 * SST1 - 182.04945 * SST2 -
13.18674 * SST3 + 33.41153 * T, with an average error of 33.79%.
6.4.4.3.5 Comparison and Validation of Gray Prediction Models
The above model was used to evaluate the amount of resources of Australian
mackerel in 2013 and 2014. The specific results are shown in Table 6.25. The
average prediction error of the GM (1, 2) model based on SST1 is the smallest,
which is only 3.73%. The average prediction error of the GM (1, 2) model based on
SST2 is 4.41%. The average prediction error of the GM (1, 2) model based on SST2
Table 6.25 Prediction of Australian mackerel resources in 2013 and 2014 (Zhang and Chen
2019) Unit: thousand tonnes
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
2013 838.95 810.26 768.48 784.87 498.39 792.62
2014 896.34 822.01 788.65 767.36 936.88 1052.95
Mean error (%) 6.72 3.73 4.41 4.78 29.56 19.38
was 4.78%. The average prediction error of the GM (1, 2) model based on the tidal
level difference was 29.56%.
The results show that the gray prediction model established based on the SST of
the feeding ground and spawning ground has a high accuracy in forecasting the
resources of Australian mackerel and can be applied to subsequent fishery produc-
tion. Analysis of the gray parameter values a and b of the GM (1, 5) model shows
that among all the factors, SST2 and SST3 have the highest impact on the amount of
resources of Australian mackerel.
6.4.4.4 Gray Prediction of Diseases of Large Yellow Croaker in Cage

Culture
Prevention of farmed animal diseases is one of the keys to the healthy and sustain-
able development of the aquaculture industry. Owen et al. (2013) introduced the gray
system theory to explore the occurrence and development of bacterial diseases in
cage-cultured large yellow croaker and their relationship with environmental factors
and established a gray model for the prediction of bacterial diseases in cage-cultured
large yellow croaker. It is expected to provide a method for the prediction of
bacterial diseases in cage cultures of large yellow croaker and provide ideas and
approaches for the disease prediction of other aquaculture organisms.
Table 6.26 shows the incidence of bacterial diseases in cage-cultured large yellow
croaker in Zhoushan, Zhejiang Province, China, from 2001 to 2008. Analysis of the
incidence of bacterial diseases in large yellow croaker from 2001 to 2006 showed
that there was a large-scale outbreak of disease every year. The diseased water body
in the whole city was above 3000 m3, and the highest incidence was 36,000 m3. If we
can predict in advance and take active preventive measures, it is possible to reduce
the scope of the disease or avoid the occurrence of the disease.
From 2001 to 2008, we monitored the aquaculture, morbidity, and mortality of
large yellow croaker in cage culture in Dinghai District, Daishan District, and Putuo
District of Zhoushan City. At the same time, sampling points were set up in these
three areas, and the physical and chemical factors and biological factors at these
sampling points were regularly measured, including water temperature, salinity,
suspended matter, dissolved oxygen, pH, phosphate, silicate, nitrogen, ammonia
nitrogen, inorganic nitrogen, chemical oxygen demand (COD), zooplankton, and
phytoplankton species and quantity. The specific calculation of gray correlation is
154
Table 6.26 The incidence of bacterial diseases in large yellow croaker in 2001–2008 (Owen et al. 2013)
Month
Year Jan. Feb. March April May June July Aug. Sep. Oct. Nov. Dec.
2001 0 0 0 0 0 0 0 0 4002 0 0 0
2002 0 0 0 0 0 372 1939 1656 3386 1322 0 0
2003 0 0 0 0 210 270 1500 5300 2400 672 0 0
2004 0 400 0 1500 0 0 0 2700 1200 30,000 1500 0
2005 0 0 0 0 0 0 8640 0 0 0 0 0
2006 0 0 0 0 0 0 0 36,000 18,000 2400 2400
2007 0 0 0 0 0 0 0 0 0 0 0 0
2008 0 0 0 0 0 0 2000 0 0 0 0 0
X. Chen et al.
shown in Chap. 2, and its resolution coefficient is set to 0.5. See Chap. 5 for the GM
(1, N ) model.
6.4.4.4.1 Correlation Analysis of the Morbidity of Large Yellow Croaker

and Environmental Factors
To make full use of the information provided by changes in environmental factors on

the occurrence and development of bacterial diseases in large yellow croaker, it is
first necessary to understand the relationship between environmental factors and
their changes and the occurrence and development of bacterial diseases in large
yellow croaker. The incidence sequence of disease X0 is related to water temperature,
salinity, suspended matter, dissolved oxygen, pH, phosphate, silicate, nitrate nitro-
gen, nitrite nitrogen, ammonia nitrogen, inorganic nitrogen, COD, zooplankton,
phytoplankton, etc. The degree of correlation γ (X0, Xi) of the environmental factors
is shown in the first column of Table 6.27. Table 6.27 shows that the occurrence of
diseases in large yellow croaker was correlated with changes in 14 environmental
factors to varying degrees. The correlation degree above 0.8 is water temperature,
suspended matter, COD, inorganic nitrogen, nitrate nitrogen, nitrite nitrogen, and
ammonia nitrogen. Changes in these environmental factors are associated with the
occurrence and development of bacterial diseases in large yellow croaker. However,
these changing trends are not necessarily synchronized. When the external environ-
ment changes, such as the climate, some environmental factors in aquaculture waters
change first. When these changes reach a certain level and exceed the physiological
limit of fish, they may cause disease. To effectively monitor and warn of the
occurrence of fish diseases, it is necessary to consider the advanced relationship
between the changes in environmental factors and the occurrence of fish diseases,
Table 6.27 Correlation Correlation degree

between the incidence of bac-
Factor γ(X0, Xi) γ′(X0, Xi)
terial diseases and environ-
mental factors (Owen et al. Water temperature/°C 0.867 0.9067
2013) Salinity 0.6657 0.767
Suspended matter/(mg/L) 0.8564 0.8684
Dissolved oxygen/(mg/L) 0.7734 0.7709
pH 0.7953 0.7817
Phosphate/(mg/L) 0.799 0.7655
Silicate/(mg/L) 0.7214 0.7704
Nitrate/(mg/L) 0.8093 0.7201
Nitrite/(mg/L) 0.8075 0.7389
Ammonia/(mg/L) 0.8027 0.7209
Inorganic nitrogen/(mg/L) 0.8114 0.9037
COD/(mg/L) 0.844 0.8372
Phytoplankton/L 0.6895 0.6758
Zooplankton/(mg/m3) 0.7034 0.7045
156 X. Chen et al.
which is the leading indicator of fish diseases. Therefore, the degree of association
between the environmental factor sequence and the morbidity sequence is calculated
in advance by one order and is denoted as γ′ (X0, Xi). The results obtained are shown
in the second column of Table 6.27. Table 6.27 shows that water temperature,
suspended matter, inorganic nitrogen, and COD are the factors with γ′ (X0, Xi)
greater than 0.8. Therefore, these four factors are used as predictors of bacterial
diseases in large yellow croaker.
6.4.4.4.2 Establishment of the GM (1, N ) Model
Considering that there are many reasons for the occurrence of fish diseases, in
addition to their own problems, such as fish constitution, the changes in the envi-
ronmental factors of aquaculture waters are the incentives for the occurrence of
diseases. The establishment of the GM (1, N ) model considers the establishment of
the dynamic relationship between the incidence of large yellow croaker and envi-
ronmental factors. The GM (1, N) model was constructed based on the characteristic
data series of the incidence rate from May to October 2003, and the series of water
temperature, suspended matter, inorganic nitrogen, and COD were used to establish
the GM (1, 5) model. The fitted values of the model were calculated, and the
obtained simulation series are listed in columns 1–3 of Tables 6.28 together with
the primary incidence series and the residuals of each observation point. The average
relative error of the calculation model was 7.9791%. The average accuracy of the
forecast is 92.0209%.
In general, when the correlation factor series with a high degree of correlation
with the feature data series is introduced into the model, it will provide more
information for the prediction of the development trend of the feature data series.
However, the increase in the series of relevant factors will also increase the risk of
forecasting, especially when the correlation degree of these series of relevant factors
is very large. Due to the fluctuation of these series, the volatility of the model may be
increased, and the forecast error will increase. At the same time, the increase in the
series of relevant factors in the model will increase the difficulty of application.
Table 6.28 Predicted values and fitting residuals of GM (1, 5), GM (1, 4), and GM (1, 3) (Owen
et al. 2013)
GM (1, 5) GM (1, 4) GM (1, 3)
Observed Simulation Simulation Simulation
value value Residual value Residual value Residual
10.05 10.0500 10.0500 10.05
1.16 0.9821 0.1779 0.8748 0.2852 1.2046 -0.0446
6.03 5.8205 0.2095 6.0831 -0.0531 5.6076 0.4224
22.08 21.3068 0.7732 22.1177 -0.0377 21.337 0.734
10.00 10.6457 -0.6457 9.8093 0.1907 10.7148 -0.7148
3.17 3.5227 -0.3527 3.7466 -0.5766 3.3335 -0.1635
Therefore, a good model should ensure the fit of the model while keeping the
forecast variables as small as possible. Therefore, three and two different combina-
tions of these four correlation factors were selected to establish a GM (1, N ) (N = 3,
4) model, and four GM (1, 4) models and six GMs were obtained. (1, 3) model. The
relative errors of each group were compared. The average relative error of the four
GM (1, 4) models was 8.0262–11.1136%, which was higher than that of the GM
(1, 5) model. The model with the smallest average relative error contains suspended
matter, inorganic nitrogen, and COD. Comparing the six GM (1, 3) models, the
average relative error of the five models was higher than that of the GM (1, 5) model,
which was 11.707–62.8392%. The model is composed of inorganic nitrogen and
COD, and the expression of its G (1, 3) model is:
1 ð1Þ ð1Þ
xð1Þ ðk þ1Þ= 10:05- -54:7598x1 ðk þ1Þþ56:4861x2 ðk þ1Þ e -2:3307k
2:2092
1 ð1Þ ð1Þ
þ -54:7598x1 ðk þ1Þþ56:4861x2 ðk þ1Þ
2:2092
xð0Þ ðk þ1Þ=xð1Þ ðk þ1Þ-xð1Þ ðkÞ
where X1(0) and X2(0) are the inorganic nitrogen and COD sequences, respectively.
The simulation values and residuals of the optimal model are shown in columns 4–7
of Tables 6.28.
According to the established GM (1, N ) model, as long as the values of the
relevant factors in the current period are measured, the incidence of bacterial
diseases in cage-cultured large yellow croaker in the next period can be predicted.
Comparing the GM (1, 5) model and the GM (1, 3) model, the GM (l, 3) model lacks
the two environmental factor sequences of water temperature and suspended matter,
but the fitted residuals of GM (1, 3) are better than those of GM (1, 5). As a leading
indicator of water temperature, the correlation between water temperature and the
occurrence of bacterial diseases in large yellow croaker was the largest. The increase
in water temperature is a prerequisite for the occurrence of bacterial diseases in large
yellow croaker. When the water temperature reaches a certain range, the occurrence
and development of the disease depends on the water quality of the aquaculture
waters. The increase in the amount of suspended matter will affect the photosynthe-
sis of phytoplankton, increase the consumption of organic matter, and change the
physical and chemical properties of the water body, such as inorganic nitrogen and
COD. Its effect on the incidence is mainly reflected in the changes in inorganic
nitrogen and COD. Therefore, two environmental factors, water temperature and
suspended matter, were added to the model, which, in contrast, increased the
uncertainty of the model and reduced the fitting accuracy.
158 X. Chen et al.
6.4.4.5 Prediction of Gray Catastrophe of Fishery Resources
6.4.4.5.1 Prediction of Gray Catastrophe for the Resource Abundance

of Ommastrephes bartramii
As an important economic cephalopod in the northwestern Pacific Ocean, the

resources of Ommastrephes bartramii have drastically changed over the years and
are difficult to predict. Therefore, Xie and Chen (2020) attempted to establish a
catastrophe prediction model for the abundance of Ommastrephes bartramii to
provide a basis for national exploitation based on the gray theory.
Data and Methods

The fishery production data were from the Chinese Squid Fishing vessels. The time
period was from June to November of 1995 to 2017. The spatial range was 35°–45°
N and 140°–179°E. The statistical contents included date, longitude, and latitude.
The spatial resolution is 1° × 1°. The catch per unit effort (CPUE) was used to
characterize the abundance index of Ommastrephes bartramii. With 1° × 1° as a
fishing area, the CPUE value of each fishing area in each year was calculated.
The total catch obtained in each year was counted, and the number of fishing
vessels in each year was used as the fishing effort. The generalized linear regression
model (GLM) was used to normalize the nominal CPUE. The expression of GLM is
glm[CPUE] ~ SST + Lon + Lat + ε.
The GM (1, 1) model was used to predict the abundance index of Ommastrephes
bartramii in the North Pacific. The specific method is as follows:
Set the original sequence X = (x(1), x(2) ..., x(n)), and take the third quantile Q3 of
the sequence (a value greater than 75%) as the upper limit catastrophe value ξ1, and
the first quantile Q1 of the series (a value less than 25%) is taken as the lower
catastrophe value ξ2. The points greater or less than ξ in sequence X are considered
outliers, and they are composed of the upper limit or the lower limit catastrophe
sequence Xξ = (x[q(1)], x[q(2)]. . ., x[q(m)]). Let the catastrophe date series
Q(0) = (Q(1), q(2). . ., q(m)), and a GM (1, 1) model is established for disaster
prediction. The GM (1, 1) model establishment method is described in Chap. 5.
CPUE Distribution and Its Catastrophe Point

The nominal average CPUE and normalized CPUE for each year from 1995 to 2017
are shown in Fig. 6.7. The normalized CPUE of the GLM showed basically the same
variation trend as the nominal CPUE, but the fluctuation was small. In 2007, the
nominal CPUE value was the largest, which was 4.302 t. d -1, and the nominal
CPUE in 2009 was the smallest, which was 1.307 t. d -1. The standardized CPUE
value was the largest in 2017, which was 2.915 t. d -1, and the nominal CPUE was
the smallest in 2002, which was 1.399 t. d -1. According to the classification of
catastrophe values, the upper catastrophe points are 2005, 2007, 2008, 2010, 2014,
4.5
Nominal CPUE
Standardized CPUE
4.0 Upper limit of catastrophic values
Lower limit of catastrophic values
CPUE (t per fishing vessel each year)
3.5
3.0
2.5
2.0
1.5
1.0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Year
Fig. 6.7 Distribution of the resource abundance of Ommastrephes bartramii in the Northwest
Pacific Ocean from 1995 to 2017 (Xie and Chen 2020)
Table 6.29 The relative error between the predicted value and the actual value of Ommastrephes
bartramii (Xie and Chen 2020)
Lower limit catastrophe Fitted Relative Upper limit catastrophe Fitted Relative
sequence number value error serial number value error
X5 5.08 1.67 X13 12.26 5.68
X7 7.47 6.73 X14 14.32 2.28
X8 10.98 37.25 X16 16.72 4.52
X21 16.14 23.16 X20 19.53 2.35
X22 23.72 7.80 X23 22.81 0.86
Average relative error 15.32 3.13
and 2017, and the lower catastrophe points are 1996, 1999, 2001, 2002, 2015, and
2016.
Establishment and Validation of the Gray Catastrophe Model

The GM (1, 1) model is constructed using the upper catastrophe point and the lower
catastrophe point. From the perspective of model accuracy (Table 6.29), the average
correlation error of the lower catastrophe prediction model is 15.32%, and the
average correlation error of the upper catastrophe prediction model is only 3.13%.
From the perspective of the relevant parameters of the model, the development
coefficient a reflects the degree of coordination between the main factor and each
subfactor. When a ≤ 0.3, the model can be used for medium- and long-term
prediction; when 0.3 < -a ≤ 0.5. The lower-bound catastrophe prediction model
(a = -0.385) can be used for short-term prediction, and the lower-bound catastrophe
prediction model (a = -0.155) can be used for medium- and long-term prediction
(Table 6.29). When the posterior ratio c < 0.35 and the small error probability
160 X. Chen et al.
Table 6.30 Relevant parameters of the gray catastrophe GM (1, 1) model (Xie and Chen 2020)
Development Posterior Small error
coefficient a ratio c probability p Response function
Lower limit catas- -0.385 0.344 1.000 X(t + 1) = 10.823exp
trophe model (0.385 t) - 8.823
Upper limit catas- -0.155 0.128 1.000 X(t + 1) = 73.058exp
trophe model (0.155 t) - 62.058
p > 0.95, the model is reliable, the accuracy level is level 1, and the c and p values of
the upper- and lower-bound catastrophe prediction models meet the requirements
(Table 6.29).
According to the response function in Table 6.30, the time when the next
occurrence number of the lower limit catastrophe is approximately 34.86, i.e., the
resource abundance under year will occur approximately 12 years after the occur-
rence; the next occurrence number of the upper limit catastrophe is approximately
26.64. The resource abundance year will occur in the fourth year after the occurrence
of (Table 6.30). From the analysis of the average error of the model (Table 6.30), the
GM (1, 1) model can effectively predict the occurrence time of a catastrophe.
6.4.4.5.2 Gray Catastrophe Prediction of the Abundance of Illex argentinus

in the Waters of the Malvinas Islands
Argentine flying squid (Illex argentinus) are shallow oceanic species and important
economic species. It is particularly abundant at 35°–52°S. It is currently one of the
most important cephalopod resources in the world. Among them, the waters of the
Malvinas Islands are one of the important fishing grounds of Illex argentinus. The
average annual catch in this sea area is 200 thousand t, of which squid production
accounts for approximately 75% of the total. In high-yield years, the catch around
the Malvinas Islands provides 10% of the world’s total squid. There is significant
interannual variation in the amount of Illex argentinus, which may be because the
early life history and habitat are highly susceptible to the impact of the marine
environment and climatic factors. To this end, Xu et al. (2022) used the gray
catastrophe prediction GM (1, 1) model in gray system theory to scientifically
predict the year of highest or lowest catch of Illex argentinus, which will provide a
reasonable scientific basis for the management and sustainable development of
fishery resources.
Data and Research Methods

The fishery data in this study are from the annual fishery statistics report of the
Fisheries Bureau of the Malvinas Islands. The data are the annual catch and the
number of vessels operated from 1995 to 2019. The annual fishing yield per vessel
(CPUE, t/ship) is used to characterize the abundance of Illex argentinus. The GM
350000 3500
Catch
CPUE
300000 3000
Upper limit
CPUE (t per fishing vessel each year)

Lower limit
250000 2500
Annual catch (t)
200000 2000
150000 1500
100000 1000
50000 500
0 0
1995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019
Year
Fig. 6.8 Catch and CPUE of Illex angentinus in the Malvinas Islands from 1995 to 2019
(1, 1) model was used to predict the abundance of Illex argentinus resources in the
Malvinas Islands. The specific modeling process and model verification are shown in
Chap. 5.
Changes in the Production and CPUE of Illex argentinus

According to the statistical analysis, during the 25 years from 1995 to 2019, the
annual production of Illex argentinus in the waters of the Malvinas Islands fluctuated
significantly (Fig. 6.8). The annual average production of Illex argentinus in 2009
was the lowest, only 3 tons. In the following years, the yield began to rise and
reached the highest yield in 2015, reaching 332,863 t. In 2016, the yield decreased
sharply to only 2297 t, and the yield in the following years was at a low level.
During the period from 1995 to 2019, the average annual CPUE of Illex
argentinus was 1118.807 t/ship, and the variation trend of CPUE was basically
consistent with the variation trend of production. In 2009, the CPUE value was the
smallest, only 0.143 t/ship. In the following years, the CPUE began to rise. In 2015,
the CPUE reached a maximum value of 3140.217 t/ship. In 2016, the CPUE
decreased sharply to only 22.086 t/ship. However, the CPUE in 2007 and 2008
was abnormal. Although the annual production was low, the CPUE was relatively
high. The reason may be related to the decrease in the number of fishing vessels
operating in that year.
Establishment and Testing of the GM (1, 1) Model

According to the calculation results of the upper and lower limits of the catastrophic
value (the upper catastrophe value is 1615 t/ship, and the lower catastrophe value is
784 t/ship), the years of rich years are determined to be 1999, 2000, 2007, 2008,
2014, and 2015. The corresponding time series are 5, 6, 13, 14, 20, and 21; the years
of poor years are 1995, 1996, 2004, 2005, 2009, 2010, 2016, 2017, 2018, and 2019.
The corresponding time series are 1, 2, 8, 10, 11, 15, 16, 22, 23, 24, and 25.
162 X. Chen et al.
Table 6.31 Relative error between the predicted value and the true value of the GM (1, 1) model
(Xu et al. 2022)
Serial Predicted Relative error/ Serial Predicted Relative error/
number value % number value %
Q13 11.214 13.741 Q8 8.887 11.089
Q14 14.168 1.200 Q10 10.313 3.132
Q20 17.901 10.497 Q11 11.968 8.800
Q21 22.616 7.698 Q15 13.888 7.411
Q16 16.117 0.731
Q22 18.703 14.986
Q23 21.704 5.634
Q24 25.187 4.945
Q25 29.228 16.913
Average relative error 8.284 Average relative error 8.182
Table 6.32 Related parameters and prediction results of the GM (1, 1) model (Xu et al. 2022)
Development Posterior Small error Catastrophe Catastrophe
Model coefficient (a) ratio(c) probability(P) point point number
Rich -0.234 0.285 1.000 X0(7) 28.57
year X0(8) 36.10
X0(9) 45.61
Poor -0.149 0.288 1.000 X0(12) 33.92
year X0(13) 39.36
X0(14) 45.677
The GM (1, 1) model for forecasting rich years is X(t) = 33.689exp(0.234 t) -

28.689.
The GM (1, 1) model for the prediction of poor years is X(t) = 57.527exp
(0.129 t) - 56.527.
The relative error between the predicted value of the catastrophe model and the
actual value is shown in Tables 6.31. For the GM (1, 1) model for forecasting rich
years, the average relative error was 1.20–13.8%, and the average relative error was
8.824%. For the GM (1, 1) model for forecasting poor years, the average relative
error was 3.13–16.92%, and the average relative error was 8.182%. The prediction
parameters of the model are shown in Tables 6.32. In the prediction model for the
abundance catastrophe of Illex argentinus in this study, both the rich year forecasting
model (a = -0.234) and the poor year forecasting model (a = -0.149) can be used
for medium- and long-term forecasting. The variance ratio was less than 0.35, the
small probability error P value was 1.000, greater than 0.95 (Table 6.32), and the
accuracy of the model was grade I.
The last abundance year was 2015, and the corresponding serial numbers for the
next three occurrences that exceeded the catastrophic point threshold were 28.57,
36.10, and 45.61 (Table 6.32). Therefore, the years of rich resources in the future are
2021, 2028, and 2038. Similarly, the last apocalyptic year was 2019, and the
numbers corresponding to the next three occurrences that exceeded the catastrophic
point threshold were 33.92, 39.36, and 45.677 (Table 6.32); therefore, the years with
poor resources in the future abundance are 2024, 2029, and 2036.
Through the processing of the original CPUE data of Illex angentinus in the
waters of the Malvinas Islands from 1995 to 2019 and the use of the gray catastrophe
model to better predict the rich years and poor years, the average relative errors of the
models for the rich years and the poor years are 8.284%. This result can provide
guidance for fishery production. According to relevant data in 2021, the fishing yield
of Illex angentinus in the waters of the Malvinas Islands reached 172,000 tons, and
the average CPUE exceeded the upper limit of the catastrophe value of 1615 t/ship,
which is a good year. The predicted value is credible.
There are many factors that affect the establishment of the catastrophe prediction
model, such as the delineation of the upper and lower limits of the catastrophe, the
length of the prediction time, and the environmental factors. The determination of
the upper and lower limits of the catastrophe, as well as the selection and optimiza-
tion of the prediction time, can be compared and determined by establishing different
catastrophe prediction models to overcome this problem. In addition, in the subse-
quent establishment of catastrophe prediction models, it is possible to consider the
climate and environmental factors that affect the abundance of Illex angentinus.
References
Chen XJ, Zhou YQ (2001) Analysis and forecast of manpower resources in Chinese Marine
Fisheries by using grey theory. J Zhanjiang Ocean Univ 21(1):22–29. (In Chinese)
Guo M (1992) Gray forecast of shrimp yield in Bohai Sea. Fish Sci 3:10–14. (In Chinese)
Liu SF, Guo TB, Dang YG (1999) Gray system theory and its application. Science Press.
(In Chinese)
Lu Q, Fang Z, Li N et al (2022) Prediction model of fisheries catch based on GM (1, N) in the Indian
Ocean. J Fish China:1–8. (In Chinese)
Owen MK, Ni HE, Wang LG et al (2013) A forecasting model for bacterial disease of cage cultured
large yellow croaker (Pseudosciaena crocea) based on grey system theory. J Fish China 37(6):
Xie MY, Chen XJ (2020) Grey catastrophe year prediction for the abundance of neon flying squid
(Ommastrephes bartramii) in the Northwest Pacific. Haiyang Xuebao 42(4):40–46. (In Chinese)
Xie MY, Chen XJ (2021a) Analysis of the fishing seasons characteristics of Ommastrephes
bartramii and prediction of the main fishing seasons based on the grey system theory. Prog
Fish Sci 42(4):1–8. (In Chinese)
Xie MY, Chen XJ (2021b) Prediction of abundance index of Ommastrephes bartramii in the North
Pacific Ocean based on different order grey system models. J Shanghai Ocean Univ 30(4):
Xu ZA, Xie MY, Chen XJ (2022) Grey catastrophe year prediction for the abundance index of Illes
angentinus in the waters near Malvinas islands. J Shanghai Ocean Univ 31(3):642–649.
(In Chinese)
Zhang C, Chen XJ (2019) Forecasting model for spotted mackerel biomass based on grey system
theory. J Shanghai Ocean Univ 28(1):154–160. (In Chinese)
Chapter 7
Gray Decision
Xinjun Chen
Abstract Decision-making is making a choice or a decision, and it is a key step in

the process of solving a problem. Decision-making has both its technical side, that is,
the search process of alternatives, and decision-making has its adaptive side, that is,
the selection process of alternatives. Decision-making is not only an individual
behavior, but also an organizational behavior. In the actual management activities,
the decision-making is influenced by the organizational environment, to avoid
making bad decisions. According to Herbert Symon, the essence of management
is decision-making, which runs through the whole process of management and
determines the success or failure of the whole management activity. Usually,
because of the complexity of the social economic system and the crisscross of
decision factors, it is difficult for any decision maker to make the optimal decision
based on intuition and experience. Therefore, in the modern scientific decision-
making, often with the help of the method of natural science, the use of mathematical
tools, the establishment of the relationship between the decision-making variables
formula and model, to reflect the essence of the decision-making problem, simplify
complex decision problems. The general form of the decision model is V = F (Ai, Sj),
in which V is the value goal, Ai is the controllable decision factor, and Sj is the
uncontrollable decision factor. In order to obtain the production plan or plan with the
greatest benefit, the decision-making model is listed first, and then the optimal one is
obtained through the model. In the process of decision-making, many uncertain
factors affect the choice of decision-making scheme, often there are gray factors and
gray process, which is especially obvious in the fishery production system, for
example, how the marine environment factors affect the change of the fishery
resources, how they affect the fishery production and the central fishery, thus
affecting the decision-making of the fishery production, so the Gray system theory
is applied to the decision-making, this problem has been well solved. Gray decision-
making is one of the important contents of gray system theory. In this chapter, the
basic concept of gray decision-making, gray correlation decision-making, gray
X. Chen (✉)
166 X. Chen
situation decision-making, and the application of gray decision-making in fishery

science will be introduced.
Keywords Gray system · Decision-making · Fishery science · Gray correlation

decision-making · Gray situation decision-making
7.1 Basic Concepts of Gray Decision-Making
Decision-making is one of the basic activities of human society, politics, and

economic life. What is decision-making? People often interpret it as “deciding
policy” or “finalizing the case” in a narrow sense. In fact, as a specific term,
decision-making has very broad meaning. Any decision to act based on a
predetermined goal can be called decision-making. More precisely, decision-making
is a decision made on the direction, principle, and goal of future practice, as well as
the methods and means to adhere to the direction, implement the principle, and
achieve the goal.
A decision-making process can be roughly divided into three basic steps (Chen
2003, 2023):
The first step is to identify the problem and propose decision-making goals.
Without goals, there is no decision-making. Due to the complexity of objective
things, the goal is often not one but the integration and coordination of several goals.
At the same time, there are contradictions between different goals. Therefore, special
attention should be given to the handling, coordination, and integration of
multiobjective problems. The processing methods are as follows: subordinate targets
are eliminated, similar targets are merged, secondary targets are reduced to con-
straint conditions, and various comprehensive methods of targets and layered
screening and elimination methods of targets are used. For example, in the manage-
ment of fishery resources, currently, the maximum sustainable yield (MSY), the
maximum economic yield (MEY), and the optimal yield (OY) are usually used as the
management objectives, but these objectives are often contradictory, so each country
should adapt to local conditions. Therefore, each country should formulate manage-
ment objectives based on local conditions and coordinate and unify different
objectives.
In the second step, through data collection, analysis and prediction research,
design plans, and technical evaluation, various possible alternative plans are
drawn up.
The third step is to select the most suitable scheme from various possible
schemes. Based on the evaluation and analysis of various schemes, the decision is
made decisively using the scientific method of selection according to certain
decision-making criteria.
7 Gray Decision 167
7.2 Gray Correlation Decision-Making
7.2.1 Basic Concepts
Now we have S = {sij| ai 2 A, bj 2 B} is the situation set, ui0 j0 =

ð1Þ ð2Þ ðsÞ
ui0 j0 , ui0 j0 , . . . , ui0 j0 is the optimal effect vector. If ui0 j0 is the corresponding
situation ui0 j02
= S, then we call ui0 j0 the ideal optimal effect vector, and its
corresponding Si0 j0 is called the ideal optimal situation.
Now, the effect vector corresponding to situation Sij is
ð1Þ ð2Þ ðsÞ

uij = uij , uij , . . . , uij ; i = 1, 2, . . . , n; j = 1, 2, . . . , m
ðk Þ ðk Þ
1. When the target effect value of k is larger, the value is better, ui0 j0 = max uij
1≤i≤n
1≤j≤m
.
2. When the target effect value of k is close to a certain moderate value u0, which is
ðk Þ
good, it is taken as good, ui0 j0 = u0 .
3. When the target effect value of k is smaller, the value is better, and it is taken as
ðk Þ ðk Þ
ui0 j0 = min uij .
1≤i≤n
1≤j≤m
ð1Þ ð2Þ ðsÞ

Then, ui0 j0 = ðui0 j0 , ui0 j0 , . . . , ui0 j0 Þ is the ideal optimal effect vector.
Let S = {sij| ai 2 A, bj 2 B} be the situation set, and the effect vector
corresponding to situation Sij is
ð1Þ ð2Þ ðsÞ

ð1Þ ð2Þ ðsÞ

ui0 j0 = ui0 j0 , ui0 j0 , . . . , ui0 j0 is the ideal optimal effect vector.
εij (i = 1, 2 ..., n; j = 1, 2 ..., m) is the gray absolute correlation degree between uij
and ui0 j0 . If εi1 j1 satisfies for any i 2 {1, 2, . . ., n} and i ≠ i1 and any j 2 {1, 2, . . ., m}
and j ≠ j1, there is always εi1 j1 ≥ εij , then ui1 j1 is the suboptimal effect vector, and Si1 j1
is a suboptimal situation.
168 X. Chen
7.2.2 Basic Steps of Gray Correlation Decision-Making

Calculation
Gray correlation decision-making can be carried out according to the following steps
(Chen 2003, 2023):
Step 1: Determine the event set A = {a1, a2, . . ., an} and the game set B {b1, b2,
. . ., bm}, and construct the situation set S = {sij = (ai , bi)| ai 2 A, bi 2 B}.
Step 2: Determine decision-making goals 1, 2, . . ., s.
Step 3: Find the effect value of different situations sij (i = 1, 2 ..., n; j = 1, 2 ..., m)
ðk Þ
under the k target uij
ðk Þ ðk Þ ðk Þ ðk Þ ðk Þ ðk Þ ðk Þ ðk Þ
uðkÞ = u11 , u12 , . . . , u1m ; u21 , u22 , . . . , u2m ; un1 , un2 , . . . , uðnm
kÞ
; k = 1, 2, . . . , s:
Step 4: Find the mean image of the situation effect sequence u(k) under the
k target, which is still denoted as
ðk Þ ðk Þ ðk Þ ðk Þ ðk Þ ðk Þ ðk Þ ðk Þ
uðkÞ = u11 , u12 , . . . u1m ; u21 , u22 , . . . u2m ; un1 , un2 , . . . , uðnm
kÞ
; k = 1, 2, . . . , s:
Step 5: Write the effect vector of situation sij through the fourth step.
ð1Þ ð2Þ ðsÞ

Step 6: Find the idealized optimal effect vector.
ð1Þ ð2Þ ðsÞ

ui0 j0 = ui0 j0 , ui0 j0 , . . . , ui0 j0
Step 7: Calculate the gray absolute correlation degree εij between uij and ui0 j0 ,
i = 1, 2, . . ., n; j = 1, 2, . . ., m.
Step 8: From max εij = εi1 j1 , we obtain suboptimal effect vector ui1 j1 and
1≤i≤n
1≤j≤m
suboptimal situation si1 j1 .
7.3 Decision-Making in the Gray Situation
Situational decision-making refers to the whole process of selecting the best for a
certain target under the premise of the unification of events, countermeasures, and
effects. When the event and the countermeasure are quantified and the event and the
decision form a paired combination of decision-making, it is called the gray situation
7 Gray Decision 169
decision-making method. If a single target is considered in decision-making, we call

it single-target situation decision-making; if multiple targets are considered at the
same time and the effect of the same situation under different targets is different,
comprehensive analysis and overall consideration are needed to integrate these
targets into one. Decision-making is called multiobjective gray situation decision-
making.
Due to the incommensurability and contradiction between multiple objectives, it
is impossible to simply merge multiple objectives into a single objective for
decision-making. Gray system theory uses methods such as effect measurement
calculation and target weighted averaging to convert multiobjective decision-making
into single-objective decision-making. Circumstance restrictions. Therefore,
multiobjective gray situation decision-making is a gray decision-making method
with a wide range of adaptations (Chen 2003, 2023). The mathematical method for
multiobjective situation decision-making is as follows:
7.3.1 Decision Element, Decision Vector, and Decision

Matrix
The event is denoted as ai, the countermeasure is bj, and its binary combination (ai,
bj) is called the situation and is denoted as Sij = (ai, bj). Its meaning is the jth
countermeasure (bj) to address the situation of the ith event (ai).
r ij r ij
=
Sij ai , b j
If there are events a1, a2, . . ., an, and there are countermeasures b1, b2, . . ., bm,
then for the same event ai, we can use b1, b2, . . ., bm, etc., m countermeasures to deal
with, thus forming (ai, b1), (ai, b2) ... (ai, bm) and m other situations. The decision-
making elements corresponding to these situations can be arranged in a row to form a
decision-making row:
r i1 r i2 r
δi = , , . . . , im
Si1 Si2 Sim
where rij is the measure of the effect of situation Sij.

Similarly, for countermeasure bi, we can use a1, a2 ..., an to match, and the
corresponding decision elements can be arranged in a row to form a decision
column:
170 X. Chen
r 1j r 2j r nj
θi = , , ...,
S1j S2j Snj
Arrange the decision rows (i = 1, 2, . . ., n) and the decision columns θj ( j = 1,

2, . . ., m) to form the matrix M.
r 11 r 12 r 1m
...
S11 S12 S1m
r 21 r 22 r
M= . . . 2m
S21 S22 S2m
r n1 r n2 r
. . . nm
Sn1 Sn2 Snm
Then, M is called the situation decision matrix, which can be denoted as M(δi, Θj).
7.3.2 Effectiveness Measure
The effect measurement is the measurement of the actual effect produced by the
situation compared to the target. The time series is the correlation coefficient of two
comparison series at the same time. The calculation formula is
Δmin þ Δmax
γ ij ðt Þ =
Δij ðt Þ þ Δmax
where Δmin, Δmax is the minimum difference and maximum difference (absolute
value) of the subtraction of the two comparison series at each time.
Δij(t) is the difference at any time t.
For a single point, the effect measure can be divided into (Chen 2003, 2023):
1. Upper limit effect measurement. The calculation formula is
uij
γ ij =
umax
uij ≤ umax
where uij is the measured effect of situation Sij; umax is the maximum value of all
the measured effects of situation Sij.
γ ij ≤ 1
2. Measurement of the lower limit effect. The calculation formula is

7 Gray Decision 171
umin
γ ij =
uij
uij ≥ umin
where umin is the minimum value of all the measured effects of situation Sij.
γ ij ≤ 1
3. Central effect measurement. The calculation formula is
min uij , uo
γ ij =
max uij , uo
where u0 is the reference point of the sample.

Or use the formula:
uo
γ ij =
uij - u0 þ u0
γ ij ≤ 1
In practical applications, which effect measurement is used depends on the nature

of the target. For example, for efficiency indicators such as yield and output value,
the larger the better, the upper limit effect measurement is used. For example, for
cost-based indicators such as investment and cost, the smaller the better, the lower
limit effect measurement is used. For example, for indicators such as fertilization and
irrigation, the appropriate amount is appropriate.
In addition, the steady-state effect measurement of the system can also be used for
the time series of the benefits of the situation. That is, the GM (1, 1) model is
established for the time series to obtain the parameters.a = au Then, 1a is the measure
of the steady-state effect of the system, denoted as:
1
γ ij =
a
7.3.3 Multiobjective Decision Matrix
When the situation has several targets, then the effect measure of the Kth target is
γ ij ðK Þ
recorded as γ ij(K ), and its corresponding decision element is Sij , for which there is a
ðK Þ ðK Þ
corresponding decision vector δi , θji and decision matrix M(K )
172 X. Chen
r ðK Þ 11 r ðkÞ 12 r ðK Þ 1m
...
S11 S12 S1m
M ðK Þ = r ðK Þ 21 r ðK Þ 22 r ðK Þ 2m
...
S21 S22 S2m
ðK Þ ðK Þ ðK Þ
r n1 r n2 r nm
...
Sn1 Sn2 Snm
Then, the comprehensive multiobjective situation decision matrix M(∑) is
r ðΣÞ 11 r ðΣÞ 12 r ðΣÞ 1m

...
S11 S12 S1m
ðΣÞ ðΣÞ ðΣÞ
M ðΣÞ = r 21 r 22 r 2m
...
S21 S22 S2m
r ðΣÞ n1 r ðΣÞ n2 r ðΣÞ nm
...
Sn1 Sn2 Snm
N ðK Þ
Σ γ ij
The elements in the matrix are calculated as follows: γ ij ðΣÞ = K = 1N .
When the K targets are biased according to the purpose and requirements of the
decision-making, different weights ωp can be given, and the elements of the matrix
can be calculated as follows:
N
ðK Þ
γ ij ðΣÞ = ωP γ ij
P=1
7.3.4 Decision-Making Criteria
Decision-making selects the best situation. If the best countermeasure is selected by

the event, the “line decision” is performed, while the “column decision” is performed
by the countermeasure matching the most suitable event. The method is
Row decision: for the decision matrix M(∑) in the decision row δi, the decision
element with the largest effect of the countermeasure is obtained:
ðΣÞ ðΣÞ ðΣÞ ðΣÞ ðΣÞ

γ ij1 = max γ ij = max γ i1 , γ i2 , . . . , γ im
j
ð ΣÞ
γ
Then, Sijij is the decision element, Sij* is the optimal decision-making situation,
i.e., bj* is the optimal countermeasure for event ai.
7 Gray Decision 173
Column game: For the decision matrix M(∑) in the decision column θj, the
decision element with the largest effect of the countermeasure is obtained:
T
ðΣÞ ðΣÞ ðΣÞ ðΣÞ ðΣÞ
γ ij = max γ ij = max λ1j , γ 2j , . . . γ nj
i
ð ΣÞ
γ
Then, Sijij is called the column decision element. Si*j is the optimal decision-
making situation, i.e., ai* is the optimal countermeasure for event ai.
In the actual decision-making process, according to the above criteria, the row
decision and column decision are made on the matrix, and the resulting decision is
often difficult to develop in a coordinated manner in the overall situation, so the goal
of the overall benefit cannot be achieved. In this case, the comprehensive matrix
needs to be adjusted. Gray target decision-making can be performed after optimiza-
tion or normalization.
7.3.5 Prioritized Decision Matrix
The optimization of the comprehensive decision matrix M(∑) can be performed in

two steps:
The first step is to arrange the decision elements in the matrix row by row from
left to right according to the size to obtain the comprehensive decision matrix M(∑)
ðΣÞ
and the matrix M 1 after the row optimization:
0:472 0:240 0:069 0:085

S11 S12 S13 S14
0:416 0:157 0:067 0:268
S21 S22 S23 S24
0:40 0:321 0:042 0:136
S31 S32 S33 S34
0:382 0:449 0:077 0:068
M ðΣÞ =
S41 S42 S43 S44
0:573 0:278 0:107 0:196
S51 S52 S43 S54
0:331 0:193 0:062 0:052
S61 S62 S63 S64
0:393 0:329 0:090 0:083
S71 S72 S73 S74
174 X. Chen
0:472 0:240 0:085 0:069

S11 S12 S14 S13
0:416 0:268 0:157 0:067
S21 S24 S22 S23
0:40 0:321 0:136 0:042
S31 S32 S34 S33
0:449 0:382 0:077 0:068
M 1 ðΣÞ =
S42 S41 S43 S44
0:573 0:278 0:196 0:107
S51 S52 S54 S53
0:331 0:193 0:062 0:052
S61 S62 S63 S64
0:393 0:329 0:090 0:083
S71 S72 S73 S74
Step 2: Next, the row-optimized ordering matrix is used to arrange the decision
ðΣÞ
elements from top to bottom in order of magnitude to obtain the matrix M 2 , which
is as follows:
0:573 0:382 0:196 0:107

S51 S41 S54 S53
0:472 0:329 0:157 0:083
S11 S72 S22 S74
0:449 0:321 0:136 0:069
S42 S32 S34 S13
0:416 0:278 0:090 0:068
M 2 ðΣÞ =
S21 S52 S73 S44
0:40 0:268 0:085 0:067
S31 S24 S14 S23
0:393 0:240 0:077 0:052
S71 S12 S43 S64
0:331 0:193 0:062 0:042
S61 S62 S63 S33
Then, we perform another check by row. If row optimization is not achieved, row
permutation can be performed again. In this way, the decision-making matrix is
gradually reduced from the upper left corner to form the optimal matrix M*.
According to the optimal ordering decision matrix, the specific principles and
methods for gray situation decision-making are as follows:
1. The optimal ordering matrix is divided into several steps along the main diagonal
direction so that the values of the elements of the previous step are all greater than
the values of the elements of the next step. The above example can be divided into
the following five steps.
7 Gray Decision 175
2. Under normal circumstances, the situation within the same rung can be selected
according to the same advantages and disadvantages; that is, the situation in the
upper rung is better than the situation in the lower rung and vice versa.
3. The order of preference should be performed from top to bottom, step by step. If
necessary, the effect measures were compared.
4. The principle of comprehensive coordination should be adhered to in decision-

making. According to the specific purpose and requirements, within the same
ladder, each event can choose 2–4 countermeasures, and each countermeasure
can match 2–4 events to avoid the “one size fits all” in the selection of the best and
prevent one event from selecting all countermeasures or one countermeasure to
match all events.
5. The principle of combining qualitative and quantitative decisions should be
adhered to. An echelon is similar to a “gray target,” in which a satisfactory
situation can be found from a nonoptimal situation, and a coordinated situation
can be found from a noninferior situation.
7.3.6 Normalized Decision Matrix
For the comprehensive decision matrix M(∑), the normalization transformation can
be divided into two categories. First, the normalization process is performed row by
row, and the calculation formula is
176 X. Chen
γ ij
γ 0ij = m i = 1, 2, . . . n
γ ij
j=1
ðΣÞ
In this way, the row-normalized matrix is obtained M 1 . This matrix can reflect
the proportion of each countermeasure in the comprehensive effect measurement of
each event.
The other is to perform normalization column by column, and the calculation
formula is
γ ij
γ 0ij = n j = 1, 2, . . . , m
γ ij
i=1
ðΣÞ
In this way, the column normalization matrix is obtained M 2 . This matrix
reflects the proportion of each event in the comprehensive effect measurement of
each countermeasure.
For the above example matrix M(∑) after row and column normalization, the
ðΣÞ ðΣÞ
matrices M 1 and M 2 are obtained.
0:545 0:277 0:080 0:098

S11 S12 S13 S14
0:458 0:173 0:074 0:295
S21 S22 S23 S24
0:445 0:357 0:047 0:151
S31 S32 S33 S34
0:391 0:460 0:079 0:070
M 1 ðΣÞ =
S41 S42 S43 S44
0:497 0:240 0:093 0:170
S51 S52 S53 S54
0:519 0:303 0:097 0:081
S61 S62 S63 S64
0:439 0:368 0:100 0:093
S71 S72 S73 S74
7 Gray Decision 177
0:159 0:122 0:134 0:096

S11 S12 S13 S14
0:140 0:080 0:130 0:302
S21 S22 S23 S24
0:135 0:163 0:082 0:153
S31 S32 S33 S34
0:125 0:228 0:150 0:077
M 2 ðΣÞ =
S41 S42 S43 S44
0:193 0:141 0:208 0:221
S51 S52 S53 S54
0:122 0:098 0:121 0:059
S61 S62 S63 S64
0:132 0:167 0:175 0:093
S71 S72 S73 S74
Using the two normalized matrices obtained, the gray situation decision can be
made. The specific method is as follows:
ðΣÞ
1. Normalize the matrix with columns M 2 to carry out decision-making and select
the best decision for each event, and the optimal situation will be found. In the
above example, the optimal positions of each row are S11, S24, S32, S42, S54, S61,
and S73.
ðΣÞ
2. Use the row-normalized matrix M 1 to select the best matching event of each
countermeasure and form the optimal situation. In the above example, the optimal
situation of each column is S11, S42, S24, and S73.
3. Following the above two steps and selecting the suboptimal (or satisfactory)
situation. For example, S13, S21, S34, S43, S53, S63, S24, and S72 are in the column
normalization matrix, and S61, S72, S63, and S54 are in the row normalization
matrix.
4. On the basis of the above results, global coordination is performed. That is, the
row normalization is coordinated, the row and column of the column normaliza-
tion matrix are coordinated, and the situations of “globally superior and local
nonoptimal” and “globally nonoptimal and locally superior” are checked and
adjusted. For example, in situation S51, in the column-normalized matrix, the
global is optimal (the proportion of the column is the largest), but the horizontal
comparison is nonoptimal in the row-normalized matrix, and it is also locally
optimal in the row-normalized matrix (according to the row-based normalization
matrix). However, the vertical comparison is not optimal according to the col-
umn, so the decision is lost, and S51 should be re-elected during coordination. S53
should be screened out for similar analysis.
178 X. Chen
7.4 Analysis of the Application of the Gray

Decision-Making System in Fishery Science
7.4.1 Application in the Fish Farming Industry
In the aquaculture industry, there are many factors that restrict the production of fish
farming, including feed factors and environmental factors, as well as the interspecific
relationships among the cultured species. To scientifically and rationally develop the
aquaculture industry, protect the aquatic environment, and improve the efficiency of
aquaculture, it is necessary to make scientific decisions on the factors and systems
that affect the fish aquaculture industry. Some scholars have used the gray decision
system to conduct research in this area and have achieved some results.
Xie et al. (1998) published an article entitled “The comparative study on factors
analysis and yield model of high-yield fish-pond for the Pearl River Delta and
Yangtze Delta.” In this study, we collected the relevant data of the “Comprehensive
High-yield Technology Experiment of Ten Thousand Mu of Continuous Fish Ponds
in the Pearl River Delta Region” in the Shunde area in 1983, including the net yield
X0, the stocking amount of bighead carp X1, the stocking amount of silver carp X2,
the stocking amount of grass carp X3, the stocking amount of mud carp X4, the
stocking amount of trash fish X5, the protein content of the concentrate X6, and the
protein content of forage X7.
In this study, the gray correlation decision-making method was used to analyze
the correlation degree of the factors affecting the yield, and the data were standard-
ized using the mean method with a resolution coefficient of 0.5. The results were as
follows:
The gray correlation between the stocking amount of bighead carp and the net
yield was r1 = 0.657.
The gray correlation between the stocking amount and the net yield of silver carp
was r2 = 0.474.
The gray correlation between the stocking amount of grass carp and the net yield
was r3 = 0.599.
The gray correlation degree between the stocking amount of carp and the net yield
was r4 = 0.709.
The gray correlation degree of trash fish stocking amount and net yield was
r5 = 0.489.
The gray correlation degree of concentrate protein amount and net yield was
r6 = 0.717.
The gray correlation degree of forage protein amount and net yield was
r7 = 0.762.
According to the gray correlation degree, the gray correlation sequence is as
follows: r7 > r6 > r4 > r1 > r3 > r5 > r2.
The above results indicate that the factors that have a greater impact on net yield
are the forage protein content, concentrate protein content, and stocking amount of
7 Gray Decision 179
Table 7.1 Dominance analysis matrix of high-yield ponds in the Shunde area (Xie et al. 1998)
Net yield of Net yields of Net yields of Net yields of
Content grass carp (X1) silver carp (X2) mup carp (X3) bighead carp (X4)
Fishing stocks of 1.0000 (r11) 0.8242 (r12) 0.4549 (r13) 0.2458 (r14)
grass carp (Y1)
silver carp (Y2)
mud carp (Y3)
bighead carp (Y4)
0.2209
silver carp bighead carp
0.7086
grass carp
0.4549
0.4759
mud carp
Fig. 7.1 Gray correlation degree of the polyculture fish relationship (Xie et al. 1998)
mud carp, followed by the stocking amount of bighead carp and grass carp, and the
least important factor is the stocking amount of trash fish and silver carp.
The study also used the collected data to analyze the dominant factors of the high-
yield ponds in the Shunde area. Using the stocking amounts of the four fish species
as the reference series, Y1, Y2, Y3, and Y4 represented the fishing stocks of grass carp,
silver carp, mud carp, and bighead carp per 1/5/ha, respectively. The net yields of
grass carp, silver carp, mud carp, and bighead carp per ha were denoted as X1, X2, X3,
and X4, respectively, with a resolution coefficient of 0.1. The analysis results are
shown in Table 7.1 and Fig. 7.1.
Based on the above analysis, the study concluded the following:
180 X. Chen
1. Forage protein and concentrate protein have the greatest impact on net yield.
2. The effect of the same parent factor (Y ) on different subfactors (X).
The effect of grass carp stocking (Y1) on the yield of various fish species:
r(11) = (gray correlation between Y1 and X1) = (grass carp stocking amount to net
yield of grass carp) = 1.000, r(12) = (grass carp stocking to net yield of sliver
carp) = 0.8242, r(13) = (grass carp stocking to net yield of mud carp) = 0.4549,
r(14) = (grass carp stocking to net yield of bighead carp) = 0.2458. This indicates
that the stocking amount of grass carp has a greater impact on the yield of silver
carp, which is consistent with the relationship that we usually think of as “three
silver carp in one grass belt.” At the same time, we also know that r(12) is the
largest value in the matrix, and the main basis of multispecies polyculture in
China is to use grass carp culture as the main management object. The stocking of
grass carp has little effect on the net yield of bighead carp, which can still be seen
from the perspective of the food chain relationship. That is, mud carp is a benthic
fish that feeds mainly on benthic phytoplankton, similar to the diet of silver carp,
but bighead carp is a zooplankton.
The effect of silver carp stocking (Y2) on the yield of various fish species was
as follows: r(22) = 1.0000 > r(21) = 0.5431 > r(23) = 0.4625 > r(24) = 0.2209.
This indicates that the stocking amount of silver carp has the greatest impact on
the net yield of grass carp. Silver carp mainly play a role in regulating water
quality, while grass carp require freshwater quality to be conducive to growth.
The stocking amount of silver carp was the second most important, indicating that
the feed of silver carp was close to that of common carp.
The effect of the stocking amount of mud carp (Y3) on the yield of various fish
specieswasasfollows:r(33) =1.000>r(32) =06937>r(31) =0.4759>r(34) =0.2102.
This indicates that the yield of silver carp has the largest relationship with the
stocking amount of carp, followed by the yield of grass carp, and the smallest
relationship is with the yield of bighead carp. As mentioned above, the feeding
habits of mud carp and silver carp are similar and therefore closely related.
The feeding habits of mud carp are similar to those of grass carp fingerlings.
The relationship between mud carp and bighead carp is basically irrelevant, so the
correlation coefficient is the smallest in the entire matrix.
The effect of bighead carp stocking (Y4) on the yield of various fish species
was as follows: r(44) = 1.000 > r(41) = 0.8075 > r(42) = 0.7086 > r(43) = 0.5671.
This indicates that the stocking amount of bighead carp has the largest relation-
ship with the net yield of grass carp, and it is second in the entire matrix. Due to
the presence of mud carp in the pond, the water quality is relatively fat. The
relationship between silver carp and bighead carp was second, and the relation-
ship between bighead carp and mud carp was the smallest.
3. Effect of different parent factors (Y ) on the same subfactor (X)
r(11) = (degree of gray correlation between Y1 and X1) = (grass carp stocking
to net grass carp yield) = 1.000, r(21) = (silver carp stocking to net grass carp
yield) = 0.5431, r(31) = (the stocking amount of bighead carp to the net yield of
grass carp) = 0.4759, r(41) = (the stocking amount of bighead carp to the net yield
of grass carp) = 0.8075. Therefore, r(11) > r(41) > r(21) > r(31). The results showed
7 Gray Decision 181
that the first major factor affecting the net yield of grass carp was the stocking of
grass carp, followed by the stocking of bighead carp, and the smallest was the
stocking of mud carp.
The effect of different stocking conditions on silver carp yield (X2):
r(21) = 1.000 > r(11) = 0.8242 > r(41) = 0.7086 > r(31) = 0.6937. This indicates
that the stocking of silver carp has the greatest impact on the yield of silver carp,
followed by the stocking of grass carp, and the smallest impact is the stocking of
mud carp.
The effect of different stocking conditions on the yield (X3) of mud carp was as
follows: r(31) = 1.000 > r(41) = 0.5671 > r(21) = 0.4625 > r(11) = 0.4549. This
indicates that the impact on the yield of mud carp was followed by the stocking of
mud carp and the stocking of bighead carp, and the smallest impact was the
stocking of grass carp.
The effect of different stocking conditions on the yield (X4) of bighead carp
was as follows: r(41) = 1.000 > r(11) = 0.258 > r(21) = 0.2209 > r(31) = 0.2102.
Studies have shown that stocking bighead carp has the greatest impact on the
yield of bighead carp, followed by stocking grass carp, and stocking mud carp has
the smallest impact.
4. Advantage analysis in the matrix
It can be seen from Table 7.1 that, from the perspective of the rows of the
matrix, each data point in the fourth row is greater than the corresponding data
point in the other rows, that is, the stocking amount of the bighead is the dominant
factor of the matrix.
The aquaculture in the Shunde area, which is located in the Pearl River Delta,
is different from that in other areas. Based on the characteristics of rich water and
fast growth of bighead carp in the ponds, the mixed culture of bighead carp was
the main type of pond, while silver carp was less common, which did not inhibit
bighead carp in food. The growth potential of bighead carp could be brought into
full play, and the total yield of the pond could be increased. Therefore, the
stocking of bighead carp is the dominant maternal factor. The least influence on
yield was the stocking amount of mud carp, which was a nondominant factor.
From the columns in Table 7.1, each data point in the second column is greater
than the corresponding data points in the other columns; that is, the net yield of
silver carp is the dominant subfactor of the matrix. From the above analysis, we
can see that the relationship between the yield of silver carp and the stocking of
several fish species is greater than that of other fish species. First, the stocking of
grass carp has the largest influence factor. Due to the importance of bighead carp
culture in the Pearl River Delta, the yield of silver carp was also affected because
of its close relationship with mud carp. Therefore, the net yield of silver carp was
the most affected factor, and the yield of silver carp was the dominant factor. The
net yield of bighead carp was the least influential factor, and the net yield of
bighead carp was the nondominant factor.
182 X. Chen
7.4.2 Application of Environmental Assessment of Fishery

Waters
The environment is a complex system with multiple factors and multiple levels.
Because there is some unclear gray information in the quality assessment of the
environmental system, it is difficult to establish a definitive mathematical model
based on the monitoring data of the assessment factors. Therefore, it is more
objective and reasonable to use gray system theory to evaluate the eutrophication
of fishery waters.
Zheng and Li (1999) published “a modified gray situation decision-making
method for the assessment of lake water eutrophication.” An example shows that
this model is applicable to lake water quality assessment, the method is feasible, and
the results are reasonable.
7.4.2.1 Determine the Event Set, Countermeasure Set, and Target Set
The 9 major lakes (Table 7.2) constitute the event set A = {a1, a2, . . ., a9} = {Qinghai
Lake, Taihu Lake, . . ., Erhai Lake}.
The eutrophication status of the lake water body is divided into five levels
(Table 7.3). The target set P = {total phosphorus..., biomass} was formed by the
five pollution parameters participating in the evaluation.
Table 7.2 Measured data of evaluation parameters of nine major lakes in China (Zheng and Li
1999)
Lakes Qinghai Taihu Hulun Lake Hongze Chaohu
Lake Lake
Total phosphorus/μgL-1 20 20 80 100 30
Chemical oxygen consump- 1.4 2.83 8.29 5.5 6.26
tion/mgL-1
Transparency/m 4.5 0.5 0.5 0.3 0.25
Total nitrogen/mgL-1 0.22 0.9 0.13 0.46 1.67
Biomass/10 thousand ind. 14.6 100 11.6 11.5 25.3
L-1
Lakes Dianchi Wuhan East Hangzhou Erhu
Lake West Lake
Total phosphorus/μgL-1 20 105 130 34
Chemical oxygen consump- 10.13 10.7 10.3 2.11
tion/mgL-1
Transparency/m 0.5 0.4 0.35 3.3
Total nitrogen/mgL-1 0.23 2.0 0.76 0.49
Biomass/10 thousand ind. 189.2 1913.7 6920 22.30
L-1
7 Gray Decision 183
Table 7.3 Lake water quality classification standards (Zheng and Li 1999)
Total Chemical oxygen Total Biomass
phosphorus consumption Transparency nitrogen 10 thousand
Level μgL-1 mgL-1 m mgL-1 ind.L-1
Extremely poor <1 <0.09 >37.0 <0.02 <4
nutrition
Poor nutrition 4 0.36 12.0 0.06 15
Medium 23 1.80 2.4 0.31 50
nutrition
Eutrophication 110 7.1 0.55 1.20 100
Very nutritious >660 >27.10 <0.17 >4.60 >1000
7.4.2.2 Calculate the Measure of Target Effectiveness
The whitening function is used to calculate the target effect measurement. For
example, the whitening function of target 1 (total phosphorus) on the water quality
of the first-class lake is
1 xi1 < 1
ð1Þ
γ i1 = ð4 - xi1 Þ=3 1 ≤ xi1 ≤ 4
0 xi1 > 4
For grade 2 water quality:
ðxi1 - 1Þ=3 1 ≤ xi1 ≤ 4

ð1Þ
γ i2 = ð23 - xi1 Þ=19 4 < xi1 ≤ 23
0 xi1 < 1 xi1 > 23
For grade 5 water quality:
0 xi1 < 110

ð1Þ
γ i5 = ðxi1 - 110Þ=660 100 ≤ xi1 ≤ 660
1 xi1 > 660
Similarly, the whitening function of each target on the eutrophication level of the
five lakes can be established.
By substituting the whitening value of each target (Table 7.2) into the
corresponding formula, the effect measurement of each target can be obtained, and
the effect measurement matrix is formed. For example, the effect measurement
matrix of target 1 (total phosphorus) is
184 X. Chen
0 0 0 0 0 0 0 0 0
0:158 0:158 0 0 0 0:158 0 0 0
Rð1Þ = 0:842 0:842 0:345 0:115 0:920 0:842 0:057 0 0:874
0 0 0:655 0:885 0:080 0 0:943 0:964 0:126
0 0 0 0 0 0 0 0:036 0
7.4.2.3 Determination of Target Weights
According to the whitening value of each target, the target weights are calculated
p
according to the formula ωik = ωik = ωik , where the standard value of the third
k=1
level (medium nutrition) is the reference value S0k, and the calculation results of each
target weight are listed in Table 7.4.
7.4.2.4 Calculate the Comprehensive Effect Measure
ðk Þ
According to the effectiveness of each target γ ij and weight ωij, the comprehensive
p
ðΣÞ ðΣÞ ðk Þ
effect measure γ ij is calculated according to γ ij = ωik γ ij and thus constitutes
k=1
the comprehensive effect measurement matrix as follows:
Table 7.4 Calculation results of the weight of each target (Zheng and Li 1999)
Lakes Qinghai Taihu Hulun Lake Hongze Chaohu
Lake Lake
Total phosphorus 0.192 0.115 0.389 0.470 0.121
Chemical oxygen 0.172 0.208 0.515 0.330 0.322
consumption
Transparency 0.414 0.028 0.023 0.014 0.010
Total nitrogen 0.157 0.384 0.047 0.161 0.500
Biomass 0.056 0.265 0.026 0.025 0.047
Lakes Dianchi Wuhan East Hangzhou West Erhu
Lake Lake
Total phosphorus 0.077 0.082 0.036 0.032
Chemical oxygen 0.501 0.107 0.036 0.025
consumption
Transparency 0.019 0.004 0.001 0.030
Total nitrogen 0.066 0.116 0.056 0.034
Biomass 0.337 0.691 0.871 0.878
7 Gray Decision 185
0:002 0 0:008 0:008 0 0 0 0 0

0:288 0:018 0:053 0:017 0:032 0:012 0 0 0:697
Rð Þ = 0:710 0:394 0:147 0:288 0:176 0:186 0:095 0 0:289
0 0:584 0:759 0:678 0:715 0:790 0:266 0:196 0:014
0 0:004 0:033 0:009 0:077 0:012 0:639 0:804 0
7.4.2.5 Determining the Optimal Situation

m
ðΣÞ
The formula H i = jγ ij was used to calculate the eigenvalues Hi of the grade
j=1
variables of each sample and determine the degree of eutrophication of each lake.
It is concluded that H1 = 2.70, H2 = 3.57, H3 = 3.76, H4 = 3.66, H5 = 4.24,
H6 = 3.54, H7 = 4.23, H8 = 4.32, and H9 = 2.72.
Therefore, it was determined that Qinghai Lake was mesotrophic, Taihu Lake and
Dianchi Lake were between grades 3 and 4, and they were mesotrophic.
References
(In Chinese).
(In Chinese)
Xie J, Xiao XZ, Huang ZH et al (1998) The comparative study on factors analysis and yield model
of high-yield fish-pond for the Pearl river delta and Yangtze delta. J Shanghai Fish Univ
7(2):102–106. (In Chinese)
Zheng CD, Li ZB (1999) Improved grey situation decision making method for lake eutrophication
evaluation. J Lake Sci 11(1):75–80. (In Chinese)
Chapter 8
Gray Linear Programming
Xinjun Chen
Abstract Linear programming is an important branch of operational research,

which has been studied earlier, developed faster, applied widely, and used more
mature methods. It is a mathematical theory and method to study the extreme value
problem of linear objective function under linear constraint, which is widely used in
economic analysis, management, engineering technology, industrial and agricultural
production, etc., it provides a scientific basis for making optimal decision with
limited human, material, and financial resources. Generally, under the condition of
linear constraint, the problem of finding the maximum or minimum of linear
objective function, the decision variable, constraint condition and objective function
are the three elements of linear programming. However, there are many uncertainties
in fisheries production and aquaculture, such as the impact of the global climate and
marine environment on the early life histories of fish, the impact of fishing practices
on target species and their ecosystems, and so on, these cannot be expressed by an
exact value, so there are many uncertain and fuzzy factors in the social economic
system and natural ecological environment system of fishery production, the phe-
nomenon is always gray, and the wrong results may appear when the traditional
linear programming is used to analyze and deal with the problem. The emergence of
Gray system provides a way to solve this problem. Gray linear programming is a
kind of dynamic linear programming, which can be used to solve the problems of
conventional linear programming and achieved good results. In this chapter, the gray
linear programming model and its calculation method are introduced, and the
application of gray linear programming in fishery science is analyzed by taking the
study on the adjustment of marine fishery structure in Shandong province as an
example, the development trend of fisheries in Shandong province is forecasted
scientifically, the power and composition of fishing vessels of different grades are
calculated, and the suggestions for the adjustment of marine fisheries structure in
Shandong province are put forward.
X. Chen (✉)
188 X. Chen
Keywords Gray linear programming · Fishery science · Marine fishing ·

Mariculture
Linear programming is a decision-making method that is widely used in the study of

multivariable systems, and it is particularly popular in socioeconomic disciplines
and fisheries sciences. However, because there are many uncertain and fuzzy factors
in the socioeconomic system, fisheries system, and the other natural ecological
environment systems, their phenomenon is often gray. Therefore, errors may occur
when using linear programming to analyze and deal with problems. Gray linear
programming is carried out under the condition that the technical coefficient is a
variable gray number and the constraint value is developed. It is a kind of dynamic
linear programming that compensates for the shortcomings of conventional linear
programming and has also been preliminarily applied in fishery science. This chapter
mainly introduces the gray linear programming model and the application of gray
linear programming in fishery science.
8.1 Gray Linear Programming Model
8.1.1 Standard Form of the Linear Programming Model
Linear programming is an important branch of operations research. It is a mathe-

matical model that is widely used and easy to implement in the study of multivar-
iable systems. It is also the most commonly used method for deterministic decision-
making. The main problem that it solves is how to maximize the role of limited
resources (including human resources) to achieve the maximum economic and social
benefits and to find effective ways for the rational use of human, material, and
financial resources. There are two main types of problems in the study of linear
programming: first, after a goal or task is determined, how to make overall arrange-
ments and use the least manpower, material resources and financial resources to
complete this goal; second, under certain conditions including human, material, and
financial resources, how to reasonably arrange and use human, material, and finan-
cial resources to achieve the most tasks and maximize the benefits. This is actually
two aspects of the same problem, that is, solving the overall optimal problem of the
system. Therefore, linear programming is often used as the main mathematical
method to adjust the industrial structure of various industries.
Linear programming is used to solve linear relationship problems. The so-called
linear relationship is a proportional relationship. For example, the relationship
between production and resource input and between cost and profit is generally
linear or close to linear. The following conditions are usually required to form a
linear programming problem (Chen 2003, 2023):
8 Gray Linear Programming 189
1. Determine the decision variables of the problem. This refers to the factors that the
decision maker can control, and their values determine the solution of the model.
2. There must be clear goals. It is required that the objective of the question can be
expressed by a numerical value, that is, the relevant question is converted into a
formula, and the criteria used by the decision maker to evaluate different answers
to the question, namely the objective function, are determined.
3. The goal to be achieved is realized under certain constraints, and there are
multiple feasible schemes to achieve the goal.
4. To clarify the limited number of limited resources, the input–output relationship
and the output-benefit relationship of each production sector are used to deter-
mine the reasonable coefficients of the decision-making variables.
5. Both the constraint condition and the objective function must have a linear
relationship. The constraint conditions reflect the limitations of the system envi-
ronment, and the objective function reflects the goals of the decision makers.
Therefore, the general linear programming model includes five parts:
1. decision variable Xj ( j = 1, 2, . . ., n);
2. constraint or resource constraint bi (i = 1, 2, . . ., n);
3. technical coefficient aij;
4. benefit coefficient cj;
5. objective function Z.
The mathematical model of linear programming is
Objective function max or min Z = c1 x1 + c2 x2 + . . .. . . + cn xn
Satisfied with the constraints:
a11 x1 þ a12 x2 þ ⋯ þ a1n xn = b1

a21 x1 þ a22 x2 þ ⋯ þ a2n xn = b2
...
am1 x1 þ am2 x2 þ ⋯ þ amn xn = bm
x1, x2, . . . , n ≥ 0
Its abbreviation is
n
Objective function max or min Z = cj xj
j=1
n
Satisfy the constraints aij xj = bi (I = 1, 2, . . ., m)
j=1
xj ≥ 0 ðj= 1, 2, . . . , nÞ
where xj is a set of unknown decision variables representing the output of various

products; aij is the technical coefficient, which represents the input quantity of i types
of production factors required to produce j types of products; cj is the efficiency
190 X. Chen
coefficient, which represents the income of the production unit j types of products; bi
is the restricted quantities of production factors.
The linear programming problem with the above structure is called the standard
form. The specific linear programming model may have many limitations and
constraints, but any linear programming problem can be transformed into the
above standard form.
8.1.2 Gray Linear Programming
Although linear programming has been widely used in fields of social and economic
development, fisheries science, etc., general linear programming has the following
problems (Chen 2003, 2023):
1. Linear programming is static and cannot reflect the change in constraint condi-
tions over time. Therefore, the obtained results often fail due to changes in
conditions.
2. If there are gray parameters (or gray numbers) in the planning model, such as the
technical coefficients and constraint values in the constraint equations, it is
difficult to address general linear programming.
3. Due to the problem of model technology or computational skills, there is often no
solution or unsolvable problem in the actual calculation process.
Due to the above problems, the application of general linear programming is
limited to a certain extent. However, these problems can be solved using the idea and
modeling method of the gray system. Linear programming combined with gray
system theory is called gray linear programming.
The form of gray linear programming is as follows:
Objective function:
Constraints: (A)X ≤ b X ≥ 0
In other words, satisfying (A)X ≤ b under the condition of X ≥ 0, a set of X is
sought to make f(X) reach the maximum value (or minimum value).
In the above relation, X is a vector:
X = ½x1 , x2 , ⋯, xn T
C is the coefficient vector of the objective function
C = ½c1 , c2 , ⋯, cn
where Ci can be a gray number, (A) is the coefficient matrix of the constraint
condition, and A is the whitening matrix of (A) and has
11 12 ... 1n

ðAÞ = ⋮ ⋮ ... ⋮
m1 m2 ... mn
a11 a12 . . . a1m
A= ⋮ ⋮ ... ⋮
am1 am2 . . . amn
b is the constraint quantity b = [b1, b2, ⋯, bm]T.

If there is a set of whitening sequences for the constraint index bi, there is a set of
whitening sequences.
ðoÞ ðoÞ ðoÞ ðoÞ

bi = bi ð1Þ, bi ð2Þ, ⋯, bi ðN Þ
ðoÞ ð1Þ ð1Þ

Then, after bi is accumulated to obtain bi , bi is used to set up the prediction
model GM (1, 1), and then the prediction value is obtained from the prediction
model.
ð0Þ
bi ðK Þ, Kin
When making the planning calculation, the following constraint conditions are
applied.
ð0Þ
b1 ð K Þ
ð0Þ
b2 ð K Þ
ðAÞ X =
⋮
bðm0Þ ðK Þ
Then, the gray linear programming value at time K can be obtained. When K > n
is set to different values, various linear programming solutions for future develop-
ment can be obtained, that is, linear programming solutions for different periods.
Gray linear programming has the following characteristics (Chen 2003, 2023):
1. It makes up for the shortcomings of general linear programming. Conventional
linear programming is a deterministic and static model that requires that the
benefit coefficient in the target coefficient, the technical coefficient in the con-
straint condition, the amount of resources, and other restrictions be fixed. In fact,
the socioeconomic relationship is uncertain and changeable, and there are many
accidental and risky factors. In practice, there is no solution. Gray linear pro-
gramming is carried out under the condition that the technical coefficients are
variable gray numbers and the constraint values are developed. It is a dynamic
192 X. Chen
linear programming that compensates for the shortcomings of conventional linear

programming.
2. It can not only guide the optimal structure under given conditions but also guide
the development and change of the optimal structure. The constraint values in the
constraint conditions may be variable, and some can be described by time series.
The GM (1,1) model is used for prediction. Such linear programming not only
reflects a specific situation but can also reflect the development and change of
constraints. Such a linear programming solution is not a value but a group of
values and is a group of time series values. Such a solution can not only guide the
optimal structure under the present condition but also provide information on the
development of the optimal structure relationship.
3. Given a set of information, a set of optimization schemes can be obtained. The
constraint condition coefficient in gray linear programming is the number of gray
intervals, which can be planned according to the lower bound, the upper bound,
or any whitening value in the interval. In the interval, as long as a set of whitening
values (information) can be obtained, a set of optimization schemes can be
obtained so that the planning is flexible and has much room for adjustment,
adapting to the development and changes of the situation and avoiding the fact
that the conventional linear programming makes a lot of specificity. No feasible
solution can be obtained for the problem.
8.2 Application of Gray Linear Programming in Fishery

Science
At present, the application of gray linear programming in fishery science is mainly in

the aspects of marine fishery structure characteristics (including aquaculture, fishing
vessel structure, etc.), industrial structure adjustment, and development planning.
Gao et al. (1999) published the “Study on the structural adjustment of marine
fisheries in Shandong province,” which studied the application of the gray linear
programming optimization method to explore the adjustment of the fishing vessel
structure to achieve the optimization of the output value.
8.2.1 Analysis of the Structure of Different Fishing Vessels
1. Target determination: This study selects the indicator reflecting the economic
benefits, maximum profit, as the objective function and calculates the suitable
marine fishing effort of each level of fishing vessel with the maximum economic
benefit within the predicted range.
2. Variable setting: The fishing effort of each level of fishing vessel is selected as the
decision variable.
3. Constraint conditions: Among the many available constraint variables, total

control and sublevel ship power control are used as dual constraint parameters.
4. Selection of coefficients: The benefit coefficient used in this study refers to the
coefficient of the decision variable in the objective function, which refers to the
profit per 1000 PS of fishing vessels.
5. Establishment of the linear programming model. In 2000, the linear programming
model of marine fishing vessels in Shandong Province was as follows (Gao et al.
1999):
Y max = 2497X 1 þ 3256X 2 þ 2818X 3 þ 1540X 4 þ 2062X 5 þ 109X 6 þ 3X 7

þ 398 X 8
X 1 þ X 2 þ X 3 þ X 4 þ X 5 þ X 6 þ X 7 þ X 8 ≤ 100
X 5 þ X 6 þ X 7 þ X 8 ≤ 400
X 1 ≤ 10, X 2 = 20 - 25, X 3 = 5 - 10, X 4 ≤ 20
X1, X2, X3, X4, X5, X6, X7, X8 ≥ 0
where X1, X2, X3, X4, X5, X6, X7, and X8 are below 19 PS, 20 PS, 21–59 PS,
60–119 PS, 120–199 PS, 200–399 PS, 400–599 PS and above 600 PS fishing
boat horsepower, respectively.
6. Calculation results: Since the planned value is an approximate number, the
calculation results are rounded to the nearest whole number (all units are 104 PS).
X 1 = 10
X 2 = 25
X3 = 5
X 4 = 20
X 5 = 10
X 6 = 10
X7 = 5
X 8 = 15
Under the structure of fishing vessels at all levels under the condition of control-
ling the total fishing effort at 100 × 104 PS, the calculation results of the above
optimization model show that the proportion of fishing vessels below 19 PS should
be reduced, and the proportion of fishing vessels at 21–59 PS and 120–199 PS
should be also reduced, the fishing vessels at 20 PS should be kept stable, and the
fishing vessels at 60–119 PS, 200–399 PS, and above 600 PS should be developed
(Gao et al. 1999).
194 X. Chen
8.2.2 The Planning of the Number of Fishing Vessels

and Their Efforts in Marine Fishing Operations
in Shandong Province
Based on the analysis of the types of fishing vessels and the allocation of efforts in
Shandong Province, the following adjustments are proposed (Gao et al. 1999):
1. Under the condition that the total marine fishing effort is controlled to be less than
100 × 104 PS, considering the carrying capacity of fishery resources and the
existing fishery productivity, the fishing effort structure with a 4:3:3 ratio is
proposed, i.e., trawling boat (including purse seine) occupies 40% of the total,
gill-net boat (including jigging boat) accounts for 30% of the total, and stake net
boat (including other fishing boats) occupies 30% of the total.
2. Effort allocation for different types of operations (Table 8.1). The total number of
fishing vessels is controlled at approximately 26,680, which is a significant
decrease from the current 35,417. At the same time, the fishing effort and the
structure of the types of fishing vessels have been significantly improved.
8.2.3 Linear Programming Model for the Marine

Aquaculture Industry
The objective function is selected to reflect the economic benefit index, which is the
maximum net income, to obtain the suitable aquaculture area for each industry. The
aquaculture area was selected as the decision variable. The aquaculture areas of fish
farming, shrimp and crab farming, algae farming, shallow sea shellfish farming, and
tidal flat farming were used as the decision variables X1, X2, X3, X4, and X5. The
constraint condition is the aquaculture area (Table 8.2). The benefit coefficient refers
Table 8.1 Effort allocation of various types of fishing boats in Shandong Province (Gao et al.
1999)
Total power (104 KW) and the number of fishing
Fishing type boats
Trawler Level Above 441 KW 184–294 KW 136–147 KW
Power (104 KW) 11 11 7.35
Number of ships (boats) 250 500 500
Drift jigging boat Level Above 44.1 KW 29.4–44 KW 15 KW
Power (104 KW) 11 3.68 7.35
Stake net Level 15.4–44 KW 15 KW 8.8 KW
Power (104 KW) 3.68 11 7.35
Table 8.2 The constraint conditions unit: 104 mu (Gao et al. 1999)
Shallow
Shrimp sea Mudflat Shallow
Fish and crab Algae shellfish shellfish sea Mudflat Harbor
Year farming farming farming farming farming farming farming farming
1994 1.32 65.98 17.66 73.67 37.66 92.40 52.82 52.13
2000 2.57 54.01 38.41 129.01 89.80 149.71 163.42 54.01
to the coefficient of each decision variable in the objective function. This study takes
the net income per unit area by industry in 1994 as the reference benefit coefficient.
The linear programming model for the year 2000 is constructed as follows (Gao
et al. 1999):
Z max = 1218X 1 þ 179X 2 þ 1243X 3 þ 1341X 4 þ 1361X 5

X 1 þ X 2 þ X 3 þ X 4 þ X 5 ≤ 367:14
X 1 þ X 2 þ X 5 ≤ 163:42 þ 54:01
X 3 þ X 4 ≤ 149:71
X 1 ≥ 2:57, X 2 ≥ 54:01, X 3 ≤ 38:41, X 4 ≤ 129:01, X 5 ≤ 89:80
Obtained by calculation
X 1 = 2:57, X 2 = 54:01, X 3 = 38:41, X 4 = 111:3, X 5 = 50:26
Total suitable farming area: X1 + X2 + X3 + X4 + X5 = 256.55 (104 mu)

Through the analysis of the average yield of each subindustry from 1985 to 1994,
the average yield of each subindustry in 2000 was calculated. According to the
above linear programming model, the suitable aquaculture area value of each
industry under the optimal benefit is obtained, and the expected output of the marine
aquaculture industry in Shandong Province in 2000 is obtained:
1. The expected yield of mariculture in 2000 based on the subindustry average yield
0:55X 1 þ 0:03X 2 þ 2:20X 3 þ 1:14X 4 þ 0:58X 5

= 1:42 þ 1:64 þ 84:46 þ 126:93 þ 29:17
= 243:62 104 t
2. The expected yield of mariculture in 2000 based on the total average yield of the
mariculture industry
0:87 × 256:55 = 223:20 104 t

On the basis of a comprehensive investigation, study, and analysis of the current
situation of marine fisheries in Shandong Province, this study used gray model
196 X. Chen
theory to simulate and predict the characteristics and trends of fishery development
in Shandong Province and used the theory and method of gray linear programming
to calculate various grades of fishery development. The optimal estimation of the
power of the fishing vessel and the optimal structure of fishing boats are calculated.
Based on the evaluation results of this model, some suggestions for adjusting the
structure of marine fisheries in Shandong Province are proposed (Gao et al. 1999).
References
(In Chinese)
(In Chinese)
Gao QL, Qiu TX, Song XF et al (1999) The study on the structure regulation of marine fishery of
Shandong Province[J]. J Ocean Univ Qingdao 29(2):47–55. (In Chinese)

Xinjun Chen - Application of Gray System Theory in Fishery Science-Springer-CAP (2023)

Uploaded by

Copyright:

Available Formats

Xinjun Chen - Application of Gray System Theory in Fishery Science-Springer-CAP (2023)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Xinjun Chen - Application of Gray System Theory in Fishery Science-Springer-CAP (2023)

Uploaded by

Copyright:

Available Formats

Xinjun Chen Editor

Application of Gray System

ISBN 978-981-99-0634-5 ISBN 978-981-99-0635-2 (eBook)

© China Agriculture Press 2023

Lingang New City, China Xinjun Chen

1 Overview of Gray System Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Keywords Gray system theory · Gray correlation · Gray clustering · Gray

1.1 Basic Concepts of the Gray System

In people’s social economic activities, they often encounter incomplete or vague

such as ﬁshery resources, marine environment, ﬁshermen’s technical level, sea

1.1.1 Gray Meaning and Gray Phenomenon

1.1.2 Gray System

1.2 Overview of the Development of Gray System Theory

1.2.1 The Scientiﬁc Background of Gray System Theory

1.2.2 Overview of the Development of Gray System Theory

information” uncertain systems with “partially known information” and “poor

1.3 Main Content of the Gray System Theory

uncertainty” phenomena with multiple possible outcomes. The starting point is a

1.3.1 Gray Relational Analysis

1.3.2 Gray Modeling

The mathematical model quantitatively expresses the mathematical relationship

1.3.3 Gray Prediction

1.3.4 Gray Decision-Making

The so-called decision-making refers to the selection of an appropriate countermea-

Gray decision-making includes gray situation decision-making, gray-level

1.3.5 Gray Linear Programming

Linear programming is an important branch of operations research. It is a

1.3.6 Gray Control

1.4 The Status and Characteristics of Gray System Theory

1.4.1 The Position of Gray System Theory in the Discipline

1.4.2 The Role of Gray System Theory in the Discipline

1.5 Research Progress on the Application of Gray System

1.5.1 Fishery Economy Industry

1.5.2 Aquaculture Industry

in the aquaculture industry mainly includes aquaculture yield prediction, aquaculture

1.5.3 Environmental Assessment of Fishery Waters

1.5.4 Fisheries Forecasting

Fishery situation forecasting is an important branch of the ﬁeld of ﬁshery science. It

Keywords Original data · Data transformation method · Case analysis

2.1 Sources and Characteristics of the Original Data

2.2 Several Methods of Raw Data Whitening and Initial

2.2.1 Collection and Whitening of Raw Data

For the vast majority of gray parameters, whitening or lightening is needed to

4. Prediction of economic indicators. The classiﬁcation and zoning of the marine

2.2.2 Several Commonly Used Data Transformation Methods

2.2.2.1 Standardized Transformation

The calculation formula is

(i = 1, 2, . . ., the number of N samples; j = 1, 2, . . ., the number of P variables)

X 1 = ðX 1978 þ X 1979 þ . . . þ X 1984 Þ=7 = ð2:64 þ 2:72 þ . . . þ 2:54Þ=7 = 2:66

Then, the data are transformed into:

Other data transformations are similar.

The sequences of the transformed resource and environment subsystems are

2.2.2.2 Range Transformation

The calculation formula is

X1max = 2.73 X1min = 2.54

Then, the data are transformed into:

X 11 - X 1 min 2:64 - 2:54

The rest of the data conversion is similar.

2.2.2.3 Averaging Transformation

The calculation formula is