Water Resour Manage (2014) 28:1937–1958

DOI 10.1007/s11269-014-0586-2

Simulation of Multiple Hydropower Reservoir Operations

Using System Dynamics Approach

Ramesh S. V. Teegavarapu & Slobodan P. Simonovic

Received: 23 February 2013 / Accepted: 12 March 2014 /

Published online: 1 April 2014
# Springer Science+Business Media Dordrecht 2014

Abstract A simulation model is developed to understand the dynamic behavior of a hydrau-

lically coupled multiple reservoir system when operated in real-time. An object-oriented
simulation environment conceived on the principles of system dynamics (SD) is used
for the development of the model. The modeling process consists of developing causal
loop, stock and flow diagrams, and carrying out simulations using difference equa-
tions to integrate stocks. Unlike traditional simulation environments, modeling using
SD principles will aid in the analysis of the system by a series of interconnecting
processes and functional relationships. Also, the feedback loops that influence variable
values and causal relationships in space and time can be clearly understood using SD-
based simulation. Performance measures are developed to quantify the system perfor-
mance in making decisions regarding actual implementation of operating rules. A
simulation model based on a real-life hydropower reservoir system in the Province
of Manitoba, Canada is developed. Dynamics of the multiple reservoir operation
suggest that in a serial system of reservoirs, the operational decision taken at one
reservoir will have effect on the other reservoir. The behavior of the reservoir system
for extreme events is also observed and conclusions relevant to real-time operation of
the systems are drawn. The simulation models developed in this study provide global
description of a hydraulically coupled reservoir system with ability to easily model
dynamic processes to obtain operational release decisions that can be adopted in
real-time.

Keywords System dynamics simulation . Hydropower . Hydraulic coupling . Real-time

operation

R. S. V. Teegavarapu (*)
Department of Civil, Environmental and Geomatics Engineering, Hydrosystems Research Laboratory
(HRL), Florida Atlantic University, 777 Glades Road, Bldg # 36, Room 217, Boca Raton, FL, USA
e-mail: [email protected]

S. P. Simonovic
Department of Civil and Environmental Engineering, University of Western Ontario,
London, ON, Canada N6A 5B9
e-mail: [email protected]
1938 R.S.V. Teegavarapu, S.P. Simonovic

1 Introduction

The field of water resources planning and management is replete with problems where
traditional methods require complex mathematical representation of the physical systems at
the same time being excessively abstract. The non-linearities inherent in hydrological process-
es and socio-economic aspects associated with water resources systems some times make them
less amenable to traditional modeling approaches. The complexity of the models is a driving
force for the research community in the field of water resources to switch from reductionist
approaches to holistic approaches in order to understand the dynamic behavior of these
systems. System dynamics simulation (Forrester 1961) is one such approach that is appropriate
in this context. The central building blocks of system dynamics paradigm—levels and rates (or
stocks and flows) can be effectively used to model many elements that govern the physical
description and behaviors of water resource, environmental and ecological systems. Modeling
the operation of reservoir systems is generally achieved by developing mathematical program-
ming formulations for optimizing the operations or by using simulation models to understand
and characterize the dynamics of reservoir operation. The use of simulation models is not
uncommon in the field of reservoir operation. Simulation models are often used in real-time to
refine the rules developed by optimization models. An earlier detailed review of traditional
simulation models used for reservoir operation is given by Wurbs (1993). Traditional simula-
tion models developed using high level programming languages do not provide insights into
the dynamics of the system. Their structure is rigid and the interaction between different
components within the models is not transparent.
It is generally known that engineering systems are predictable and the dynamics of
managed (engineered) systems is well understood. Counter-intuitive behavior that is common
to natural, social and economic systems is not usually observed in engineering systems.
Reservoir system operations are no exception to this. However, a counter-intuitive behavior
is observed if the dynamics of system are influenced by the social and economic components.
The human element involved in the management of reservoirs governed by judgment,
experience and rules of thumb can at times lead to unpredictable or counter-intuitive behavior.
This is true in case of short-term or real-time operation of reservoir systems where the reservoir
operators make changes to rules provided by optimization models. Random events that
influence the system also introduce certain element of uncertainty that cannot be contemplated.
Very often the operation of multiple reservoir systems in real-time is influenced by the
operator’s judgment.
Behavior of systems that are influenced by a number of processes and factors can be
understood by developing models using system dynamics (SD) principles. Some of the most
recent real-life case studies provide applicability of various contemporary modelling tech-
niques in water resources engineering. These techniques that are used for prediction employ
artificial neural networks (ANNs) and their variants (Chau et al. 2005; Muttil and Chau 2006;
Cheng et al. 2005; Wu et al. 2009). A more general and easy performance evaluation of water
resource systems can be achieved by developing models that use object-oriented simulation
(OOS) environments (Wurbs 1993). Recent applications of OOS to water resource planning
and policy analysis (Lund and Ferreira 1996; Simonovic et al. 1997; and Fletcher 1998;
Simonovic and Fahmy 1999) are benefited by adopting this approach. Keyes and Palmer
(1993) indicate the advantages of such an approach to problems in water resources by
demonstrating its utility in drought planning policy scenario generation. The principles of
system dynamics are well suited for modeling and application to water resources problems, a
fact that is confirmed by the previous works. Simonovic and Fahmy (1999) provide a general
approach for policy analysis that uses the principles of system dynamics. Several studies (Ford
Simulation of Multiple Hydropower Reservoir Operations 1939

1999; Deaton and Winebrake 1999; Moffatt 1991; Nirmalakhandan 2002; Coyle 1996)
indicate that the principles of system dynamics are well suited for modeling and application
to water resources and environmental problems.
System behavior in space as well as time can also be simulated using system dynamics
framework (Hugget 1993). Huang and Chang (2003) in a survey of tools for environmental
systems analysis indicate the potential of using system dynamics for improved understanding
of the environmental systems. The field of water resources has definitely benefited by system
dynamics approach (Simonovic 2009). Simonovic et al. (1997) apply system dynamics
simulation to the operation of High Aswan Dam in Egypt. The advantages of using a
simulation environment over developing a traditional simulation model are clearly evident
from their study. Lund and Ferreira (1996) use SD approach to develop a rule based reservoir
operation model. They compare the performance of this model with that of HEC-PRM
(Hydrologic Engineering Center–Prescriptive Reservoir Model). A comprehensive treatise
on system dynamics approach applications to water resources management is provided by
Simonovic (2009).
Ahmad and Simonovic (2000) used SD approach to a reservoir operation problem for flood
control. The impacts on flood management capacity of the reservoir are investigated by
simulating a gated spillway in addition to an existing un-regulated spillway. They indicate
the use of SD is more attractive than other techniques because of ease of modification of
system structure and ability to perform sensitivity analyses. Saysel et al. (2002) evaluated
water resources development options for regional agricultural projects using system dynamics
simulation. Their emphasis was on impact assessment on social and natural environments.
Components affecting water demand and supply in Yellow River basin in China were analyzed
by Xu et al. (2002) using SD approach. Public understanding of value of water conservation
was simulated using SD for a water management study in Las Vegas, USA, by Stave (2003).
Several studies related to seawater intrusion (Fernandez and Selma 2004), global water
assessment (Simonovic 2002), regional water resource assessment (Simonovic and
Rajasekaram 2004), flood evacuation strategies (Simonovic and Ahmad 2005) and water
conflict resolution (Nandalal and Simonovic 2003) and hydrologic systems (Sehlke and
Jacobson 2005) highlighted the advantages of using SD approach over traditional simulation
methods. Simultaneous consideration of dynamic interactions between quantitative character-
istics of water resources and water use constrained by socio-economic, population and
physiographic heterogeneity were documented in few SD models reviewed (e.g., Simonovic
and Rajasekaram 2004; Simonovic 2002) from available literature. SD approaches were also
used for watershed modeling and management for nutrient and pathogen impairment of
streams by Teegavarapu et al. (2005) and Elshorbagy et al. (2005, 2006). Teegavarapu et al.
(2005) simulated the dynamics of a nutrient impaired stream and evaluated source load
reductions using a total maximum daily load (TMDL) approach. The water quality simulation
model was developed, calibrated and validated using SD approach.
The main objective of the current study is to develop a simulation model using system
dynamics approach (SD) to understand the behavior of the reservoir system in a variety of
situations. The model developed in the current research study is different in comparison with
the earlier works (Simonovic et al. 1997; Simonovic and Fahmy 1999): (i) in terms of
addressing the issue of real-time operation of multiple reservoir system considering hydrologic
and hydraulic coupling; (ii) exhaustive modeling of the system and the use of two indices
(Hashimoto et al. 1982) to quantify the system performance. The study also concentrates on
understanding of the behavior of the system rather than pure simulation of the processes
governing the system. The motivation to develop simulation model is to understand: (i) the
behavior of the system for short-term changes in the inputs and (ii) the heuristic operating
1940 R.S.V. Teegavarapu, S.P. Simonovic

procedures or rules of thumb and their effect on the overall system performance. An exhaustive
literature review reveals that issues of hydraulic and hydrologic coupling were not addressed in
a simulation model meant for real-time operation of hydropower systems in the past.
The model in the current study is developed using an object-oriented simulation environ-
ment, STELLA (Structural Thinking, Experiential Learning Laboratory with Animation
(STELLA 2009). The environment provides several generic building blocks through which
specific components of water and environmental resource systems can be modeled. These
basic building blocks referred to as stock, flow, converter and connector are shown in Fig. 1.
The generic properties of these objects can be used for modeling a variety of dynamic systems.
Table 1 provides a list of processes that influence operation of hydropower systems, and their
equivalent modeling objects in STELLA environment. In literature there are no operation
models (especially simulation models) available for real-time operation of coupled hydropow-
er systems. This simulation model developed using object-oriented simulation environment
can help plant operators to simulate conditions that occur at finer time intervals that are less
than 1 h and understand the dynamics of power generation when water availability, hydraulic
coupling influencing energy generation head and operation constraints in maintaining a
specific level (stage) of water in a any given reservoir within a system of reservoirs.

2 Modeling Reservoir Systems

The reservoir system considered in the present study is a series of four hydropower reservoirs,
in Manitoba, Canada. However, a schematic representation of a two-reservoir system is
provided in Fig. 2 that is representative of the multiple reservoir system handled in the current
study. The notation used will be explained later when the governing equations are discussed.
The reservoirs are linked both hydrologically and hydraulically. The former link indicates that
the release from an upstream reservoir becomes an input to the immediate downstream
reservoir while the latter link suggests that any decision taken at one reservoir has an effect
on the other in way that influences the performance of both reservoirs. This linking is referred
to as hydraulic coupling, which is mainly due to hydraulic characteristics of the system. The
link and the nature of system can be better explained using a causal loop diagram. Details of
such diagram are discussed in the next Section.

Fig. 1 Basic building blocks of the object-oriented simulation environment, STELLA

Simulation of Multiple Hydropower Reservoir Operations 1941

Table 1 Processes and physical elements that affect hydropower reservoir operations and corresponding
modeling objects in STELLA simulation environment

Processes and physical elements Modeling

objects
in STELLA

Lakes, Stream storage Stocks

Power generation
Inflows (stream flows) Flows
Mathematical Relationships (e.g. efficiency-discharge, stage-storage, forebay-tailwater elevation Converters
curves), preference functions for performance measures
System boundaries: reaches, outlets, lateral flows Sources
and sinks
Transfer of relationships and links Connectors

2.1 Causal Loop Diagram

Influence or causal loop diagrams representing the inter-relationships between various ele-
ments can help understand the behavior of system. Standard notation following Roberts et al.
(1983) is used for the development of the causal loop diagram. A causal loop diagram of a
hydraulically and hydrologically coupled two-reservoir system is shown in Fig. 3. The
relationships between different elements of the system and the effect one element has on the
other is evident from the Fig. 3. The arrows indicate the relationships between different
elements and the sign (+ or −) attached at the front of the arrow suggests the type of effect
one element has on the other. For example, an increase in inflow would lead to an increase in
storage. Similarly an increase in tailwater elevation would lead to a decrease in the head
required for power generation. The term head is defined as the difference in the level between
the forebay and tailwater elevations of the reservoir. Basic description of hydropower gener-
ation process can be found elsewhere (Wood and Wollenberg 1984). The spills, releases,
reservoir levels and heads as shown in Fig. 3 are defined for different reservoirs using different

j=1

j=2

Fig. 2 Schematic diagram of hydraulically coupled hydropower reservoirs

1942 R.S.V. Teegavarapu, S.P. Simonovic

Inflow1

Controlled Inflow2
Flow0
+
+ +
Reservoir
Storage1 Reservoir -
+ - + Storage2
Reservoir +
Level1 +
Release1/Spill1 + Reservoir
Local Level2
Inflow2 +
+
Head1
- Release2/Spill2

- + + +
Tailwater -
Elevation1
Fig. 3 Causal loop diagram of processes for a hydraulically coupled two-reservoir hydropower system

superscripts (e.g., 1 and 2). A similar diagram can be used to explain the behavior of the next
two reservoirs in a cascade.
Few observations can be made based on the causal loop diagram. Shocks (sudden
changes) can be introduced by the exogenous inputs. These inputs are the inflows
(streamflows) that are both controlled and natural. While the first input is known a
priori, the second one is stochastic. Feedbacks are evident in the system and they are
balancing or negative feedbacks. The balancing is due to controlled (release decisions)
and uncontrolled (spill) actions in the system. For example, a rise in the reservoir level
due to higher inflow has an effect on the release made. However, as the release increases
the tailwater elevation increases and thus reducing the head required for energy gener-
ation. If the release is restricted, the uncontrolled spill will produce the same effect.
Positive feedback behaviors are uncommon in engineering systems where human inter-
vention mechanisms always control the dynamics of the system.

2.2 Model Structure

The structure of the model can be developed as a set of mathematical expressions and
computer coded to form a simulation model. The expressions that govern the physical system
can be incorporated via flow diagrams and finally transformed into object structure using any
simulation environment (e.g. STELLA; ISEES 2009). The object structure here refers to series
of objects (stocks and flows) with specific properties that are used to describe the model of a
real system. The model developed is for a four reservoir system in the Province of Manitoba,
Canada. A schematic of this system of reservoirs and case study area is shown in Fig. 4.
Different components of the model developed in this study are specific to this system.
However, the representation is generic and is easily transferable to any other system of similar
configuration. A brief description of the system processes that need to be represented is
provided here.
Simulation of Multiple Hydropower Reservoir Operations 1943

(a)

(b)

Fig. 4 a Details of the four hydropower reservoir system (Province of Manitoba and location of hydropower
generating stations); b schematic representation of reservoirs in the case study region (Data Source for map:
http://mli2.gov.mb.ca)

2.2.1 Hydraulic Coupling

It is evident from the causal loop diagram (Fig. 3) that the tail-water elevation of an upstream
reservoir is effected by the reservoir stage elevation of the downstream reservoir. Hydraulic
coupling is a major factor that should be considered in developing operation rules. In case of
isolated reservoirs (reservoirs separated by large distances), the hydraulic linking (coupling) is
generally absent and need not be considered while developing models.

2.2.2 Tailwater Elevations

If the hydraulic linking is considered to be strong, then the dynamic variation of the tail-water
elevation needs to be considered in the modeling process. This is possible by the use of tail-
water elevation curves and incorporating them in the model to calculate the head required for
energy generation.
1944 R.S.V. Teegavarapu, S.P. Simonovic

2.2.3 Local Inflow

Local inflows that join the stream in between any two reservoirs should be considered as
inputs to the system as they contribute to the mass balance. Local inflows may influence the
tailwater elevations, if they join the stream close to the tailwater pool of the reservoir.

2.2.4 Boundaries of the System

The reservoir system considered in the current study is a set of four reservoirs. The first
reservoir receives a controlled flow from the upstream reservoir that can be regarded as a
source. The last reservoir in the system is discharging into a lake (referred to as sink in
modeling environment) that is considered to be the boundary of the system. The local inflows
into the reservoirs are also considered to be boundaries of the system.

2.2.5 Governing Equations

The operations of the system can be expressed as a set of equations that can help capture the
dynamics of the operation process. The object structure (with stocks, flows, converters and
connectors) can be used to model these equations using any simulation environment (e.g.
STELLA). Variation of elevation within any time interval is difficult to model. Hence a
relationship to obtain average elevation is used.
 
h j;t ¼ k ij;t þ k fj;t 0:5 j ¼ 1; n &t ¼ 1; T ð1Þ

The above equation is used for obtaining the average forebay elevation at each of the
generating stations. An average value is used as the forebay elevation fluctuates within any
time interval. Here kij,t and kfj,tare the initial and final forebay elevations for the time interval, t,
associated with the station, j, respectively.
 
h jþ1;t ¼ k ijþ1;t þ k fjþ1;t 0:5 j ¼ 1; n &t ¼ 1; T ð2Þ

The value of hj+1,t in Eq. 2 refers to average forebay elevation of reservoir, j+1 and this is
to select tailwater elevation curve. For any generating station, tailwater elevation curves
represent discharge-elevation curves for different downstream forebay elevations. A schematic
diagram showing these elevation curves is provided in Fig. 5. A general form of these curves is
given by the Eq. 3.

T j;t ¼ k l;o jþ1 þ C l; jþ1 Gtj ∀ j; ∀l; t ¼ 1; T ð3Þ

Tj,t is the tailwater elevation, kol,j+1 is a discrete downstream reservoir’s forebay elevation, which is
taken as downstream condition for deriving the curves. Curves are developed for different discrete
o
values (i.e. l) of kl,j+1 . A specific value of kol,j+1 and tailwater curve is selected using the value of hj+
t
1,t. The variable Gj represents the sum of outflow and spill from plant j, while the variable Cl,,j+1 is a
constant in the linear expression. If the local inflow influences the tail water elevation, it has to
be included in calculating the total discharge (or project discharge), Gtj. A set of
tailwater elevation curves are used for different forebay elevations.
Simulation of Multiple Hydropower Reservoir Operations 1945

Fig. 5 Representation of storage intervals and tailwater elevation curves

 
β j;t ¼ f Q j;t ∀j; t ¼ 1; T ð4Þ

Equation 4 defines the relationship between efficiency, βj,t, and the outflow, Qj,t. This
constraint specifies the functional relationship between the outflow, Qj,t and overall plant
efficiency, βj,t for generating station, j, for time interval t. Efficiency, in general is also a
function of head. In simulation a non-linear functional relationship can be easily used by the
inclusion of relationship which relates efficiency with both discharge and head poses no
conceptual difficulty. However, the relationship (Eq. 4) is used to facilitate the comparison
of the present model with the already existing weekly scheduling model, EMMA (Energy
Management and Maintenance Analysis) model (Barrit-Flatt and Cormie 1988) that uses
similar relationship.
The relationship to calculate power generated at any hydropower generating reservoir is
given by Eq. 5 as:


ε j;t ¼ γ o h j;t − T j;t Q j;t β j;t ∀ j; ∀t ð5Þ

where γo is a constant and the term (hj,t −Tj,t) provides the head for power, εj,t, in any time
interval, at any reservoir, j, calculations. The mass balances for reservoirs are given by given
Eqs. 6–8.

S j;tþ1 ¼ S j;t þ I j;t þ R j−1;t − R j;t ∀ j; ∀t ð6Þ

R j;t ¼ SP j;t þ Q j;t ∀ j; ∀t ð7Þ

S tþ1
j ¼ S tj þ I j;t þ R j−1;t − O j;t j ¼ n þ 1; ∀t ð8Þ
1946 R.S.V. Teegavarapu, S.P. Simonovic

The variables, Sj,t and Sj,t+1 represent reservoir storages at the beginning of time intervals t
and t+1 in volume units respectively. The variables Rj,t and SPj,t are the release and the spill
values from reservoir j respectively, while Ij,t is the forecasted value of local inflow. The Eq. (8)
represents the continuity equation for the lake into which the last plant is discharging. The
variable, Oj,t, represents the controlled flow out of the lake. All the reservoirs considered are
run-off-the river plants with very little storage and strong hydraulic coupling. Also, the
simulation model developed in this study is mainly aimed at real-time or short-term operation
and therefore evaporation is neglected. Inclusion of the evaporation and any other losses is
straight forward in the simulation model.

2.2.6 Tailwater Curves

The model developed considers the strong hydraulic coupling that exists between any two
hydropower reservoirs in the system. This feature is taken into consideration while calculating
the head required for energy generation at each of the hydropower plants. The selection of
tailwater curve based on below conditional constraint (Eq. 9) related to the forebay elevations
of the downstream reservoirs.

 
if h jþ1;t > k l;o jþ1 ∩ h jþ1;t < k lþ1;
o
jþ1 then T j;t ¼ k l; jþ1 þ C l;; jþ1 G j;t
o
∀ j; l; t ð9Þ

The conditional constraint (Eq. 9) is modeled using an IF-THEN-ELSE construct in the

objected oriented modeling environment.

2.3 Modeling Environment and Representation of System

A simulation environment, STELLA (ISEES 2009), is used to model the operation of

a multiple reservoir system. Functional relationships and dependencies are defined
using converters and connectors respectively. Few built-in mathematical, logical and
statistical functions are used in the objects. In the current study, functions such as
RANDOM, IF-THEN-ELSE and DELAY are used. The environment specific features
such as sensitivity analysis, graphical inputs and other are used as well. The model
structure is aimed at reflecting the operation of a system of reservoirs in series, with
the final reservoir discharging into a lake with a strong hydraulic coupling existing
between the power plants. Three major components (i.e., system configuration, hydraulic
coupling and decision process) and five minor components for reservoirs and lakes are
developed in the modeling environment. These components can also be referred to as sectors.
A brief description of the model components is given next.

2.4 Reservoirs

Each reservoir is represented as a minor model component that has reservoir specific
properties. The structure of one of such reservoir given in Fig. 6 provides description of
the Seven Sisters reservoir from the Winnipeg Reservoir System. Similar structures are
required for all the other reservoirs and a lake that add up to a total of five minor
components. The reservoir specific characteristics for a particular component include,
Simulation of Multiple Hydropower Reservoir Operations 1947

Fig. 6 Object structure for one of the reservoirs (Seven Sisters Reservoir) in the four reservoir system

stage-storage relationships, local inflow, controlled flows from any other reservoirs,
discharge-efficiency curves, spill calculations, plant release, average storage level, flow
transport delay and calculation of head required for power generation. In optimization
and simulation models developed in the past for hydraulically coupled reservoirs, similar
rules were used and they are referred to as look-up tables. Similar rules exist for all the
reservoirs that are considered in the current study. If hydraulic coupling is neglected the
average storage value can be used for head calculation. This eliminates all the converters
that incorporate the IF-THEN-ELSE rules.

2.5 Lake System

The object structure of a lake (Winnipeg Lake) is shown in Fig. 7. The structure includes a
stock as a water body into which the final reservoir (Pine Falls reservoir) is discharging and
also a controlled outflow from the lake. Converters are used to calculate the average volume of
the lake and for determination of tailwater elevations curves. Apart from these major compo-
nents few others are used to incorporate different features in the model. These include selection
process for variable values, graphical relationships and all other calculations that are not
included in the major components. Link exists between different components if and only if
the variables in these components are interdependent. This can be seen in the main mapping
layer of the STELLA modeling environment.

2.5.1 Tailwater Curve Selection

This major component uses objects with logical functions (e.g. IF-THEN-ELSE) to
determine the appropriate tailwater elevation curve for use in the calculation of the head.
These objects are shown in Fig. 7. A sample set of rules using IF-THEN-ELSE construct
(for Mc Arthur generating station) are given below.
1948 R.S.V. Teegavarapu, S.P. Simonovic

Fig. 7 Object structure for a lake system (Lake Winnipeg)

IF ðaverage storage 3 >¼ 246:583Þ AND ðaverage storage 3 < 246:736Þ THEN
ð246:583 þ ðTotal flow at Mc Arthur  0:067ÞÞ ELSE

IF ðaverage storage 3 >¼ 246:736ÞANDðaverage storage 3 < 246:888ÞTHEN

ð246:736 þ ðTotal flow at Mc Arthur  0:064ÞÞ ELSE
Endif

2.6 Model Application

The real-time operation model is applied to a series of four reservoirs on the

Winnipeg River in the Manitoba province in Canada. These four reservoirs in cascade
form a part of a much more complex network of hydropower reservoirs maintained by
the local hydropower corporation, Manitoba Hydro. Only the plants on the Winnipeg
River maintained by Manitoba Hydro are used as test bed for the developed model. A
schematic representation of the plants in the case study region is given in Fig. 4. The
first plant, Seven Sisters, receives the controlled flow from the Slave Falls reservoir
located upstream, while the last plant under consideration, Pine Falls reservoir, drains
into the Lake Winnipeg. The McArthur reservoir has the largest storage of all the
plants. Strong hydraulic coupling between the hydro-generating plants on the Winni-
peg River is one of the important features of the present system. More details of the
case study region can be obtained from earlier works by Teegavarapu and Simonovic
(2000) and Barritt-Flatt and Cormie (1988). The weekly target power production for
all the power plants, initial and final forebay levels and the forecasted values of
inflows are assumed to be known. To facilitate the development, the model is divided
into three major components (i.e., system configuration, hydraulic coupling and
decision process).
Simulation of Multiple Hydropower Reservoir Operations 1949

3 Results and Discussion

In general, the models developed using the system dynamics principles are validated by
several tests that include: (1) replication; (2) sensitivity; and (3) prediction. These tests will
confirm the structure of the model with the physical system that is modeled. The replication
test is carried out by using real-life data and also a simulation experiment to compare the
results obtained using release decisions by an already existing optimization model
(Teegavarapu and Simonovic 2000). The simulation model developed in the current study is
not dependent or based on the optimization model. However, the optimal releases obtained
from the model were used in the simulation model to evaluate the system performance with or
without changes to these releases. The optimization model (Teegavarapu and Simonovic 2000)
was originally developed in General Algebraic Modeling System (GAMS) modeling language
(Brooke et al. 1996). The original optimization model development required a specialized
language referred to as algebraic modeling language (AML) and a GAMS language-based
syntax. The model is executed using a mixed integer nonlinear programming (MINLP) solver
(DICOPT (DIscrete Continuous OPTimizer) and the formulation required binary variables
which resulted in computational intractability and execution time issues relevant to real-term
operation of the same coupled hydropower system. The optimization model (Teegavarapu and
Simonovic 2000) was used for short-term optimal operation and not applicable for real-time
operation due to computational issues. The current model reported in this paper uses system
dynamics simulation principles for simulation of a hydrologically and hydraulically coupled
hydropower system. The main advantage of this model is that it can be used for simulation in
real-time, scenario generation and performance evaluation of hydropower energy generation
system.
Validation of the model results carried out by comparison of SD model results and those
available from actual operations of the system by the provincial hydropower corporation. The
power generation values at a daily temporal scale obtained from the SD model are within
0.4 % of the values realized from real-life operational results. Power generation values for sub-
daily time steps cannot be compared as hydraulic coupling is not exhaustively modeled in the
current SD model developed in this study. The simulation model is also tested by using the
optimal release decisions of the optimization model to replicate the storage variations as
obtained by the previous model. This test can also be used to validate the prediction aspect
of the system. The optimal total power production obtained from the use of optimization model
(Teegavarapu and Simonovic 2000) is 44.8 GW h. The power is obtained based on an
objective of minimizing the overall cost of power production with initial reservoirs levels of
274.1, 252.7, 247.4, 229.1 and 217.7 m at Seven Sisters, Mc Arthur, Great Falls, Pine Falls
and Winnpeg Lake respectively. The local flows to Seven Sisters, Mc Arthur and Pine Falls are
0.586, 0.014 and 0.101 103 m3/s respectively.
Sensitivity of the model to different inputs is tested by using extreme events (high and low
inflows). Various other tests that validate the dimensional consistency of the equations used
and robustness of the model in handling extreme conditions are carried out before the model is
adopted for implementation. The model developed in the current study is validated using all
these tests. Different initial storage conditions and the value of controlled flow into the system
are considered for performing the sensitivity analysis. More scenarios can also be obtained by
changing the flow transport delay between the plants. The outflow at each reservoir can be
varied either graphically or based on any existing operating policy provided by an optimization
model. This will result in different power generation values. The model can be run to obtain
the operation rules using different initial values of initial storage states while retaining a
constant inflow scheme to the reservoir system. The key model parameters are: available
1950 R.S.V. Teegavarapu, S.P. Simonovic

Table 2 Variation of forebay elevations of Seven Sisters reservoir based on different flow scenarios

Scenario

1 2 3
Day Forebay elevation (ft.)

1 274.107 274.107 274.107

2 273.808 273.720 273.631
3 273.509 273.332 273.156
4 273.211 272.945 272.680
5 272.912 272.558 272.205
6 272.613 272.171 271.729
7 272.314 271.784 271.272

water (conditioned on initial starting storages, later flows and outflows upstream of any
reservoir), travel delay and number of tail-water elevation curves used. The curves are critical
in characterizing hydraulic coupling between runoff the river hydropower generating plants.
In the current study, the outflow values that are used are optimal and are obtained from the
optimization study by Teegavarapu and Simonovic (2000). However, these values are not
always implemented in real-time and need to be refined adaptively Oj,t. Sensitivity analysis can
be easily performed using simulation environment to generate a variety of operation rules. One
such experiment is carried out where the outflow for the first reservoir (Seven Sisters) is varied
starting with the initial value and three different levels within the interval [0.504, 0.586]
103 m3/s. The variation of the reservoir level is given in the Table 2. Each scenario is based on
a particular value of outflow. Three scenarios are provided in the Table 2. It can be observed
that the variation is higher for the scenario 3 as the value of discharge is high compared to
other two scenarios.
The variations will be observed only if the controlled inflow into the reservoir is reduced to
0.424 103 m3/s. The variations of tailwater elevations for three scenarios corresponding to
different release patterns are given in the Table 3. The variation in the tailwater elevation will
not be observed if the inflow is high enough to cause spill when release is restricted. For inflow

Table 3 Variation of tailwater elevations at Seven Sisters reservoir based on different flow scenarios

Scenario

1 2 3
Day Tailwater elevation (ft.)

1 254.953 254.977 255.002

2 254.825 254.852 254.877
3 255.358 254.852 254.877
4 255.358 255.379 255.404
5 255.358 255.379 255.404
6 255.358 255.379 255.398
7 255.358 255.379 255.279
Simulation of Multiple Hydropower Reservoir Operations 1951

level of 0.586 103 m3/s (controlled inflow), no variation in the tailwater elevation is observed.
The use of IF-THEN-ELSE rules to select an appropriate curve is valid in the model for
addressing the issue of hydraulic coupling. The tailwater curves are derived for specific
downstream forebay elevations and outflow conditions and these rules cannot be avoided.
The forebay variations for Seven Sisters are very small compared to those at all other
reservoirs. The Seven Sisters and Pine Falls behave more like run-of-the-river plants and the
two middle reservoirs in the series play a regulatory role.
The flow transport time between the plants can be used to evaluate the effect of the delay
time on the total amount of power produced at each of the plants. The model developed in the
current study uses a fixed delay approach. A built-in function, DELAY or an object, Conveyor
(in STELLA environment), can be used to introduce the flow transport delay. Conveyor
as Stock object has special properties that allow it to convey material (in this case flow)
at a pre-defined time intervals. The modeling environment can also be used to incorpo-
rate a routing model. A number of stocks can be used in between the two reservoirs to
model the storage effect of the stream. A fixed delay of 1 day is used to understand its
effect on power production at each of the plants. Details of the power production at each
of the hydropower plants with and without flow transport delay are given in the Table 4.
In general, Flow transport delay reduces the amount of power that can be generated
within the operating horizon, as some amount of water is not available to provide
adequate head for power generation. The total power production values without and
with flow delay are equal to 44.83 and 38.73 GW h respectively. Approximately 12.5 %
reduction in the power is realized during the time horizon of 7 days due to flow delays.
The power generation values will change based on the location where the delay is
introduced.

3.1 Influence of Hydraulic Coupling

Hydraulic coupling between any two hydropower reservoirs is addressed in the current
study using tail-water elevation curves which are conditioned on the downstream forebay
elevations. A total of 5, 11, 11 and 15 tailwater curves are used for each of the reservoirs
in the order from the upstream reservoir to the lake. When no coupling exists, the energy
generation head is calculated using the average forebay and tailwater elevations or using
constant tailwater elevations. The total power production values with no hydraulic
coupling, and without and with flow delay of 1 day are equal to 46.40 and
40.87 GW h respectively. Details of the power production at each of the hydropower
plants with and without flow transport delay are given in the Table 5. In a hydropower
system where strong hydraulic coupling exists, assumption of no coupling has resulted in

Table 4 Variation of power generation (GWhrs) at four reservoirs considering flow delay with hydraulic
coupling

Power generation (GWhrs)

Hydropower reservoir No delay Delay

Seven Sisters 15.320 13.850

Mc Arthur 5.850 6.030
Great Falls 13.810 11.620
Pine Falls 9.850 7.230
1952 R.S.V. Teegavarapu, S.P. Simonovic

Table 5 Variation of power generation (GWhrs) at four reservoirs considering flow delay with no hydraulic
coupling

Power generation (GWhrs)

Hydropower reservoir No delay Delay

Seven Sisters 15.800 14.380

Mc Arthur 6.060 6.820
Great Falls 14.420 12.240
Pine Falls 10.130 7.430

increased power production values due to no influence of tailwater elevation variations

on available head calculations. These energy values are 3.5 and 5.5 % higher than those
from similar conditions with coupling. However, the reduction of total energy value in
the scenario of no coupling is 8.82 % when delay is considered which is lower than the
case when coupling is incorporated in the model. It is evident from the Tables 4 and 5,
the energy production values have increased at each of the plants when no hydraulic
coupling was considered. The increase in energy production value is higher than for the
second plant (McArthur) compared to that of others when delay was considered in case
of no hydraulic coupling. The McArthur plant with highest storage has the advantage of
storing the delayed flows to improve the power production from a no delay condition.

3.2 Heuristic Operating Rules

Reservoir operating rules are often refined and re-defined based on a variety of
situations in real-time. Judgment and experience of reservoir operators become useful
in many situations. One such case where heuristics are used would be to increase the
release whenever the forebay elevation of a reservoir is increasing to avoid spill.
However, the reservoir operator’s objective is to release as much water as possible to
maximize the hydropower generation may influence the tailwater elevation which
affects the head. Incorporation of different heuristic rules at different reservoirs would
lead to a system behavior that is not easily predictable or counter intuitive. If
hydraulic linking is negligible, the tailwater rises or variations will not be observed.
In case of extreme events (high or low inflows), the operating rule would depend on
the energy demand and storage available.

3.3 System Performance Measures

Two indices, namely reliability and vulnerability (Hashimoto et al. 1982) are used to
measure the system performance. Reliability of the system can be assessed by the
number of times the target demand is met during the whole operating horizon.
Vulnerability is defined in terms of a monetary value attached to a particular failure
decided by a penalty function provided by the user. These penalty functions are
similar to membership functions (Teegavarapu and Elshorbagy 2005) where the
decision makers preferences are attached to specific performance measure are modeled
as linear, nonlinear or sigmoid curves with penalty value confined to a specific range.
These two indices are appropriate for hydropower systems as the performance of the
Simulation of Multiple Hydropower Reservoir Operations 1953

system will be measured based on the number of times the target electric power
demand is met and the specific time interval in which the target demand is not met.
Vulnerability identifies the severity of failure in case a failure occurs. In the current
study, loss functions are developed to obtain a penalty value (monetary) associated
with a particular failure (e.g. failure to meet the target demand). These curves can be
modified by user with the help of a Graphical User Interface (GUI) provided by the
simulation environment.
The reliability and vulnerability of the system largely depend on inflows that form the
major inputs to the system. Variants of reliability indices are possible and are associated with
the number of times the pre-specified demand is met. Two reliability indices (λ and λo)
developed in the current study are described by Eqs. 11 and 12.
XT Xn
φ¼ ε j;t ð10Þ
t¼1 j¼1

 
if φ > φT arget ; then λ ¼ 1; else λ ¼ 0 ð11Þ

The variable φ is the total energy produced during the entire time horizon of operation
given by the summation of energy produced at each reservoir and in all time intervals within
the total time horizon and λ is the reliability of the system, and φTarget is the target energy from
the system. The value of λis binary and can take only values of one or zero.
Alternatively, a variant of reliability index (λ) can be expressed as a ratio of number of time
intervals the energy production exceeded the target to the total number of time intervals
(months). The refined definition of reliability, λo, is given by:

ν
λo ¼ ð12Þ
T

where ν the number of times is the energy production exceeded the target and T is the total
number of time intervals in operating time horizon under consideration.
The vulnerability level or achievement ratio is defined based on the total energy value (φ)
produced in comparison to the target value, φTarget. The vulnerability index indicates the cost/
penalty associated with a failure. The vulnerability index (ϕ) is calculated based on a
performance function, f(ψ), defined based on total deficit

φ
ψ¼ ð13Þ
φTarget

ϕ ¼ f ðψÞ ð14Þ

A typical penalty function defined for the current study is shown in Fig. 8. This function is
similar to a membership function generally defined in fuzzy set theory. The values of the
indices (reliability and vulnerability) for an arbitrarily selected set of inflows, associated with a
1954 R.S.V. Teegavarapu, S.P. Simonovic

0.9

0.8

0.7
Vulnerability ( )

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Achievement Ratio ( )

Fig. 8 Penalty function for defining vulnerability index

specific reservoir (Mc Arthur) are given in the Table 6. The values for vulnerability index are
based on accumulated values for all time intervals in the time horizon. The reliability values
are calculated based on Eq. 11. Smaller values of the index suggest better performance of the
system. It should be noted that there is no conceptual difficulty in developing these criteria for
the whole system. One of major advantages of the system dynamics simulation environment is
that qualitative information can be handled. For example, if two operating polices are
available, then operator’s willingness to use any one of the policies can be incorporated
through graphical functions within the simulation environment. The sensitivity analysis feature
of simulation environment is particularly useful for reservoir/plant managers to generate
operating rules for various uncertain conditions. This feature is also useful in instances where
the time interval within which decisions are required is too small to run an optimization model.
The model can be run to obtain the operation rules using different values of initial storage
states while retaining a constant inflow scheme to the reservoir system. The model can be
extended to include the simulation of individual turbine operations at each of the reservoirs. To
achieve a complete representation of the physical system (e.g. incorporation of hydraulic
coupling and calculation of average storage elevations), additional objects are defined in the
model.

Table 6 Performance indices for different inflow schemes

Inflow scheme Reliability Vulnerability

1 1 0.24
2 1 0.35
3 1 0.29
Simulation of Multiple Hydropower Reservoir Operations 1955

The use of IF-THEN-ELSE rules for selection of tailwater curves introduces a form of
rigidity into the system that has influence on the dynamics of the system. The rigid structure of
these rule constructs may lead to discontinuity in some instances, especially where the
variation in system conditions is difficult to model. The dynamic variation of tailwater
elevation due to changes in forebay elevation can only be modeled for discrete intervals of
these elevations as the tailwater elevation curves are available for these discrete water levels.
Any behavior of the system within the intervals can be associated with the use of one of
tailwater elevation curves. The system representation can be improved if the number of
tailwater elevation curves is increased. In summary, the model uses objects with specific
properties that provide an accurate representation of the physical system under consideration.
The simulation environment provides an easy mechanism to include physical aspects of the
system which otherwise are difficult to incorporate into traditional simulation models. The
scenarios generated will be useful for real-time implementation of the operating rules. Even
though the model is specific to reservoir operation related to hydropower generation, the
modeling concepts can be extended to any type of reservoir system with any type of objective.

3.4 Sensitivity Analysis

The object-oriented simulation environment provides sensitivity analysis feature to generate

multiple runs of the simulation model by perturbing a single parameter or variable value in the
model. The feature includes the advantage of generating the value based on a probability
distribution. Uniform and normal distributions are allowed in STELLA modeling environ-
ment. By using the distribution of the inputs, sensitivity analysis process therefore becomes a
Monte Carlo simulation exercise. In the system under consideration, an important variable that
effects the operations and power generation is the flow to the first reservoir (Seven Sisters
Reservoir) in the cascading system of reservoirs. This flow is important as the system has little
storage at all the reservoirs. Monte Carlo simulation is conducted by perturbing this flow
variable by assuming uniform distribution of values within the range of [0.504–0.586] 103 m3/
s. The random number generator available in the simulation environment is used to generate
1000 of these flow values. The distribution of flows and total power generation for the entire
cascading reservoirs for 7 days are given in Fig 9a and b respectively. The power generation
values are skewed to right as the frequency of flow values in the range of [0.504–0.586]
103 m3/s are higher. Using different values of flows, it is possible to generate an ensemble of
power generation scenarios and also exceedance probabilities associated with these energy
levels. The latter is possible using a cumulative density function based on the Monte Carlo
simulation runs. The distribution of total power generation values was near normal when flow
values to the first reservoir used in the system follow a Gaussian distribution.

4 General Remarks

The modeling of the multiple reservoir system using the system dynamics approach has helped
in understanding the dynamics of the operation. The modeling approach has many advantages
compared to the traditional techniques where the models are developed using formal program-
ming languages. The simulation model developed in the current study will have enormous
utility in variety of real-time operational conditions. Advantages of the model include: (1)
generation and evaluation of different operating rules based on a variety of conditions related
to inflows, initial storage levels and tailwater conditions; (2) evaluation of system performance
through different indices; and (3) usage for real-time operational decisions. The object-oriented
1956 R.S.V. Teegavarapu, S.P. Simonovic

(a)

0.497 0.511 0.525 0.539 0.553 0.567 0.582 0.596

m

(b) 250

200

150
Frequency

100

50

0
44 44.1 44.2 44.3 44.4 44.5 44.6 44.7 44.8 44.9
Power (GWhrs)

Fig. 9 Distribution of (a) randomly generated flows and (b) power generation using Monte Carlo simulation

simulation environment is ideal for application to reservoir operation problems. Reliability and
vulnerability used as indicators to quantify the system performance in response to different
inflow conditions can benefit decision makers involved in the economics side of system
operation.
In the current study a dynamic modeling environment is used to develop a conceptually
simple model for hydraulically coupled reservoir system. Object-oriented simulation environ-
ments based on SD principles such as STELLA is often referred to as dynamic modeling
environment which can handle temporal complexity (Costanza and Gottlieb 1998; Hannon and
Ruth 1997). However, spatial representation of the physical system is also possible using
STELLA as demonstrated by the model developed in this study. Simulation models developed
using system dynamics principles are designed to understand the basic structure and behavior
of the physical system. Therefore, trends in the results or alternative policy decisions derived
based on the model results are often times more necessary than the actual numerical values
obtained through the simulation. Simulation models developed using spreadsheet programs
lack transparency while models developed using object oriented modeling environment with
Simulation of Multiple Hydropower Reservoir Operations 1957

SD principles allow the modeler to study the structural and behavioral aspects of the processes
included in the model.

5 Conclusions

A system dynamics simulation model is developed for modelling the dynamics of a hydrau-
lically coupled hydropower reservoir system operation. The system is simulated using an
object-oriented simulation environment using generic objects that represent reservoir storages
and inflows. The hydraulic coupling feature is modeled using conditional constraints incor-
porated via functional relationships linking tailwater elevations and forebay elevations of
upstream and downstream reservoirs. Incorporation of hydraulic coupling in system results
in conceptually accurate representation of the physical system with reduction in the power
generation value compared to that from a condition when no coupling is considered. Flow
delays have considerable impact on the power generation amounts in a run-of-the river
hydropower plants with very little storage variations. Performance indices such as reliability
and vulnerability are defined to assess the system status to different inflow conditions. The
vulnerability index is defined using fuzzy membership resembling functions to help incorpo-
rate the operator preferences attached to the overall system performance. Monte Carlo
simulation is used to assess the variability of power generation due to changing system
conditions. Two major limitations of the model developed in this study are: (1) the lack of
procedures for explicit travel time calculation of flow between any two reservoirs and (2)
limited number of functional relationships used to represent the link between tailwater and
forebay elevations. Future improvements in the model should focus on accurate modeling of
flow transport delays and hydraulic coupling.

References

Ahmad S, Simonovic SP (2000) System dynamics modeling of reservoir operations for flood management. J
Comput Civil Eng, ASCE 14(3):190–198
Barrit-Flatt PE, Cormie AD (1988) A comprehensive optimization model for hydro-electric reservoir operations,
Proceedings of Computerized Decision Support systems for Water Managers, Labadie JW et al., ASCE:
463–477
Brooke A, Kendrik D, Meeraus A (1996) GAMS: a user’s guide, 286 pages
Chau KW, Wu CL, Li YS (2005) Comparison of several flood forecasting models in Yangtze River. J Hydrol
Eng ASCE 10(6):485–491
Cheng CT, Chau K, Sun Y, Lin Y (2005) Long-term prediction of discharges in Manwan Reservoir using
artificial neural network models. Lect Notes Comput Sci 3498:1040–1045
Costanza R, Gottlieb S (1998) Modeling ecological and economic systems with STELLA: part II. Ecol Model
112(2–3):81–84
Coyle RG (1996) System dynamics modeling: a practical approach. Chapman and Hall, London
Deaton ML, Winebrake JI (1999) Dynamic modeling of environmental systems. Springer, New York
Elshorbagy A, Teegavarapu RSV, Ormsbee L (2005) Framework for assessment of relative pollutant loads with
limited data. J IWRA Water Int 30(3):350–355
Elshorbagy A, Teegavarapu RSV, Ormsbee L (2006) Assessment of pathogen pollution in watersheds using
object-oriented modeling and probabilistic analysis. J Hydroinf 7:51–63
Fernandex JM, Selma MAE (2004) The dynamics of water scarcity on irrigated landscapes: Mazarron and
Aguilas in south-eastern Spain. Syst Dyn Rev 20(2):117–137
Fletcher EJ (1998) The use of system dynamics as a decision support tool for the management of surface water
resources. First International Conference on New Information Technologies for Decision Making in Civil
Engineering, Montreal, Canada: 909–920
1958 R.S.V. Teegavarapu, S.P. Simonovic

Ford A (1999) Modeling the environment. Island Press, USA

Forrester JW (1961) Industrial dynamics. Productivity Press, Portland
Hannon B, Ruth M (1997) Modeling dynamic biological systems. Springer, New York
Hashimoto T, Stedinger JR, Loucks DP (1982) Reliability, resiliency and vulnerability criteria for water resource
system performance evaluation. Water Resour Res 18(1):14–20
Huang GH, Chang NB (2003) Perspectives of environmental informatics and systems analysis. J Environ Inform
1(1–6):2003
Huggett R (1993) Modeling the human impact on nature. Oxford University Press, NY
Keyes AM, Palmer RN (1993) The role of object-oriented simulation models in the drought preparedness studies,
Proceedings of the 20th Annual National Conference, Water Resources Planning and Management Division,
ASCE, Seattle, WA:479–482
Lund JR, Ferreira I (1996) Operating rule optimization for Missouri river reservoir system. J Water Resour Plan
Manag, ASCE 122(4):287–295
Moffatt I (1991) Causal and Simulation Modeling using System Dynamics, Concepts and Techniques in Modern
Geography, CATMOG, Australia
Muttil N, Chau K-W (2006) Neural network and genetic programming for modelling coastal algal blooms. Int J
Environ Pollut 28(3–4):223–238
Nandalal KDW, Simonovic SP (2003) Resolving conflicts in water sharing: a systemic approach. Water Resour
Res 39(12):1362–1373
Nirmalakhandan N (2002) Modeling tools for environmental engineers and scientists. CRC Press, Boca Raton
Roberts N, Andersen D, Deal R, Garet M, Shaffer W (1983) Introduction to computer simulation, a system
dynamics approach. Productivity Press, Portland
Saysel AK, Barlas Y, Yenigun O (2002) Environmental sustainability in an agricultural development project: a
system dynamics approach. J Environ Manag 64:247–260
Sehlke G, Jacobson J (2005) System dynamics modelling of transboundary systems: the Bear River basin model.
Ground Water 43(5):722–730
Simonovic SP (2002) World water dynamics: global modeling of water resources. J Environ Manag 66(3):249–
267
Simonovic SP (2009) Managing water resources: methods and tools for a systems approach, UNESCO and
Earthscan Publication
Simonovic SP, Ahmad S (2005) Computer-based model for flood evacuation emergency planning. Nat Hazards
34(1):25–51
Simonovic SP, Fahmy H (1999) A new modeling approach for water resources policy analysis. Water Resour Res
35(1):295–304
Simonovic SP, Rajasekaram V (2004) Integrated analysis of Canada’s water resources: a system dynamics model.
Can Water Resour J 29(4):223–250
Simonovic SP, Fahmy H, ElShorbagy A (1997) The use of object-oriented modeling for water resources planning
in Egypt. Water Resour Manag 11:243–261
Stave KA (2003) A system dynamics model to facilitate public understanding of water management options in
Las Vegas, Nevada. J Environ Manag 67:303–313
STELLA (2009) STELLA (Structural Thinking, Experiential Learning Laboratory with Animation): model
building and simulation tutorials, NH, USA
Teegavarapu RSV, Elshorbagy A (2005) Fuzzy set-based error measure for hydrologic model evaluation. J
Hydroinf 7:199–208
Teegavarapu RSV, Simonovic SP (2000) Short-term operation model for coupled hydropower plants. J Water
Resour Plan Manag, ASCE 126(2):98–106
Teegavarapu RSV, Tangirala AK, Ormsbee L (2005) Modeling water quality management alternatives for a
nutrient impaired stream using system dynamics and simulation. J Environ Inform 5(2):73–81
Wood AJ, Wollenberg BF (1984) Power generation, operation, and control. John Wiley and Sons, New York
Wu CL, Chau KW, Li SY (2009) Predicting monthly streamflow using data-driven models coupled with data-
preprocessing techniques. Water Resour Res 45(W08432):2009. doi:10.1029/2007WR006737
Wurbs R (1993) Reservoir-system simulation and optimization models. J Water Resour Plan Manag, ASCE
116(1):52–70
Xu ZX, Takeuchi K, Ishidaira H, Zwang XW (2002) Sustainability analysis for yellow river water resources
using the system dynamics approach. Water Resour Manag 16:239–261