Energy Conversion and Management: X 23 (2024) 100610

Contents lists available at ScienceDirect

Energy Conversion and Management: X

journal homepage: www.sciencedirect.com/journal/energy-conversion-and-management-x

Development of a dynamic mathematical model of PEM electrolyser for

integration into large-scale power systems
Siavash Asiaban a, b, *, Dimitar Bozalakov a, b, Lieven Vandevelde a, b
a
Department of Electromechanical, Systems & Metal Engineering, Faculty of Engineering & Architecture, Ghent University, Tech Lane Ghent Science Park–Campus
Ardoyen, Technologiepark-Zwijnaarde 131, B-9052 Ghent, Belgium
b
FlandersMake@UGent–Corelab MIRO, Flanders Make, B-9052 Ghent, Belgium

A R T I C L E I N F O A B S T R A C T

Keywords: Proton exchange membrane (PEM) electrolyser stands as a promising candidate for sustainable hydrogen pro­
PEM electrolyser duction from renewable energy sources (RESs). Given the fluctuating nature of RESs, accurate modelling of the
Hydrogen PEM electrolyser is crucial. Nonetheless, complex models of the PEM electrolyser demand substantial time and
Modelling
resource investments when integrating them into a large-scale power system. The majority of introduced models
Dynamic
Static
in the literature are either overly intricate or fail to effectively reproduce the dynamic behaviour of the PEM
Thermal electrolyser. To this end, this article aims to develop a model that not only captures the dynamic response of the
PEM electrolyser, crucial for conducting flexibility studies in the power system, but also avoids complexity for
seamless integration into large-scale simulations without comprising accuracy. To verify the model, it is vali­
dated against static and dynamic experimental data. Compared to the investigated experimental cases, the model
exhibited an average error of 0.66% and 3.93% in the static and dynamic operation modes, respectively.

extracted from compounds that contain it, such as water, coal, natural
1. Introduction gas or biomass [7]. Typically, fossil fuels and RESs are the two major
primary energy sources for hydrogen production, with the former
With the depletion of fossil fuels, rising awareness of energy security, emitting carbon dioxide (CO2 ) as a by-product. Therefore, to achieve
growing electricity demands and considerable concerns about environ­ sustainable hydrogen production from fossil fuels, the CO2 needs to be
mental issues, renewable energy sources (RESs) are attracting significant captured and stored [8]. On the other hand, water electrolysis, which is
attention worldwide. Electrification is widely recognised as the most the focus of this article, distinguishes itself as a highly favourable path to
efficient approach to utilise RESs. Nevertheless, due to the power fluc­ producing hydrogen from RESs.
tuations of renewables and electricity transmission constraints, several Based on the hydrogen production method, the four main types of
obstacles emerge when a high share of RESs is integrated into the power electrolysers are alkaline (AE), proton exchange membrane (PEM), solid
system [1, 2]. These challenges negatively affect multiple characteristics oxide (SO) and anion exchange membrane (AEM). While SO and AEM
of the electricity grid including voltage, frequency, reliability and power electrolysers are rather in their infancy, AE is the most mature tech­
quality. In recent years, the production of green molecules has gained nology, followed by the PEM type [9, 10].
attention as a means to address this issue [3]. For instance, in the case of Considering the two classes of electrolysers that are further along in
a mismatch between the generation and load, chemical storage/release development, PEM has several advantages over its alkaline counterpart.
of energy in the form of hydrogen (H2 ), or other molecules like methane To start with, the PEM electrolyser is able to operate at higher current
and methanol, helps the power system operators to balance the grid densities of more than 2 A/cm2 , which typically reduces the footprint of
[4–6]. the PEM electrolyser to about one-fifth that of the same capacity alkaline
Hydrogen is the simplest and most abundant element on earth that electrolyser [11].
can store and deliver energy. It is an odourless, colourless, tasteless and Furthermore, the response time of the AE in case of dynamic oper­
nonpoisonous gas a release of which is not known to pollute air or water. ation is slower compared to the PEM electrolyser [12]. This stems from
However, it generally does not exist in its free form and must be the fact that AE employs a liquid electrolyte rather than the solid

* Corresponding author at: Department of Electromechanical, Systems & Metal Engineering, Faculty of Engineering & Architecture, Ghent University, Tech Lane
Ghent Science Park–Campus Ardoyen, Technologiepark-Zwijnaarde 131, B-9052 Ghent, Belgium.
E-mail address: [email protected] (S. Asiaban).

https://doi.org/10.1016/j.ecmx.2024.100610

Available online 6 May 2024

2590-1745/© 2024 Ghent University. Published by Elsevier Ltd. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-
nc/4.0/).
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

Nomenclature Greeks
α Charge transfer coefficient
A Area, m2 δ Thickness, m
C Concentration, mol/m3 ∊ Porosity
Cl Lumped thermal capacitance, J/K ε/k Lennard-Jones potential, K
cp Specific heat capacity, J/kg.K Θ Mean molecular radii, Å
cʹ
p Specific heat capacity, J/mol.K Λ Coolant to water mole ratio
λ Water content of membrane
D Binary diffusion coefficient, m2 /s
ξ Tortuosity
Dw Diffusion coefficient, m2 /s
ρ Density, kg/m3
Eact Activation energy, J/mol
σ Conductivity, 1/Ω.m
E0 Standard cell reversible voltage, V
τ Dimensionless temperature
e Error
Ω Diffusion collision integral
F Faraday constant, A.s/mol
H Enthalpy, J/mol Subscripts and superscripts
h Heat transfer coefficient, W/m2 .K act Activation
I Current, A amb Ambient
j Current density, A/m2 an Anode
j0 Exchange current density, A/m2 ca Cathode
M Molar mass, kg/mol ch channel
Ṁ Flow rate, kg/s con Concentration
N Number of moles cons Consumed
n Number of cells diff Diffusion
Ṅ Flow rate, mol/s eod Electro-osmotic drag
ṅ Molar flux, mol/m2 .s eff Effective
nd Electro-osmotic drag coefficient exp Experiment
P Pressure, Pa gen Generated
p Partial pressure, Pa hp Hydraulic pressure
Q Heat flow, W in Inlet
R Resistance, Ω.m2 k Knudsen
Ru Universal gas constant, J/K.mol mem Membrane
r Mean pore radius, m ohm Ohmic
T Temperature, K out Outlet
V Voltage, V ref Reference
rev Reversible
v Volume, m3
sim Simulation
X Molar fraction
st stack
Zexp Number of experiments

electrolyte of the PEM electrolyser, which results in higher inertia and heat loss in the anode. Moreover, the model incorporates a constant
thus slower ion transportation [13]. This issue becomes of utmost rele­ voltage source and a series resistor to simulate the reversible voltage and
vance when the electrolyser is connected to an intermittent RES as it has membrane loss, successively. Finally, using experimental tests and sys­
to operate under fast dynamics. Hence, AE is chiefly used in fixed power tem identification techniques, the values of the circuit elements are
mode, while the PEM electrolyser only requires several minutes to start determined. Nevertheless, the main drawback of the model is its limited
up and a few seconds to ramp up from idle mode to the rated power. applicability. That is, the adapted values for the circuit elements are only
Additionally, thanks to the low gas crossover rate of a polymer valid for the specific operation conditions including input current,
electrolyte membrane, the PEM electrolyser is able to operate under a temperature and pressure, making the model less reliable when used
wide range of power densities. Besides, PEM electrolysers have higher under different operating conditions.
efficiency than alkaline [14, 15]. Nevertheless, it is crucial to precisely In order to improve the accuracy of the aforementioned model, the
examine the behaviour of the PEM electrolyser when subjected to dy­ authors in [17] have proposed an adaptive algorithm in which the pa­
namic operating conditions. rameters of the circuit are calculated based on different electrical input
To this end, there are ongoing endeavours in the literature to model currents. However, even with utilising the adaptive approach, it remains
the performance of the PEM electrolyser. These models can be divided challenging to model the effects of other operating conditions such as
into two major categories: circuit-based models which are built upon temperature and pressure.
electrical elements, and mathematical models, which are grounded in As far as the second group is concerned, in [18], a mathematical
electrochemical principles. Concerning the first group, the authors in model of the PEM electrolyser consisting of four compartments, namely
[16] have modelled a static-dynamic model of the PEM electrolyser the anode, cathode, membrane and voltage, was introduced and simu­
using circuit elements. In this approach, two parallel RC branches are lated in Matlab/Simulink. The model is created based on the conser­
employed to represent the behaviour of the anode and cathode. The vation of the mole balance of associated substances at the anode and
capacitors are used to mimic the capacitance behaviour of the cell, ac­ cathode. Regarding the membrane, electro-osmosis and diffusion are
counting for the specific time needed by the charges to travel inside the taken into account as the driving forces for water mass transfers through
anode and cathode. While the resistor of the cathode solely models the the membrane. This approach enables the calculation of the partial
heat loss, the resistor of the anode represents both the Gibbs energy and pressure of all substances at the anode and cathode, which is later used

2
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

for computing the reversible cell voltage. restricts simulating the electrolyser behaviour at the non-similar anode
The same concept of modelling was followed in [19] and [20]. and cathode pressures.
However, as opposed to what is claimed in the mentioned articles, the Generally, the literature faces a scarcity of a fast and precise model of
models replicate the static response of the PEM electrolyser and do not the PEM electrolyser, which can effectively replicate the static response
account for its dynamic behaviour. This is confirmed by [13] and does of the electrolyser as well as its dynamic behaviour. In this article, a
stem from the fact that the capacitance behaviour of the PEM electro­ developed model of the PEM electrolyser is presented. This article dis­
lyser is not modelled in any of the compartments. Thus, the close rela­ tinguishes itself from prior literature models by incorporating detailed
tionship between transport and electricity flow via Faraday’s law yields mathematical modules not only for the anode, cathode, membrane and
a static response. voltage but also for the thermal compartment. The deliberate simplifi­
Following these articles, the authors in [21] have incorporated bal­ cation, excluding non-essential mechanisms in each module, ensures
ance of plants (BoP) components, namely water pump, cooling fan, smooth integration of the model into large-scale simulations. This is of
storage tank, water tank, power supply, control units and sensors to the significant importance for electrical engineers engaged in flexibility
same electrolyser model. Similar to its predecessors, this model solely studies as complex elements like generators and transformers are
simulates the static response of the PEM electrolyser. already part of power system models. Notably, unlike some of the
A detailed static model has been presented and validated in [13]. In referenced articles such as [18–21], the model accurately replicates the
this article, mechanisms and phenomena that are generally neglected in genuine dynamic response of the PEM electrolyser, rather than solely its
other articles have been taken into account. In the membrane, in addi­ static behaviour. Moreover, the precision of the model is validated
tion to the electro-osmosis and diffusion mechanisms, the water flow against experimental data in both static and dynamic operation modes.
from the cathode to the anode due to the pressure gradient is consid­ By following the outlined steps in this article, researchers can simply
ered. Within the voltage ancillary, the concentration overvoltage is simulate a fast and reliable model of the PEM electrolyser. Lastly, with
addressed. Meanwhile, in the ohmic overvoltage sub-module, the model minor adjustments, this model can simulate the dynamic and static
goes beyond the commonly considered membrane resistance, taking response of the alkaline electrolyser as well.
into account the resistances of the bipolar plates and electrodes in a As mentioned earlier, the anode, cathode, membrane and voltage
detailed manner. Nevertheless, the values of the mentioned parameters modules are important since they enable to simulate the behaviour of
are generally insignificant when contrasted with other associated fac­ the electrolyser under various operating pressures. This further helps to
tors, and do not have a notable impact on the accuracy of the model [21, calculate the partial pressures of the substances at the electrodes,
22]. impacting the deviation of the reversible cell voltage from its standard
Although not validated against the experimental data, the study value, thereby the calculations of efficiency and cooling requirements of
conducted in [23] delved into the dynamic behaviour of a high-pressure the electrolyser. The significance of the thermal module lies in enabling
PEM electrolyser. The investigation involved considering the non-ideal the model to mimic the dynamic behaviour of the PEM electrolyser
behaviour of hydrogen and incorporating the unsteady state mass bal­ which becomes of substantial importance when coupled with the RESs.
ance equation of substances at the anode and cathode. Additionally, the Consequently, the derived results from the model align more closely
energy balance equations of the anode, cathode and membrane elec­ with the actual power consumption and produced hydrogen quantity by
trolyte assembly (MEA) are included. In this article, a one-dimensional the electrolyser. This is intriguing when it comes to conducting flexi­
dynamic model of the PEM electrolyser is presented where physio- bility studies on the PEM electrolyser as the evaluation of ancillary
electrochemical phenomena such as water flow through the mem­ services like frequency containment reserve (FCR) or automatic fre­
brane via electro-osmotic drag, diffusion and permeation are consid­ quency restoration reserve (aFRR) provision can be assessed more pre­
ered. Furthermore, the gas permeation through the membrane, e.g. cisely [26–28].
hydrogen, gas volume variations in the anode and cathode channels, gas Besides, as the feed water is an input to the model, it is capable of
compressibility, and water vaporisation are taken into account. The accounting for the dynamic changes in the feed water system, an
model allows for characterising gas purity at the cell/stack output. Yet, influential factor when considering the employment of non-pure water
as admitted by the authors, assumptions such as similar oxygen and like seawater. By coupling the electrolyser with the seawater desalina­
hydrogen concentrations on the membrane surface reduce the accuracy tion system, e.g. seawater reverse osmosis desalination system, the dy­
of the model. namic behaviour of the entire system can be explored. Furthermore, the
None of the previously mentioned models has considered the thermal model is able to effectively capture the dynamic changes in the produced
characteristics of the PEM electrolyser which is of paramount impor­ hydrogen. This facilitates the assessment of producing other green
tance when it comes to assessing its dynamic response. However, several molecules such as methanol while investigating the impact of the elec­
models in the literature have taken this into consideration. trolyser on the subsequent systems. Finally, while retaining a high level
The authors in [24] modelled a PEM electrolyser consisting of a of accuracy, it has been tried to ensure the model remains simple so that
steady-state voltage module as well as a linear dynamic thermal model. it can be easily integrated into the power system simulations smoothly
In this article, using an identification approach, the electrical and and limit simulation time. This simplicity is particularly appealing for
thermal parameters of the model are determined. The main purpose of large-scale and long-term simulations. In Table 1, the main character­
the article is to develop monitoring algorithms. The authors in [22] have istics of the developed model and existing models in the literature are
modelled the voltage ancillary beside a thermal module which is based highlighted.
on the lumped thermal capacitance. In a nutshell, this article pursues three main goals: 1) Developing a
In [25], a thermal module consisting of the heat exchange between static-dynamic model that behaves nearly similar to a real-life PEM
the PEM electrolyser and the surrounding area, the sum of entropic flows electrolyser, enabling flexibility studies, integrating it as an asset in
of both heat sources and sinks, water tank thermal energy accumulation, microgrids providing flexibility, power systems, etc.; 2) Keeping the
and heat loss via conduction is introduced. Thereafter, the model is model as simple as possible to facilitate its utilisation in extensive and
validated experimentally. long-term studies; 3) Achieving a trade-off between accuracy and
However, none of these models have taken into account the accurate simplicity in the model.
models of the anode, cathode or membrane compartments, which are of The rest of the article is organised as follows: Section 2 is dedicated to
use as the partial pressure of the substances at the anode and cathode are a thorough modelling methodology of each simulated module. This
used in the voltage calculations. Moreover, this prevents the monitoring encompasses the anode, cathode, membrane, voltage, and thermal
of partial pressures at the anode and cathode, the water flow through the components. Section 3 presents the results obtained from the model
membrane which affects the performance of the electrolyser and under various operating conditions. Additionally, the results undergo

3
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

Table 1
Comparison between the existing literature and the proposed model (L: Low, BA: Below Average, A: Average, AA: Above Average, H: High, NA: Not Applicable).
Reference Developed model [13] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25]

Anode/cathode module ✓ ✓ X X ✓ ✓ ✓ ✓ X ✓ X X
Membrane module ✓ ✓ X X ✓ ✓ ✓ ✓ X ✓ X ✓
Thermal module ✓ X X X X X X X ✓ ✓ ✓ ✓
Voltage module ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Dynamic ✓ X ✓ ✓ X X X X ✓ ✓ ✓ ✓
Accuracy in operation range H AA L BA A A A A AA H H A
Complexity A AA L BA A BA A A A H AA A
Static validation (Polarisation curve) ✓ ✓ X X X ✓ ✓ ✓ X X ✓ ✓
Dynamic validation ✓ NA ✓ ✓ NA NA NA NA ✓ X ✓ ✓

validation against experimental data to ensure the accuracy of the

model. To achieve this, a comparison is made between the static results
of the model and those of the experimental data, followed by a similar
comparison for the dynamic results. Finally, in Section 4, the obtained
results are discussed, and conclusions are drawn.

2. Model development

In the PEM electrolyser, water is fed into the anode, and due to the
applied current in the oxidation reaction, decomposed to hydrogen
protons, electrons and oxygen, as shown in Eq. (1). Later, while the
electrons migrate through the electric circuit towards the cathode, the
produced positive ions travel through the proton-conductive membrane
and reach the cathode as well to participate in the reduction reaction,
expressed in Eq. (2). As a result of this reaction, hydrogen is produced at
the cathode. The overall reaction, known as redox, is shown in Eq. (3).

2H2 O→4H+ + 4e− + O2 (1)

4H+ + 4e− →2H2 (2)

1
H2 O→H2 + O2 (3)
2

The developed model is composed of five various compartments, namely

the anode, cathode, membrane, voltage and thermal. The anode and
cathode modules are modelled based on the molar balance of relevant
substances, specifically water and oxygen for the anode, in addition to
water and hydrogen for the cathode. In the membrane compartment, the Fig. 1. Overview of the developed model in Simulink.
mechanisms via which the water traverses the membrane are modelled,
i.e., electro-osmotic drag and diffusion. The voltage ancillary in­ migrated from anode to the cathode through the membrane, and ṄO2 an,gen
corporates the reversible voltage calculations along with the ohmic stands for the molar flow rate of the generated oxygen at the anode.
overvoltage of the membrane and activation overvoltages of the anode As mentioned earlier, the fluctuations of the input water, if any, can
and cathode subsystems. Lastly, within the thermal module, the heat be incorporated into the model using Eq. (4). It is essential to note that in
sources and sinks as well as the injected water into the electrolyser for this model, the anode inlet molar flow rate of oxygen is assumed to be
cooling purposes are simulated. zero and its inclusion is merely for the sake of having a complete molar
A general overview of the interactions between the modules, simu­ balance equation.
lated in Matlab/Simulink, is shown in Fig. 1. In the forthcoming sub­ According to Eq. (1), Eq. (2) and Faraday’s first law of electrolysis,
sections, an exhaustive elaboration of each module is presented. the molar flow rate of the consumed water and generated oxygen at the
anode of the PEM electrolyser are given as follows:
2.1. Anode module
nI
ṄH2 Oan,cons = (6)
Based on Eq. (1), the molar balances of oxygen and water are 2F
considered in the anode module. The oxygen and water molar flow rate nI
equations at the anode can be expressed as follows: ṄO2 an,gen = (7)
4F
ṄH2 Oan = ṄH2 Oan,in − ṄH2 Oan,cons − ṄH2 Omem − ṄH2 Oan,out (4)
where n is the number of cells, I is the applied current and F is the
Faraday constant.
ṄO2 an = ṄO2 an,in + ṄO2 an,gen − ṄO2 an,out (5) By substituting Eq. (6) into Eq. (4) and Eq. (7) into Eq. (5), and
solving the algebraic loops, the partial pressures of water and oxygen at
where ṄH2 Oan,in and ṄH2 Oan,out are the anode inlet and outlet molar flow the anode can be derived via the ideal gas law as follows:
rate of water, ṄO2 an,in and ṄO2 an,out denote the anode inlet and outlet molar
NH2 Oan Ru T
flow rate of oxygen, ṄH2 Oan,cons is the molar flow rate of the consumed pH2 Oan = (8)
van
water at the anode, ṄH2 Omem represents the molar flow rate of water

4
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

NO2 an Ru T pH2 ca
pO2 an = (9) XH2 ca = (20)
van Pca

where NH2 Oan and NO2 an are the number of moles of water and oxygen at 2.3. Membrane module
the anode, respectively, Ru is the universal gas constant, T represents the
operating temperature of the electrolyser, and van denotes the anode When the water is introduced into the anode of the PEM electrolyser,
volume. a substantial portion of it undergoes the oxygen evolution reaction,
The pressure at the anode is equal to the total partial pressures of leading to the generation of protons and oxygen. However, a part of the
water and oxygen, described as: water passes through the membrane which influences the performance
of the electrolyser. Generally, three mechanisms cause the passage of
Pan = pH2 Oan + pO2 an (10)
water through the membrane, namely electro-osmotic drag, diffusion
Then, the molar fraction of each substance at the anode can be calcu­ and hydraulic pressure effect. While the first two mechanisms result in
lated as follows: the water flow from the anode to the cathode, the hydraulic pressure
effect yields water passage in the opposite direction. As a result, the net
pH2 Oan molar flow rate of water through the membrane is given as follows:
XH2 Oan = (11)
Pan
ṄH2 Omem = ṄH2 Omem,eod + ṄH2 Omem,diff − ṄH2 Omem,hp (21)
pO2 an
XO2 an = (12)
Pan where ṄH2 Omem,eod , ṄH2 Omem,diff and ṄH2 Omem,hp denote the water passage
through the membrane due to the electro-osmotic drag, diffusion and
2.2. Cathode module hydraulic pressure effect, respectively.
In [23], it is shown that the electro-osmotic drag is the dominant
In the cathode compartment, where hydrogen is evolved, a similar mechanism in the water transport through the membrane. Commonly,
approach to the anode module is followed, with hydrogen replacing electro-osmotic drag and diffusion are taken into account in the models,
oxygen. Thus, the molar flow rates of water and hydrogen at the cathode given that the influence of hydraulic pressure effect is negligible
are represented as follows: compared to them [18, 21].
As protons traverse the proton-conducting membrane, they drag
ṄH2 Oca = ṄH2 Oca,in + ṄH2 Omem − ṄH2 Oca,out (13)
water molecules with them, a phenomenon known as electro-osmotic
drag. This process is quantified using the dimensionless electro-
ṄH2 ca = ṄH2 ca,in + ṄH2 ca,gen − ṄH2 ca,out (14) osmotic drag coefficient, denoted as nd , which represents the number
of water molecules per proton travelling from the anode to the cathode
In Eq. (13) and Eq. (14), ṄH2 Oca,in and ṄH2 Oca,out stand for the cathode inlet (mol[H2 O]/mol[H+ ]). The molar flow rate of water through the mem­
and outlet molar flow rate of water, subsequently, ṄH2 ca,in and ṄH2 ca,out are brane due to the electro-osmotic drag can be described as follows [13]:
the inlet and outlet molar flow rate of hydrogen at the cathode,
nd jA
respectively, and ṄH2 ca,gen represents the molar rate of the generated ṄH2 Omem,eod = (22)
F
hydrogen at the cathode.
In this model, both the inlet hydrogen and water flow rates to the Here, j is the current density and A represents the active area of the MEA.
cathode are set to zero. There are several methods in the literature to deal with the electro-
Considering the intermediate reactions taking place in the electro­ osmotic drag coefficient. Despite being influenced by factors such as
lyser and Faraday’s first law of electrolysis, the molar flow rate of the water input pressure, cathode pressure, current density, and tempera­
generated hydrogen at the cathode is described as follows: ture, there seems to be no satisfactory definitive quantitative equation
for determining this coefficient in the literature. For instance, nd has
nI
ṄH2 ca,gen = (15) been considered as a constant parameter, as in [29], a fitting parameter,
2F
as illustrated in [13, 20] or derived via empirical equations, as seen in
By substituting Eq. (15) into Eq. (14), the partial pressures of water and [18, 21]. In this article, the empirical equation as a function of tem­
hydrogen at the cathode can be obtained using the ideal gas law, perature which is obtained in [30] has been utilised, where the authors
expressed as: propose a linear fit of nd as a function of temperature, expressed in K,
based on measured electro-osmotic drag for a solid polymer electrolyte,
NH2 Oca Ru T given as:
pH2 Oca = (16)
vca
nd = − 2.89556 + 0.016T (23)
NH Ru T
pH2 ca = 2 ca (17)
vca Stemming from the concentration gradient between the anode and
cathode sides of the membrane, the flow rate of water through the
where NH2 Oca and NH2 ca are the number of moles of water and hydrogen membrane via diffusion can be derived by integrating Fick’s law of
at the cathode, subsequently and vca is the cathode volume. diffusion between the anode and cathode sides of the membrane in­
The pressure at the cathode is equal to the total partial pressures of terfaces as follows:
water and hydrogen, given as:
ADw ( )
ṄH2 Omem,diff = CH2 Omem,cat − CH2 Omem,an (24)
Pca = pH2 Oca + pH2 ca (18) δmem

Using the partial pressures of the associated substances and the total where Dw denotes the diffusion coefficient, δmem is the membrane
pressure of the cathode, the molar fractions of water and hydrogen at the thickness and CH2 Omem,cat and CH2 Omem,an are the water concentration at the
cathode can be given as: cathode and anode side of the membrane, respectively. The diffusion
pH2 Oca coefficient can be expressed as follows [21]:
XH2 Oca = (19)
Pca

5
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

[ ( )]
1 1
Dw = Dʹw exp 2416 − (25) In Eq. (33) and Eq. (34), the terms Deff,O2 − H2 O and Deff,H2 − H2 O , referred to
303 T as the effective molecular diffusion coefficient for the O2 -H2 O and
⎧ − 10 H2 -H2 O binary systems, reflect the molecular diffusion mechanism. On
⎪ 10 λmem λmem < 2 the other hand, Deff,H2 O− k , known as the effective Knudsen diffusion
⎪
⎨ 10− 10 [1 + 2( λmem − 2)]
⎪
2⩽λmem < 3 coefficient for water, accounts for the Knudsen diffusion effect on the
Dʹw = [ ( )] (26)
⎪ 10− 10 3 − 1.67 λmem − 3
⎪
⎪ 3⩽λmem < 4.5 rate of water flow through the membrane. Additionally, ∊ is electrode
⎩
1.25 × 10− 10 λmem ⩾4.5 porosity and ξ denotes tortuosity.
Regarding the effective Knudsen diffusion coefficient for water,
Here, Dʹw is the diffusion coefficient at the reference temperature and introduced in Eq. (33) and Eq. (34), as the molecules collide with the
walls of the pores frequently, the movement of molecules can be
λmem represents the water content of the membrane. In this model, λmem
modelled using kinetic theory, described as:
is computed in the membrane module using the following temperature-
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅
dependent expression [31]:
4 8Ru T
Deff,H2 O− k = r (35)
λmem = 0.08533T − 6.77632 (27) 3 πMH2 O

It is noteworthy that diffusion through a porous medium, as in the where r is the mean pore radius. On the other hand, to calculate
membrane of the PEM electrolyser, is principally a function of two Deff,O2 − H2 O and Deff,H2 − H2 O , Chapman-Enskog theory of the ideal gas has
mechanisms: molecular diffusion and Knudsen diffusion. When the been utilised, giving:
mean free path of molecules is shorter than the pore size of the mem­ √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
brane, the molecules tend to collide more with each other rather than T 3/2 1 1
Deff,O2 − H2 O = 0.00133 + (36)
with the pore walls, leading to molecular diffusion being more promi­ Pan Θ2O2 − H2 O ΩD MO2 MH2 O
nent than Knudsen diffusion. Conversely, if the pore size is much smaller
with respect to the mean free path of the species, the probability of √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
T 3/2 1 1
molecules colliding with the walls of the pores is higher than with each Deff,H2 − H2 O = 0.00133 + (37)
Pca Θ2H2 − H2 O ΩD
MH2 MH2 O
other. In this case, Knudsen diffusion becomes the dominant term over
molecular diffusion [13]. In this article, both mechanisms are taken into
Here, ΘO2 − H2 O and ΘH2 − H2 O stand for the mean molecular radii of the
consideration.
associated substances, Ω is the dimensionless diffusion collision integral
The water concentration at the anode and cathode sides of the
and MO2 and MH2 denote the molar weight of oxygen and hydrogen,
membrane can be expressed as a function of water concentration in the
respectively.
electrode channels, represented as:
By calculating the remaining parameters, namely ΘO2 − H2 O , ΘH2 − H2 O
δan ṅH2 O,an and ΩD , the flow rate due to diffusion can be calculated. To start with,
CH2 Omem,an = CH2 Och,an − (28)
Deff,an ΘO2 − H2 O and ΘH2 − H2 O can be written as:
ΘO2 + ΘH2 O
δca ṅH2 O,ca ΘO2 − = (38)
CH2 Omem,ca = CH2 Och,ca − (29) H2 O
2
Deff,ca
Θ H2 + Θ H2 O
where CH2 Och,an and CH2 Och,ca are the water concentrations in the anode Θ H2 − H2 O = (39)
2
and cathode channels, δan and δca represent the anode and cathode
electrode thicknesses, ṅH2 O,an and ṅH2 O,ca are the molar fluxes of water On the other hand, ΩD can be expressed analytically as follows:
through the anode and cathode, and Deff,an and Deff,ca stand for the
1.06 0.193 1.036 1.765
effective binary diffusion coefficient at the anode (O2 -H2 O) and cathode ΩD = + + + (40)
τ0.156 exp(0.76τ) exp(1.53τ) 3.894τ
(H2 -H2 O), consecutively. The water concentrations in the anode and
cathode channels can be expressed as follows: where the dimensionless temperature τ can be represented as follows:
ρH2 OT
CH2 Och,an = CH2 Och,ca = (30) kT
MH2 O τ O2 − H2O = (41)
ε O2 − H2 O

where ρH2 OT is water density at temperature T and MH2 O is molar mass of

kT
water. τ H2 − H2O = (42)
εH2 − H2 O
Considering the second terms in Eq. (28) and Eq. (29), water molar
fluxes at the anode and cathode can be written as follows: where εO2 − H2 O and εH2 − H2 O , referred to as Lennard-Jones energies, are
ṄH2 Omem + ṄH2 Oan,cons given as:
ṅH2 O,an = (31)
A √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
εO2 − H2 O = εO2 εH2 O (43)

ṄH2 Oca √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅

(44)
ṅH2 O,ca = (32) εH2 − H2 O = εH2 εH2 O
A
In summary, the process involves substituting Eq. (41–44) into Eq. (40)
The effective binary diffusion coefficients, i.e, Deff,an and Deff,ca as
to derive diffusion collision integral. Thereafter, by inserting Eq.
described in Eq. (28) and Eq. (29), respectively, are represented as
(38–40) into Eq. (36) and Eq. (37) the effective molecular diffusion
follows:
coefficient for the O2 -H2 O and H2 -H2 O binary systems is computed.
( )
1 ∊ 1 1 Next, Eq. (35–37) are substituted into Eq. (33) and Eq. (34) to obtain the
= + (33) effective binary diffusion coefficient at the anode and cathode, facili­
Deff,an ξ Deff,O2 − H2 O Deff,H2 O− k
tating the determination of water concentration at both sides of the
( )
1 ∊ 1 1 membrane expressed in Eq. (28) and Eq. (29). Finally, by substituting
= + (34)
Deff,ca ξ Deff,H2 − H2 O Deff,H2 O− k Eq. (23) into Eq. (22), as well as Eq. (28) and Eq. (29) into Eq. (24), the

6
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

net molar flow rate of water through the membrane due to the molecular coefficients. Hence, in this model, these four parameters are considered
diffusion and Knudsen diffusion can be expressed as follows: as fitting parameters. Apart from that, the exchange current density
⎡⎛ ⎞ values are temperature-dependent which can be expressed as follows:
ADw ⎣⎝ρH2 OT δca ṅH2 O,ca ⎠ [ ( )]
ṄH2 Omem = + Eact 1 1
δmem MH2 O Deff,ca j0,an = j0,an− ref exp − − (52)
Ru T Tref
⎛ ⎞⎤ (45)
ρ H O T
δ an ṅH 2 O,an nd jA [ (
Eact 1 1
)]
− ⎝ 2 − ⎠⎦ +
j0,ca = j0,ca−
MH2 O Deff,an F ref exp − − (53)
Ru T Tref

While Eact is the activation energy of the electrodes, j0,an− ref and j0,ca− ref
2.4. Voltage module indicate the reference exchange current densities of the anode and
cathode, at the reference temperature Tref , respectively.
Ideally, a minimum voltage which is affected by the operating con­ Ohmic overvoltage. In the PEM electrolyser, there are several sources
ditions of the electrolyser, including the pressure and temperature, is contributing to the ohmic overvoltage of the cell, which can be divided
required to initiate the electrolysis process This minimum voltage is into two types: ionic overvoltage, arising from the membrane resistance,
referred to as the reversible or open-circuit voltage. However, in prac­ and electronic overvoltage, attributed to electronic components such as
tice, a higher voltage than reversible voltage must be applied to the bipolar plates, electrodes, etc. Nonetheless, the conductivity of the
electrolyser due to the several voltage drops occurring within the cell. membrane is significantly lower compared to that of electronic parts.
Thus, the total cell voltage comprises the reversible voltage as well as the Therefore, the ionic overvoltage dominates the ohmic overvoltage. In
additional overvoltages, as shown below: this model, similar to many other models, the electronic ohmic over­
voltage is neglected to avoid unnecessary complexity [20, 34, 35].
Vcell = Vrev + Vact + Vohm + Vcon (46)
Since the ionic conduction path in the PEM electrolyser is rather
where Vrev is the reversible cell voltage, Vact denotes the activation short and uniform, the ionic resistance can be simply modelled by
overvoltage, Vohm stands for the ohmic overvoltage and Vcon indicate the assuming constant resistivity, given as [36]:
concentration or diffusion overvoltage. The concentration overvoltage δmem
happens in the case of the accumulation of oxygen and hydrogen in the Rohm,i = (54)
σ mem
catalyst layer of the anode and cathode. Nonetheless, it is insignificant if
the flow field is suitable for gas removal [21]. Furthermore, for current where σmem is the membrane conductivity in S.m− 1 , which can be
densities less than 3 A/cm2 , which is the case in this article, the acti­ described as follows [37]:
vation and ohmic overvoltage are notably higher than concentration [ ( )]
1 1
overvoltage [21, 32]. Thus, it is neglected in this model. σ mem = (0.514λmem − 0.326).exp 1268 − (55)
303 T
Reversible voltage. Using the Nernst equation, the reversible voltage
can be expressed as follows [21]:
Finally, the ohmic overvoltage can be derived using Ohm’s law:
( √̅̅̅̅̅̅̅ )
Ru T pH2 pO2 Vohm = Rohm,i × j (56)
Vrev = E0 + ln (47)
2F pH2 O
2.5. Thermal module
where E0 is the cell reversible voltage in standard conditions. It is worth
mentioning that in Eq. (47), the second term expresses the deviation of By implementing the equations that have been mentioned so far, the
the reversible voltage from its standard value at various operating static response of the PEM electrolyser can be simulated. Although the
conditions. The standard reversible voltage can be empirically described response of the PEM electrolyser is fast, it is not instantaneous and ex­
as follows [13]: hibits a certain degree of dynamics. Hence, considering the dynamic
( ) behaviour of the PEM electrolyser is essential to enhance the accuracy of
E0 = 1.229 − 0.9 × 10− 3 T − 298 (48)
the model. To achieve this, the model incorporates a thermal module to
Activation overvoltage. As the name suggests, the activation over­ account for the temperature variations of the PEM electrolyser during its
voltage describes the required voltage to activate the electrochemical operation. This enables the model to simulate the thermal inertia of the
reactions taking place at the anode and cathode sides within the elec­ system, meaning how fast the electrolyser heats up or cools down in
trolyser. In other words, it accounts for the energy losses arising from the response to the changes in the operating condition. In the thermal
charge transfer kinetics in the electrolysis process. Thus, the activation module, the cell/stack temperature is considered to be homogeneous
overvoltage is composed of two terms each representing the activation and is treated as a lumped thermal capacity, given as:
overvoltage at the anode and cathode. Commonly, the Butler-Volmer ∑
equation is used to express the activation overvoltages, given as [33]: Cl,stack = cp,k vk .ρk (57)
k

Vact = Vact,an + Vact,ca (49)

where Cl,stack is the lumped thermal capacitance of the stack, cp,k rep­
( )
Ru T j resents the specific heat, vk stands for volume and ρk indicates the
Vact,an = arcsinh (50) density of the species k = {H2 O, O2 , H2 }. Taking into account the inlet
αan F 2j0,an
and outlet enthalpy flows in the electrolyser, the stack enthalpy balance
( ) equation can be written as [34]:
Ru T j
Vact,ca = arcsinh (51) dT ∑
αca F 2j0,ca Cl,stack = Qgen + Ṁin,k cp,k Tin − Qloss
dt k
(58)
where αan and αca are charge transfer coefficients, and j0,an and j0,ca stand ∑
− Qcool − Ṁout,k cp,k Tout
for the exchange current density of the anode and cathode, consecu­ k
tively. In the literature, a wide range of values is reported for the ex­
change current densities, and to a lesser extent, for the charge transfer where Qgen indicates the generated heat due to the electrical work done

7
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

in the electrolyser, Ṁin,k and Ṁout,k are the inlet and outlet flow rate of respectively.
the substance k, Qloss is the heat exchanged between the electrolyser and The heat transfer between the electrolyser and the environment, in
environment, Qcool stands for the heat extracted by the coolant while Tin the absence of any insulation, can be expressed as follows:
and Tout represent the inlet and outlet temperatures, successively. Qloss = hAst (T − Tamb ) (67)
Despite Tin being a known parameter, specifically the temperature of the
inlet water, Tout can be determined by solving Eq. (58). Furthermore, in where h is the heat transfer coefficient, Ast represents the outside surface
this model, the operational temperature of the electrolyser is considered area of the stack and Tamb indicates the ambient temperature.
to be the average of the inlet and outlet temperatures. In general, compared to the term Qloss , the other three terms on the
The overall heat generated within the cell must be extracted for right-hand side of Eq. (58) have a more pronounced impact on the cell
several reasons. First of all, it is essential to maintain the cell/stack temperature. Additionally, as water is chosen to be the coolant of the
temperature below a specific threshold, typically 80–90 ◦ C. This value is electrolyser, and is present in a significantly higher quantity compared
primarily dictated by the thermal stability of perfluoro-sulfonic acid to oxygen and hydrogen in the outlet flows, its impact on the cooling of
(PFSA) materials that are employed as polymer electrolytes in the the electrolyser is substantially greater. Thus, neglecting the mentioned
electrolyser. Furthermore, avoiding the formation of temperature gra­ terms and modifying Eq. (58), ΛH2 O can be expressed in terms of the
dients within the cell is of utmost importance. This originates from the operating cell voltage and desired ΔT as follows:
fact that not only temperature gradients within the cell cause an ( )
increased rate of ageing and degradation but also result in non- F ΔH
Λ H2 O = Vcell − (68)
homogeneous loading of the cell which affects the performance of the MH2 O .cp,H2 O .ΔT 2F
electrolyser [38].
Among the cooling methods utilised for the electrolyser, using an For various operating cell voltages, the required number of moles of
excess flow of water for cooling purposes is selected in this model. This water for each mole of the electrolysed water can be depicted, shown in
choice is made due to the several advantages offered by water as a Fig. 2.
coolant, such as the efficient heat transfer between the flowing water
and the electrolyser. To effectively determine the required amount of 3. Results
water needed for the cooling, the inlet and outlet temperature difference
of the electrolyser, denoted as ΔT, is considered. The aim is to keep ΔT In this section, the developed model is validated using the set of
less than 10 to avoid large temperature gradients within the cell due to experimental data reported in [41] and [17]. The choice of these ref­
the mentioned undesired effects. To this end, by introducing the erences for validation is based on the availability of the experimental
parameter ΛH2 O , which is the ratio of the actual water flow through the data under a range of operating conditions. Initially, the static response
electrolyser for every mole of water used for the electrolysis process, the of the model is validated at various constant temperatures, without
required amount of water can be related to the temperature difference in considering the thermal module. The validation is carried out employing
the cell. the static polarisation curves from [41]. Subsequently, the dynamic
The inlet flows in Eq. (58) can be expressed as [39]: response of the model, taking into account the variations in the cell
( ) temperature, is validated using the results reported in [17].
I
Ṁin,H2 O,an = 1 + ΛH2 O MH2 O (59)
2F
3.1. Static model validation
I
Ṁin,H2 O,ca = ΛH2 O MH2 O (60) The specifications of the PEM electrolyser utilised for the validation,
2F
along with the parameters used for the simulation are presented in
Similarly, the outlet flows can be represented as follows: Table 2.
Apart from the fixed model parameters summarised in Table 2, the
I
Ṁout,H2 O,an = Ṁout,H2 O,ca = ΛH2 O MH2 O (61) aforementioned fitting parameters were allowed to vary within their
2F feasible region to ensure that the model aligns with the experimental
I results as closely as possible. Based on the literature, j0,an− ref , j0,ca− ref , αan
Ṁout,O2 ,an = MO2 (62) and αca were varied in the ranges of 10− 11 –10− 4 A/cm2 , 10− 6 –10− 1
4F
A/cm2 , 0.4–2.5, and 0.2–1, respectively. Moreover, the combinations in
I which either j0,an− ref > j0,ca− ref or αan < αca were excluded from the
Ṁout,H2 ,ca = MH2 (63)
2F search space since typical values for the fitting parameters adhere to
these criteria. This arises from the fact that the anode contribution to the
The heat produced due to the work done in the electrolyser can be
activation overvoltage is higher than the cathode. In Table 3, the fitted
expressed as follows:
( )
ΔH
Qgen = Vcell − I (64)
2F

where ΔH is the total reaction enthalpy, which can be described as

follows for the electrolysis process [40]:
( )
1
ΔH T, p = ΔHH2 + ΔHO2 − ΔHH2 O (65)
2

In Eq. (65), the enthalpy of each substance can be calculated as follows:

̂0 + cʹ (T − T )
̂k = ΔH
ΔH (66)
k p 0

where cʹp is the specific heat capacity and ΔH

̂0 is the standard enthalpy
k Fig. 2. The required number of water moles for various operating cell voltages
of the substance k at the pressure and temperature of P0 and T0 , as a function of ΔT.

8
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

Table 2 1.91 × 10− 2 A/cm2 , αan = 2.47 and αca = 0.93 were derived, resulting in
Specifications of the PEM electrolyser used for the static validation [41]. the relative error of 0.55 %. Given the materials in use for the electrodes,
Parameter Value Unit i.e., Pt-Ir for the anode and Pt for the cathode, the obtained values hold
validity [33, 42].
Anode material Pt-Ir -
Cathode material Pt - It is essential to emphasise that the sets of combinations that
Membrane Nafion 1110 - exhibited high errors, such as those exceeding 100%, are specific to the
A 100 cm2 particular electrolyser employed for obtaining the experimental data.
F 96485.33 A.s/mol For a different electrolyser with distinct specifications, these combina­
MH2 0.002 kg/mol
MH2 O 0.018 kg/mol
tions might yield lower errors compared to what they demonstrated in
MO2 0.032 kg/mol this case. In this article, by exploring within the ranges specified in
n 1 - Table 3, it is ensured that the obtained best point, shown with a green
Pan− ca 1 bar star in Fig.3, is indeed the most optimal point.
Ru 8.31 J/K.mol
In Fig. 4, the polarisation curves obtained from the experiment and
Tin 80 ◦
C
δan− ca 0.08 mm model are compared at different temperatures. Evidently, the static
δmem 0.254 mm behaviour of the model corresponds closely with the experimental re­
∊ 0.3 - sults of the PEM electrolyser. The slight discrepancy between the
εH2 /k 59.7 K simulation and experimental curves can be attributed to the neglected
εH2 O /k 809.1 K
εO2 /k 106.7 K
factors in the model, like concentration overvoltage and electronic
ΘH2 2.827 Å
resistance. However, the resulting relative error, depicted in Fig. 5, is
ΘH2 O 2.641 Å entirely insignificant, especially when considering the simulation time
ΘO2 3.467 Å of the mathematical model.
ξ 4 - The average relative error between the model and experiment at 40
◦
C, 60 ◦ C and 80 ◦ C are 0.68%, 0.75%, and 0.55%, consecutively. The
higher relative error observed at 40 ◦ C and 60 ◦ C is believed to arise
Table 3 from the fact that the fitting parameters were fine-tuned at 80 ◦ C to
Fitted parameters for static validation. match the experimental data. Subsequently, these parameters were
maintained unchanged for other temperatures. Thus, the adjusted fitting
Parameter Considered range Fitted value Unit
parameters of the model are not optimal for other operating tempera­
j0,an− 10− 11 –10− 4 2.00 × 10− 8
A/cm2
ref
tures. Additionally, considering Fig. 5, the most substantial relative
j0,ca− 10− 6 –10− 1 1.91 × 10− 2
A/cm2
ref
error is attributed to the current densities lower than 0.1 A/cm2 . This
αan 0.4–2.5 2.47 -
αca 0.1–1 0.93 - stems from the dependency of reversible voltage on the partial pressures
of water, hydrogen and oxygen. During the initial steps of the simula­
tion, the partial pressures take several steps to stabilise, resulting in the
parameters along with their conventional feasible ranges are shown. highest relative error observed at the outset of the simulation. Never­
To illustrate the deviation of the results of the model from the theless, taking 80 ◦ C as an example, the reversible voltage eventually
experimental dataset, the relative error parameter is used. The general converges to 1.29 V after multiple iterations in the simulation. Consid­
expression for the relative error in percentage can be written as follows: ering the enthalpy of water vapour at 80 ◦ C, approximately 239.5
Zexp kJ/mol which corresponds to 1.24 V, the ability of the model to replicate
100 ∑ Vsim,z − Vexp,z the reversible voltage after several iterations is deemed to be acceptable
e= (69)
Zexp z=1 Vexp,z [43].

where Zexp is the number of experiments, and Vsim,z and Vexp,z indicate
3.2. Dynamic model validation
the results obtained from the zth simulation and experiment, respec­
tively.
In this section, the model is validated against the experimental data
The relative error obtained from different combinations of fitting
reported in [17]. Since the dynamic response of the PEM electrolyser is
parameters, irrespective of their sequence of iterations, is depicted in
examined, the thermal module, hence the temperature variations, are
Fig. 3. As 80 ◦ C is a typical operating temperature for the PEM elec­
included in the simulation. The specifications of the PEM electrolyser
trolyser, the curve fitting was implemented at this temperature, and the
employed for the dynamic verification as well as the values of the pa­
polarisation curve with the least average relative error was chosen. As
rameters utilised for the simulation are summarised in Table 4.
can be seen in Table 3 and Fig. 3, j0,an− ref = 2.00 × 10− 8 A/cm2 , j0,ca− ref =
Not unlike the section dedicated to the static validation, the fitting

Fig. 3. The average relative error derived from various combinations of the Fig. 4. Comparison of the polarisation curves obtained from the best point of
fitting parameters for static validation. the model and experiment at various temperatures and atmospheric pressure.

9
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

Fig. 5. The relative error between the polarisation curves obtained from the Fig. 6. The average relative error derived from various combinations of the
best point of the model and experiments at various temperatures and atmo­ fitting parameters for dynamic validation.
spheric pressure.

Table 4
Specifications of the PEM electrolyser used for the dynamic validation [17].
Parameter Value Unit

A 50 cm2
cp,H2 14337 J/kg.K
cp,H2 O 4179 J/kg.K
cp,O2 920.5 J/kg.K
F 96485.33 A.s/mol
MH2 0.002 kg/mol
MH2 O 0.018 kg/mol
MO2 0.032 kg/mol
n 3 -
Pan 1 bar
Fig. 7. Comparison of the stack voltage between the best point obtained from
Pca 10.5 bar
Ru 8.31 J/K.mol the model and experiment with the applied currents of 3, 6 and 10 A.
Tin 20 ◦
C
̂0
ΔH 0 J/mol the model, the case with the applied current of 6 A was chosen for the
H2
̂
ΔH0 -285830 J/mol curve fitting.
H2 O
̂0 0 J/mol The relative error of the obtained results from the simulation is
ΔH
shown in Fig. 8. Considering Table 5 and Fig. 8, j0,an− ref = 3.90 × 10− 6 ,
O2
δan− ca 0.08 mm
δmem 0.30 mm j0,ca− ref = 2.52 × 10− 2 , αan = 0.49 and αca = 0.12 were obtained, leading
∊ 0.3 - to an average relative error of 2.03%. Moreover, the average error for
59.7 K
εH2 /k the applied step currents with amplitudes of 3 and 10 A are 5.21% and
εH2 O /k 809.1 K
εO2 /k 106.7 K
4.55%, respectively.
ΘH2 2.827 Å As the exact operating temperature of the electrolyser in the exper­
ΘH2 O 2.641 Å imental setup is unspecified, in all simulations the inlet temperature is
ΘO2 3.467 Å assumed to be 20 ◦ C. Additionally, the inlet and outlet temperature
ξ 4 - difference of the electrolyser, ΔT, is set to 10. It is noteworthy to
emphasise that these parameters exert a considerable influence on the
performance of the electrolyser, thereby the results of the model. The
parameters were varied so that the results closely aligned with the
effects of ΔT variations under a consistent inlet temperature of 20 ◦ C,
experimental data. Also, the same range of variations were considered
along with the impact of inlet temperature changes at a constant ΔT of
for the parameters. The fitted parameters alongside their evaluated
10 on the performance of the electrolyser, while applying the step cur­
feasible range are tabulated in Table 5.
rent of 6 A, are represented in Fig. 9 and Fig. 10, respectively.
Fig. 6 illustrates the relative error resulting from various combina­
tions of fitting parameters, regardless of their order of iteration.
Following the determination of the fitting parameters, they were held
constant for other scenarios.
The dynamic response of the model to three distinct step current
waveforms characterised by amplitudes of 3, 6, and 10 A, along with the
experimental results are depicted in Fig. 7. In the dynamic validation of

Table 5
Fitted parameters for dynamic validation.
Parameter Considered range Fitted value Unit

j0,an− ref 10− 11 –10− 4 3.90 × 10− 6

A/cm2
j0,ca− ref 10− 6 –10− 1 2.52 × 10− 2
A/cm2
αan 0.4–2.5 0.49 -
Fig. 8. The relative error between the best point obtained from the model and
αca 0.1–1 0.12 –
experiments with the applied currents of 3, 6 and 10 A.

10
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

Conversely, a decrease in ΔT lowers the operating temperature, which in

turn causes an elevation in stack voltage.
Alternatively, in Fig. 10, by holding ΔT unchanged, when the inlet
temperature is raised, the operating temperature of the electrolyser in­
creases, since the operating temperature is the average of the inlet and
outlet temperatures. This results in a decrease in stack voltage. On the
contrary, lowering the inlet temperature while ΔT remains constant,
causes the operating temperature to decrease, leading to a reduction in
stack voltage.
This clearly demonstrates the impact of thermal parameters on the
behaviour of the electrolyser in dynamic operation mode. As a result, the
discrepancy between the results of the model and experiments is pri­
marily attributed to the unspecified thermal parameters. In [17],
different reversible stack voltages observed in the experiments highlight
the dissimilarity of thermal parameters for the explored scenarios.
It is worth noting that, in practical scenarios, the transient behaviour
of the electrolyser changes under different operating conditions. This
variation is clearly reflected in Fig. 9 and Fig. 10 where, for instance, the
regression of temperature vs. time varies across different operating
conditions, as shown in Fig. 9b and Fig. 10b.
Apart from the mentioned reasons, the higher relative error for other
cases than the reference one arises from the fitting parameters not being
optimal for them. Furthermore, given the impact of the thermal pa­
rameters on the performance of the electrolyser, aligning temperature
with its actual value can lead to a reduction in error. Nevertheless, the
model still shows decent accuracy at different operating conditions.

Fig. 9. The impact of varying ΔT on the stack voltage and operating temper­ 4. Conclusion
ature with the applied current of 6 A.
This article presented a simplified yet accurate static-dynamic model
of the PEM electrolyser, enabling its seamless integration into large-
scale simulations of the power system. For instance, it can effectively
contribute to modelling the ancillary service provision of the PEM
electrolyser alongside other units or examining the flexible operation of
the PEM electrolyser in the presence of intermittent RESs, without
introducing excessive complexity. Upon deriving the fitting parameters
for the model, it can be integrated into more complex simulations, such
as those involving a large-scale power system.
The model comprised five distinct modules including the anode,
cathode, membrane, voltage and thermal each utilising physical/
empirical relations to replicate the behaviour of the PEM electrolyser.
Furthermore, water-based cooling for the electrolyser was taken into
account, mirroring the practical operation of electrolysers. As antici­
pated, given the temperature-dependent nature of several mechanisms
taking place in the electrolyser, the thermal characteristic of the elec­
trolyser demonstrated a notable influence on the average operating
temperature, thereby the performance of the electrolyser.
The model, implemented in Matlab/Simulink, underwent validation
against experimental data in both static and dynamic operation modes.
Regarding the former case, the experimental data of a single cell with
100 cm2 active area operating at temperatures of 40, 60, and 80 ◦ C was
employed and the model exhibited an average relative error of 0.68%,
0.75%, and 0.55%, successively. Thus, across the investigated operating
conditions, an overall average relative error of 0.66% was obtained. In
light of the dynamic response of the PEM electrolyser, the experimental
data from a 3-cell stack featuring an active area of 50 cm2 was utilised.
Under applied currents of 3, 6, and 10 A, the developed model showed
an average relative error of 5.21%, 2.03%, and 4.55%, consecutively.
Fig. 10. The impact of varying inlet temperature on the stack voltage and This culminated in an overall average relative error of 3.93% among the
operating temperature. examined cases. Consequently, it can be concluded that the omission of
factors such as concentration overvoltage, electronic resistance, and
In Fig. 9, when maintaining a constant inlet temperature, an increase water passage through the membrane due to the hydraulic pressure ef­
in ΔT results in a subsequent elevation in the operating temperature of fect did not significantly impact the precision of the model in either
the electrolyser, defined as the average of the inlet and outlet temper­ static or dynamic operations. Yet, it did lead to enhanced computational
atures in this model. This, in turn, leads to a reduction in stack voltage. efficiency.

11
S. Asiaban et al. Energy Conversion and Management: X 23 (2024) 100610

Nonetheless, the model still can be improved through the incorpo­ [15] David M, Ocampo-Martínez C, Sánchez-Peña R. Advances in alkaline water
electrolyzers: A review. J Energy Storage 2019;23:392–403.
ration of more physical-based relationships instead of the empirical
[16] Guilbert D, Vitale G. Experimental validation of an equivalent dynamic electrical
ones, for example for the calculation of electro-osmotic drag coefficient. model for a proton exchange membrane electrolyzer. In: in: 2018 IEEE
For future studies, it would be interesting to integrate the model into International Conference on Environment and Electrical Engineering and 2018
more intricate systems, exploring how the dynamic characteristics of the IEEE Industrial and Commercial Power Systems Europe (EEEIC/I&CPS Europe);
2018.
BoP components, like water desalination systems, shape the dynamic [17] Hernández-Gómez Á, Ramirez V, Guilbert D, Saldivar B. Development of an
response of the PEM electrolyser. Moreover, in scenarios where the adaptive static-dynamic electrical model based on input electrical energy for PEM
synthesis of other green molecules is considered, the impact of the dy­ water electrolysis. Int J Hydrogen Energy 2020;45(38):18817–30.
[18] Görgün H. Dynamic modelling of a proton exchange membrane (PEM) electrolyzer.
namic behaviour of the PEM electrolyser on the subsequent systems, Int J Hydrogen Energy 2006;31(1):29–38.
such as the methanol production unit, holds substantial interest. [19] Lee B, Park K, Kim H-M. Dynamic simulation of PEM water electrolysis and
comparison with experiments. Int J Electrochem Sci 2013;8(1):235–48.
[20] Awasthi A, Scott K, Basu S. Dynamic modeling and simulation of a proton exchange
Declaration of Competing Interest membrane electrolyzer for hydrogen production. Int J Hydrogen Energy 2011;36
(22):14779–86.
The authors declare that they have no known competing financial [21] Yigit T, Selamet OF. Mathematical modeling and dynamic simulink simulation of
high-pressure PEM electrolyzer system. Int J Hydrogen Energy 2016;41(32):
interests or personal relationships that could have appeared to influence 13901–14.
the work reported in this paper. [22] García-Valverde R, Espinosa N, Urbina A. Simple PEM water electrolyser model
and experimental validation. Int J Hydrogen Energy 2012;37(2):1927–38.
[23] Kim H, Park M, Lee KS. One-dimensional dynamic modeling of a high-pressure
Data availability
water electrolysis system for hydrogen production. Int J Hydrogen Energy 2013;38
(6):2596–609.
Any utilised data is appropriately cited in the article. [24] Lebbal M, Lecøeuche S. Identification and monitoring of a PEM electrolyser based
on dynamical modelling. Int J Hydrogen Energy 2009;34(14):5992–9.
[25] Aouali FZ, Becherif M, Ramadan HS, Emziane M, Khellaf A, Mohammedi K.
Acknowledgments Analytical modelling and experimental validation of proton exchange membrane
electrolyser for hydrogen production. Int J Hydrogen Energy 2017;42(2):1366–74.
This work was supported by the MuSE project, funded by the Energy [26] Samani AE, D’Amicis A, De Kooning JDM, Bozalakov D, Silva P, Vandevelde L.
Grid balancing with a large-scale electrolyser providing primary reserve. IET
Transition Fund of the Belgian federal government managed by the FPS Renew Power Gener 2020;14(16):3070–8.
Economy. [27] Alex A, Petrone R, Tala-Ighil B, Bozalakov D, Vandevelde L, Gualous H. Optimal
techno-enviro-economic analysis of a hybrid grid connected tidal-wind-hydrogen
energy system. Int J Hydrogen Energy 2022;47(86):36448–64.
References [28] Dadkhah A, Bozalakov D, De Kooning JDM, Vandevelde L. Techno-economic
analysis and optimal operation of a hydrogen refueling station providing frequency
[1] Escamilla A, Sánchez D, García-Rodríguez L. Techno-economic study of Power-to- ancillary services. IEEE Trans Ind Appl 2022;58(4):5171–83.
Power renewable energy storage based on the smart integration of battery, [29] Marangio F, Santarelli M, Calı ̀ M. Theoretical model and experimental analysis of a
hydrogen, and micro gas turbine technologies. Energy Convers Manage: X 2023;18: high pressure PEM water electrolyser for hydrogen production. Int J Hydrogen
100368. Energy 2009;34(3):1143–58.
[2] Asiaban S, Kayedpour N, Samani AE, Bozalakov D, De Kooning JDM, Crevecoeur G, [30] Li X, Qu S, Yu H, Hou M, Shao Z, Yi B. Membrane water-flow rate in electrolyzer
Vandevelde L. Wind and solar intermittency and the associated integration cells with a solid polymer electrolyte (SPE). J Power Sources 2009;190(2):534–7.
challenges: A comprehensive review including the status in the Belgian power [31] Hernández-Gómez Á, Ramirez V, Guilbert D. Investigation of PEM electrolyzer
system. Energies 2021;14(9):2630. modeling: Electrical domain, efficiency, and specific energy consumption. Int J
[3] Onodera H, Delage R, Nakata T. Systematic effects of flexible power-to-X operation Hydrogen Energy 2020;45(29):14625–39.
in a renewable energy system-A case study from japan. Energy Convers Manage: X [32] Majumdar A, Haas M, Elliot I, Nazari S. Control and control-oriented modeling of
2023;20:100416. PEM water electrolyzers: A review. Int J Hydrogen Energy.
[4] Roy A, Olivier J-C, Auger F, Auvity B, Schaeffer E, Bourguet S, Schiebel J, Perret J. [33] Falcão D, Pinto A. A review on PEM electrolyzer modelling: Guidelines for
A combined optimization of the sizing and the energy management of an industrial beginners. J Clean Prod 2020;261:121184.
multi-energy microgrid: Application to a harbour area. Energy Convers Manage: X [34] Olivier P, Bourasseau C, Bouamama PB. Low-temperature electrolysis system
2021;12:100107. modelling: A review. Renew Sustain Energy Rev 2017;78:280–300.
[5] Bartels J, Varela C, Wassermann T, Medjroubi W, Zondervan E. Integration of [35] Espinosa-López M, Darras C, Poggi P, Glises R, Baucour P, Rakotondrainibe A,
water electrolysis facilities in power grids: A case study in northern Germany. Besse S, Serre-Combe P. Modelling and experimental validation of a 46 kW PEM
Energy Convers Manage: X 2022;14:100209. high pressure water electrolyzer. Renew Energy 2018;119:160–73.
[6] Andersson J, Krüger A, Grönkvist S. Methanol as a carrier of hydrogen and carbon [36] Järvinen L, Puranen P, Kosonen A, Ruuskanen V, Ahola J, Kauranen P,
in fossil-free production of direct reduced iron. Energy Convers Manage: X 2020;7: Hehemann M. Automized parametrization of PEM and alkaline water electrolyzer
100051. polarisation curves. Int J Hydrogen Energy 2022;47(75):31985–2003.
[7] Syed MB. Technologies for renewable hydrogen production. Bioenergy Resources [37] Springer TE, Zawodzinski T, Gottesfeld S. Polymer electrolyte fuel cell model.
and Technologies. Elsevier; 2021. p. 157–98. J Electrochem Soc 1991;138(8):2334.
[8] Ti W, Ng DK, Andiappan V. Optimal integration of hydrogen production process [38] Villagra A, Millet P. An analysis of PEM water electrolysis cells operating at
with carbon dioxide capture, utilisation and storage. J Clean Prod 2023:137697. elevated current densities. Int J Hydrogen Energy 2019;44(20):9708–17.
[9] Sebbahi S, Nabil N, Alaoui-Belghiti A, Laasri S, Rachidi S, Hajjaji A. Assessment of [39] Palathingal J. Mathematical modelling and dynamic simulation of a PEM
the three most developed water electrolysis technologies: Alkaline water electrolyser. Master thesis. University of Lorraine; 2017.
electrolysis, proton exchange membrane and solid-oxide electrolysis. Mater Today: [40] Agbli K, Péra M, Hissel D, Rallières O, Turpin C, Doumbia I. Multiphysics
Proc 2022;66:140–5. simulation of a PEM electrolyser: Energetic macroscopic representation approach.
[10] Vincent I, Bessarabov D. Low cost hydrogen production by anion exchange Int J Hydrogen Energy 2011;36(2):1382–98.
membrane electrolysis: A review. Renew Sustain Energy Rev 2018;81:1690–704. [41] Selamet ÖF, Acar MC, Mat MD, Kaplan Y. Effects of operating parameters on the
[11] Lickert T, Kiermaier ML, Bromberger K, Ghinaiya J, Metz S, Fallisch A, Smolinka T. performance of a high-pressure proton exchange membrane electrolyzer. Int J
On the influence of the anodic porous transport layer on PEM electrolysis Energy Res 2013;37(5):457–67.
performance at high current densities. Int J Hydrogen Energy 2020;45(11): [42] Koundi M, El Fadil H, EL Idrissi Z, Lassioui A, Intidam A, Bouanou T, Nady S,
6047–58. Rachid A. Investigation of hydrogen production system-based PEM EL: PEM EL
[12] Zeng Y, Luo M, Qin C, Liu C, Chen B. Investigation on the performance of proton modeling, DC/DC power converter, and controller design approaches. Clean
exchange membrane water electrolyzer coupled with a catalyst layer pore network Technol 2023;5(2):531–68.
model. Energy Convers Manag: X 2024:100523. [43] O’hayre R, Cha S-W, Colella W, Prinz FB. Fuel cell fundamentals. John Wiley &
[13] Abdin Z, Webb C, Gray EM. Modelling and simulation of a proton exchange Sons; 2016.
membrane (PEM) electrolyser cell. Int J Hydrogen Energy 2015;40(39):13243–57.
[14] Bhandari R, Trudewind CA, Zapp P. Life cycle assessment of hydrogen production
via electrolysis–a review. J Clean Prod 2014;85:151–63.

12