i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.elsevier.com/locate/he

Development of an adaptive static-dynamic

electrical model based on input electrical energy for
PEM water electrolysis


Angel Herna
ndez-Go

mez a, Victor Ramirez a,b,*, Damien Guilbert c,
b,d
Belem Saldivar
a
Department of Renewable Energy, Centro de Investigaci
on Cientı́fica de Yucata

n (CICY), Yucata

n, Mexico
b

tedras CONACYT, Ciudad de M
exico, M
exico
Ca
c
Universit
e de Lorraine, GREEN, Nancy F-54000, France
d
Facultad de Ingenierı́a, Universidad Aut
onoma del Estado de M
exico (UAEM), Toluca, M
exico

highlights

Development and validation of a static-dynamic model for PEM electrolyzer voltage.

Analysis of the voltage dynamic behavior in a PEM electrolyzer.

The dynamic behaviors have been modeled through an electronic circuit.

Parameter estimation for a static-dynamic mathematical model.

Simulation and comparison of a mathematical model and PEMelectrolyzer voltage.

article info abstract

Article history: Compared to alkaline electrolyzers, PEM electrolyzers offer high current densities and can
Received 30 January 2020 be coupled to renewable energy sources because of their fast responses to dynamics. The
Received in revised form modeling of PEM electrolyzers is a challenging issue to reproduce its behavior and to design
19 April 2020 properly power electronics and its control without damaging a real electrolyzer. The input
Accepted 20 April 2020 current may have an impact on the dynamics of the electrolyzer and must be taken into
Available online 9 July 2020 consideration to make a model more reliable. In this work, an equivalent electrical circuit
to replicate accurately the dynamic behavior of the PEM electrolyzer subject to fast current
Keywords: change is investigated. Based on the input current, the parameters of the model can not be
PEM electrolyzer considered as constant. Hence, to improve the accuracy of the model, an adaptive static-
Modeling dynamic electrical model is proposed and takes into consideration the change of input
Static current. This model is validated by using a commercial-400 W PEM electrolyzer. The ob-
Dynamic tained results demonstrate the effectiveness of the model to predict the PEM stack voltage.
Renewable energy sources © 2020 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.
Adaptive parameters

* Corresponding author.
E-mail address: victor.ramirez@cicy.mx (V. Ramirez).
https://doi.org/10.1016/j.ijhydene.2020.04.182
0360-3199/© 2020 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.
18818 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0

thermal model. Following this approach, a recent review for

Introduction alternative dynamic models has been presented in Ref. [21],
this work covers systems for alkaline electrolyzer too. How-
Water electrolysis is considered a promising way to generate ever, dynamic models for the electrolyzer voltage are very
clean and efficient hydrogen [1,2], since the development of its scarce; in Ref. [7] a static-dynamic model for voltage based on
technology has allowed its coupling with other renewable the thermodynamic, activation, double-layer, and ohmic ef-
energy sources (RES) [3,4]. The water electrolysis is based on fects has been developed, unfortunately, this model only ap-
the principle of dissociation of water, where two molecules of plies for alkaline electrolyzer. Based on the current state-of-
water are separated into two molecules of hydrogen and one the-art about electrolyzer modeling, it can be noted that the
molecule of oxygen using electricity, this dissociation is an dynamic issues regarding the activation over-potential are not
endothermic reaction. The water electrolysis is performed by taken into consideration. The developed models are mainly
applying a voltage to two electrodes that are submerged in based on the reversible potential and the over-potentials
water, to which an electrolyte of high conductivity is added (activation, ohmic, and concentration) change depending on
[5,6]. the gas pressure, temperature, relative humidity, and mem-
Usually, water electrolysis is carried out by electrolyzers, brane thickness. This modeling allows assessing the different
which are reliable devices that do not require continuous over-potentials according to the operating conditions. These
maintenance since they hardly include mobile elements. Also, models are useful when considering only static conditions.
they are silent and have a high degree of modularity, which However, when considering a PEM electrolyzer coupled to
makes them suitable for decentralized applications in resi- renewable energy sources (e.g. wind energy), the dynamics of
dential, commercial, and industrial areas [7]. Currently, there the electrolyzer must be taken into account to enhance the
are three main types of electrolyzers: Alkaline electrolyzer, reliability of the model. This dynamic model allows predicting
Proton Exchange Membrane (PEM) electrolyzer, and Solid accurately the voltage response of a PEM electrolyzer supplied
Oxide (SO) electrolyzer. by dynamic current profiles. For this reason, the use of a static
In recent years, the PEM electrolyzer has attracted a lot of or dynamic model is guided by the objectives of the study (e.g.
attention from researchers due to its simplicity, high energy static operation, dynamic operations when coupling with
efficiency, compact system design, and specific production renewable energy sources, energy management, the accuracy
capacity [8,9]. Besides, compared to alkaline technology, PEM of the results). For example, in Refs. [22,23], the authors have
technology is particularly fit to be coupled with RES since it chosen to use a static model to investigate the effects of the
offers high flexibility by responding quickly to dynamics as it power supply on the energy efficiency of an alkaline electro-
has been reported in the literature [10,11]. This feature is lyzer. In comparison, in Ref. [24], the authors have used a
important to capture energy during dynamic operations. In dynamic model to analyze the effects of AC-DC converters on
Ref. [12], a survey of the most significant scientific and tech- the specific energy consumption and gas quality of an alkaline
nological materials to build it has been reported. During the electrolyzer during dynamic operations.
operation, the PEM electrolyzer is commonly integrated into Recently, a first work [25] has introduced a dynamic model
the electrical grid including RES to face the intermittent for the activation over-potential assimilating the PEM elec-
electrical production from RES. In the case of surplus energy, trolyzer like an electronic circuit so, the authors have devel-
this energy can be converted into hydrogen and stored in oped a static-dynamic model to model the stack voltage. The
tanks [13]. Compared to classical energy storage devices (i.e. authors have shown that dynamic behavior is strongly linked
batteries, supercapacitors), hydrogen has a higher specific to the input current. However, the parameters used in the
energy, around 142 MJ,kg1 [14]. Therefore, its operation is paper [25] have been adapted just for one specific electrical
inherently intermittent and mathematical models that cap- input current range. As a result, the developed model with
ture its dynamics are required even at the design stage since constant parameters is less reliable for another input current.
they ensure efficient and reliable operation of the electrolysis For this reason, based on this previous work [25], this work
[13,15]. Besides, the PEM electrolyzer mathematical modeling aims at developing a static-dynamic model for PEM electro-
is a powerful tool for simulation, exploring control strategies, lyzer voltage and proposing an algorithm to compute its pa-
prediction, and understanding of the behavior of hydrogen- rameters based on different electrical current inputs. Hence,
generation systems [16,17]. the accuracy of the model in replicating the dynamic behavior
Dynamic models for PEM electrolyzers have already been of the electrolyzer can be improved. To develop the adaptive
reported in the literature [13]. In Ref. [18], the authors have model, experimental tests have been carried out on a com-
developed a model including four subsystems: anode, cath- mercial-400 W PEM electrolyzer. The tests consist of
ode, membrane, and an auxiliary component that models the increasing and decreasing the current with a step of 1 A. This
relation between voltage and electric current. The work [16] methodology allows better modeling of the dynamics of the
has presented the development of a complete model based on electrolyzer.
modules describing the behaviors of the anode, cathode, This work is divided into five sections. After providing the
membrane, and cell voltage. In Ref. [19], an electrochemical current state-of-the-art and motivations to carry out this work
model of the electrolyzer stack to calculate the theoretical in the introduction, PEM electrolyzer technology: features and
open-circuit using thermodynamic analysis has been devel- modeling issues aims at introducing the PEM technology, and
oped. In Ref. [20], the authors have reported a model based on to investigate its behavior for dynamic operations, which is an
thermodynamics and electrochemical equations. This model important issue for modeling purposes. Then, in equivalent
has fitted a steady-state electric model with a dynamic electrical circuit and mathematical modeling, the PEM
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0 18819

equivalent electrical circuit is presented, and its mathemat- for the end-user, thus requiring less energy to further
ical modeling is provided. After that, in parameter estimation, compress and store the hydrogen [26].
the algorithm to compute the parameters of the model based Problems related to higher operational pressures in PEM
on different electrical current inputs is introduced and electrolysis are also present, such as cross-permeation phe-
explained in detail. Finally, in experimental validation of the nomenon which increases with pressure. The corrosive acidic
model, the equivalent electrical model including adaptive regime provided by the proton exchange membrane requires
parameters is validated experimentally by using a commer- the use of distinct materials. These materials must not only
cial-400 W PEM electrolyzer. Besides, a discussion is provided resist the hard corrosive low pH (around 2) condition but also
to summarize and comment on the obtained results. sustain the high applied over-potential (around 2 V), espe-
cially at high current densities. Corrosion resistance applies
not only for the catalysts used but also for current collectors
PEM electrolyzer technology: features and and separator plates. Only a few materials can be selected that
modeling issues would perform in this hard environment [26]. For this reason,
its cost is high, system durability is inferior and technology is
Features of PEM electrolyzer more suitable for low-scale applications compared with con-
ventional alkaline technology [7,9].
The PEM electrolyzer cell is essentially a PEM as an electrolytic In the next subsection, experimental tests have been car-
conductor between an anode and a cathode. The water mol- ried out to emphasize the real behavior of a commercial-400
ecules and ionic particles are transferred across the mem- W PEM electrolyzer during dynamic operations (generally met
brane from the anode to the cathode, where it is decomposed by coupling the electrolyzer with RES which are very dynamic
into oxygen, protons, and electrons. In the reaction process, sources). The obtained results enable demonstrating clearly
electrical energy is supplied to the system and converted into that the dynamics are important issues and have to be taken
chemical energy. The electrons come out of the cell through into consideration for modeling purposes.
an external circuit. After that, the electrons and protons
recombine at the cathode to produce hydrogen [17], see Fig. 1. Experimental test setup
In summary, the reactions in the PEM electrolyzer are:
To investigate the dynamics of a commercial PEM electrolyzer,
1
Anode : H2 Ol / O2g þ 2Hþ
g þ 2e
 an experimental test setup has been realized as shown in Fig.
2
2. The test setup is composed of the following devices and
components: (1) a laptop with a virtual control panel to control
Cathode : 2Hþ þ 2e /H2g
the DC power supply, (2) a DC power supply, (3) a 4-channel
oscilloscope, (4) a pure water tank, (5) a commercial-400 W
1
Overall reaction : H2 Ol /H2g þ O2g PEM electrolyzer, (6) a current probe to acquire the current at
2
the input of the electrolyzer, and finally (7) a voltage probe to
The PEM technology can offer high proton conductivity, acquire the stack voltage of the electrolyzer. The specifica-
low gas crossover, compact system design, and high-pressure tions of the studied commercial PEM electrolyzer NMH2 1000
operation. The low membrane thickness enables minimizing from HELIOCENTRIS ® Company are provided in Table 1. It has
the ohmic losses, and as a result, the energy efficiency can be to be noted that the PEM electrolyzer NMH2 1000 system is
enhanced. PEM electrolyzers can operate at much higher combined with power electronics based on a single-phase
current densities, capable of achieving values above 2 A/cm2, diode rectifier and a DC/DC step-down converter. To investi-
this reduces the operational costs and potentially the overall gate the dynamic behavior of this PEM electrolyzer, only the
cost of electrolysis. The low gas crossover rate of the PEM al- stack is used and is connected to an external DC power supply
lows for the PEM electrolyzer to operate under a wide range of as depicted in Fig. 2. Besides, the PEM electrolyzer uses a solid
power input. The high-pressure operation of an electrolyzer polymer electrolyte based on fluoropolymer Nafion material
brings the advantage of delivering hydrogen at high pressure from DuPont ® company. The thickness of the membrane is
very thin, resulting in lower ohmic losses. To meet the

Fig. 1 e Schematic illustration of a basic water PEM

electrolysis system. Fig. 2 e Experimental test setup.
18820 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0

From Figs. 3 and 4, it can be noticed that the PEM electro-

Table 1 e Specifications of the PEM electrolyzer.
lyzer is able to respond very quickly as a result of dynamic
Parameters Value Unit operations (i.e. rise or fall current). Fast dynamic response is
Rated electrical 400 W one of the most important features requested for electrolyzers
power when coupling with RES. Indeed, since RES are very dynamic
Rated stack 8 V sources strongly depending on weather conditions, PEM
voltage
electrolyzers have to offer a high flexibility to respond to dy-
Stack current 0e50 A
namic operations so that it can absorb the energy during
range
Max. Outlet 10.5 bar transients. The initial and final stack voltage for a current
pressure equal to 0 A, respectively in Figs. 3 and 4, represent the
Cells number 3 e reversible voltage of the PEM electrolyzer (i.e. for a given
Active area 50 cm2 temperature and gas pressure) that is equal around 4.2 V.
section When a step current occurs, a sudden rise in PEM electrolyzer
Hydrogen flow 0e1 SLPM (Standard
stack voltage followed by a slow rise can be observed before
rate at STP Liter Per Minute)
(Standard
reaching the steady-state operation. The sudden voltage stack
Temperature rise is related to the ohmic over-potentials due to the mem-
and Pressure, brane since only the protons can go through it; while the slow
20 C and rise emphasizes the dynamics both at the anode and the
1 bar) cathode (related to the activation over-potentials). The dy-
namics of the PEM electrolyzer are mainly governed by the
input current supplying the electrolyzer. In other words, the
pressure requirements of the storage tanks based on metal
speed of the dynamics strongly depends on the current. Ac-
hydride material, the outlet pressure of the PEM electrolyzer is
cording to its value, the duration of the slow rise in PEM stack
set to 10.5 bar.
voltage may be faster or slower before reaching a steady-state
The next subsection presents the responses of the PEM
value. These dynamics issues are particularly noticeable in
electrolyzer according to the input current.
the obtained experimental results. In the first test (i.e. Fig. 3),
the final steady state-value (around 8 V) is reached roughly in
Dynamic issues
23 s; while for the second test (Fig. 4), it takes 30 s so that the
PEM stack voltage reaches its steady-state operation (around
Based on the experimental test setup presented in the previ-
4.2 V). It has to be noted in Fig. 3 a low voltage overshoot
ous subsection, two dynamic tests have been carried out: one
(around 8.3 V, namely 3.75% of the rated stack voltage), and a
of the tests considers a current rise (from 0 to 10 A); whereas
slow dynamic before reaching the steady-state stack voltage
the second considers a current fall (from 10 to 0 A). The aim is
(around 8 V). This voltage overshoot can be explained since
to show the different dynamic behaviors, which are strongly
the rated stack voltage is reached for an input current equal to
linked with the input energy supplying the electrolyzer. The
7 A. If a step current is applied from an initial current lower
results are shown in Figs. 3 and 4. Since the expected dy-
than 7 A to a final current higher than 7 A, dynamics can be
namics are relatively slow, a high time scale (i.e. 5 s) has been
observed including a voltage overshoot. The voltage over-
tuned to highlight the transient and steady-state operation.
shoots strongly depend on the input current. The higher the

Fig. 3 e Response of the electrolyzer as a result of a rising electrical current from 0 to 10 A.

 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0 18821

Fig. 4 e Response of the electrolyzer as a result of a falling electrical current from 10 to 0 A.

input current, the higher the voltage overshoots. The

maximum voltage overshoot observed in experiments is equal Equivalent electrical circuit and mathematical
to 9 V (around 12.5% of the rated stack voltage). For these tests, modeling
a current probe with a sensitivity (10 mV A-1) and a voltage
probe with a sensitivity (1/20 V) have been used. As a result, As it is widely known, the real cell voltage Vel in a PEM elec-
given that the channel 1 (iel ) and 2 (Vel ) scale are equal trolyzer can be expressed as the sum of the reversible poten-
respectively to 50 mV and 100 mV, the real measured current tial and its over-potentials:
and voltage (taking into consideration the sensitivity of the Vel ¼ Vrev þ hact þ hohm þ hcon (1)
two probes) are equal to 5 A div-1 and 2 V div-1.
In summary, since the dynamics of the PEM electrolyzer where Vrev is the reversible potential, hact ; hohm ; and hcon are
may change according to the operating conditions, the pa- the activation, ohmic and concentration over-potentials.
rameters of the model can not be considered constants. To Eq. (1) represents a static model of the electrolyzer voltage.
develop accurate models and minimize the errors, it is crucial This static model has been used by many authors as reported
to take into account these dynamic issues in the modeling of in the literature [27,28]. However, in Ref. [25], the authors have
the electrolyzer. Therefore, the parameters of the model must assimilated the electrolyzer like an electronic circuit to
be assessed in real-time according to the operating conditions. develop a static-dynamic model. The equivalent electrical
In this work, only the change in input current has been circuit is sketched in Fig. 5.
considered to develop an adaptive static-dynamic electrical The equivalent circuit model is composed of the following
model. Other parameters such as the gas pressure and tem- components:
perature can also be taken into consideration as operating
conditions, but make more complex the development of an - Two resistance-capacitor branches to model the dynamics
adaptive model. Based on experiments, a current range of up both at the anode and cathode (activation over-potential).
to 8 A has been considered to develop an adaptive static-
dynamic electrical model. Indeed, over 8 A, no dynamics
have been reported despite the rated current is equal to 50 A.
Some explanations are provided regarding this issue in
parameter estimation from the author’s point of view.
In the next sections, the dynamic model of the PEM elec-
trolyzer is provided and the methodology to model the dy-
namics of the electrolyzer based on constant step current
change as well. Besides, the algorithm to estimate the
different parameters of the model is given. Finally, the ob-
tained results with the developed model are compared with
the experimental data to validate the effectiveness of the
model in replicating accurately the real static-dynamic
behavior of the electrolyzer. Fig. 5 e Equivalent circuit for a PEM electrolyzer.
18822 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0

model, it is needed to solve Eqs. (3) and (4). So, the solution is
hact ¼ hact;c þ hact;a (2) calculated for Eq. (3) and the solution of Eq. (4) is analogous.
Then, applying the Laplace’s transform to Eq. (3), it yields:
where the dynamic equations for hact;a and hact;c are:
Z ∞
exp½zt,iel ,dt
dhact;c 1 1 L½hact;c 
¼ ,iel  ,hact;c ; (3) z , L½hact;c   hact;c ð0Þ ¼ 0
 (9)
dt C1 t1 C1 t1
where hact;c ð0Þ is the initial value of hact;c . Clearing L½hact;c :
dhact;a 1 1
¼ ,iel  ,hact;a ; (4) Z ∞
dt C2 t2


exp½zt,iel ,dt 
where iel is the electrolyzer current ½A; C1 and C2 are the ca- t1
L½hact;c  ¼ , 0 þ hact;c ð0Þ : (10)
z,t1 þ 1 C1
pacitors for cathode and anode in ½F: The electrical time
constants t1 and t2 in ½s governing the dynamics are variables Therefore, applying the inverse transform:
according to the operating conditions at the input of the Z ∞
electrolyzer.

  exp½zt,iel ,dt  
t1 t
hact;c ¼ , L1 0 þ hact;c ð0Þ,exp : (11)
C1 z,t1 þ 1 t1
- One resistance to model the membrane (ohmic over-
potential). In this work, the current function iel is determined by the
step function:

A1 if t < tc ;
iel ¼ (12)
hohm ¼ iel ,Rint (5) A2 if t  tc ;

where Rint is an internal resistance in the circuit in ½U. where both A1 and A2 are constants in ½A and tc is the time
when the input A1 changes to the input A2 expressed in ½s.
- A DC voltage source to model the reversible potential in ½V. Thus, using Eq. (12) like a function of iel and applying the
second Theorem of translation to Eq. (11), hact;c becomes:

8

>
> t1
>
< ,f1 ðtÞ þ g1 ðtÞ if t < tc;
C1
hact;c ðtÞ ¼
 
   (13)
> t1
> tc  t
>
: , ðA2  A1 Þ, 1  exp þ f1 ðtÞ þ g1 ðtÞ if t  tc;
C1 t1

Vrev ¼ Vint (6)

and for hact;a :

8

>
> t2
>
< ,f2 ðtÞ þ g2 ðtÞ if t < tc;
C2
hact;a ðtÞ ¼
 
   (14)
>
> t2 tc  t
>
: , ðA2  A1 Þ, 1  exp þ f2 ðtÞ þ g2 ðtÞ if t  tc;
C2 t2

It has been reported in Ref. [18,29] that hcon is much lower

than hohm and hact , for this reason, hcon ¼ 0. The equivalent where
resistances R1 and R2 are determined based on the activation
 
 
t t
voltage and input current of the electrolyzer, also they are f1 ðtÞ ¼ A1 , 1  exp ; f2 ðtÞ ¼ A1 , 1  exp (15)
t1 t2
related with the electrical constants t1 and t2 .

 and
hact;c
t1 ¼ C1 , R1 ¼ C1 , ; (7)    
iel t t
g1 ðtÞ ¼ hact;c ð0Þ,exp ; g2 ðtÞ ¼ hact;a ð0Þ,exp : (16)
t1 t2


hact;a
t2 ¼ C2 , R2 ¼ C2 , : (8) So, the equivalent static-dynamic model for the electro-
iel
lyzer voltage can be expressed as:
To have the complete solution of the electrolyzer voltage
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0 18823

operation. For this reason, since the studied PEM electrolyzer

Vel ðtÞ ¼ Vint þ iel ðtÞ,Rint þ hact ðtÞ (17) has been supplied with high current ripple over a long time,
some deteriorations have occurred, which can explain the
obtained static curve in Fig. 6.
Parameter estimation The voltage dynamics have been analyzed by increasing
the input current with a step of 1 A starting from 0 A up to
In this section, the parameters of Eq. (17) are estimated. These achieve 8 A and by decreasing it starting from 8 up to achieve
parameters are; Vint , Rint , hact;c ð0Þ, hact;a ð0Þ, C1 , C2 , R1 , R2 , t1 , and 0 A (keeping the same current step). The obtained results are
t2 . Using the Ohm’s law, hact;c ð0Þ and hact;a ð0Þ are obtained as: shown in Figs. 7 and 8, respectively. On one hand, in Fig. 7, it
can be noted that the voltage dynamics are preponderant
hact;c ð0Þ ¼ R1 ,iel ; hact;a ð0Þ ¼ R2 ,iel : (18) from 0 to 6 A; whereas the voltage dynamics are not significant
from 6 to 8 A. On the other hand, in Fig. 8, it can be observed a
The parameters R1 and R2 can be determined in terms of C1 ,
similar behavior, the voltage dynamics are not meaningful
C2 , t1 , and t2 by Eqs. (7) and (8). To make easier the construc-
from 8 to 6 A; while the dynamics are more significant from 6
tion of an electrical circuit, the capacitors must be equal, so it
to 0 A. Hence, in both tests (increasing and decreasing the
is supposed C1 ¼ C2 . Indeed, this assumption to consider the
input current), the voltage dynamics are preponderant in a
two capacitances equal is motivated by previous works re-
current range from 0 to 6 A, and are not particularly significant
ported in the literature regarding this important issue [30,31].
for a current range including between 6 and 8 A. This obser-
To calculate the remaining five parameters, the behavior of
vation supplements the results obtained with a static char-
the experimental data has been analyzed, both static and
acterization (Fig. 6). Finally, it is important to point out that for
dynamic. First of all, the static voltage-current curve is shown
currents higher than 6 A (Figs. 7 and 8), the stack voltage is
in Fig. 6. From this figure, it can be noted that the PEM stack
lower than 8 V, compared to the obtained voltage for the static
voltage increases and follows the increase in input current.
characterization (Fig. 6). This issue can be explained by the
However, from a current roughly equal to 7 A (i.e. a current
degradations occurred as a result of low and high-frequency
density equal to 0.14 A cm-2), the stack voltage remains con-
current ripple. Besides, these experiments have been carried
stant to 8 V. Even though this phenomenon has not yet been
out after a long period of tests. From the author’s point of
reported in the literature, some explanations can be provided
view, it may be a reason for which the performance (after a
from the author’s point of view. Indeed, as mentioned in PEM
long period of tests) of the electrolyzer is lower compared to
electrolyzer technology: features and modeling issues, the
the obtained results in Fig. 6 for the static characterization
studied PEM electrolyzer system is composed of power elec-
(these measurements have been carried out when the elec-
tronics to meet the requirements (i.e. low DC voltage) to
trolyzer stack was “cold”, meaning that no other tests have
supply the electrolyzer from an AC source (i.e. power grid). For
been performed before). As it can be observed in Figs. 7 and 8,
this reason, the power electronics part is based on a single-
no dynamics are reported, from 6 to 8 A, and from 8 to 6 A,
phase diode rectifier and a DC/DC step-down converter to
since the stack voltage varies slightly. In summary, further
supply the electrolyzer. On one hand, the use of these con-
investigations are required to understand the effects of a long
verters generates low and high-frequency current ripple. The
period of operation on the performance of electrolyzers,
low-frequency current ripple comes from the single-phase
which could be very helpful for modeling purposes.
diode rectifier (i.e. around 100 Hz from an AC signal equal to
After the analysis, a script has been developed by using
50 Hz); whereas the high-frequency current ripple comes from
MATLAB software to calculate the differential Eqs. (3) and (4)
the DC-DC step-down converter (i.e. due to the high switching
to solve Eq. (17). The script and the MATLAB command
frequency of this type of converter; in our case, the frequency
lsqcurvefit (based on the least-square regression algorithm)
is equal to 20 kHz) [32]. On the other hand, many works have
have been applied to adjust the parameters of the model to the
been reported regarding the effects of low and high-frequency
data obtained experimentally for the different tests. The
current ripples on PEM fuel cell stack [32e35], and more
adjusting of the parameters has been performed three times
recently on alkaline electrolyzer from the energy efficiency
to all different tests. In the first adjusting of the parameters,
point of view [22,23]. In these previous works, it has been
the results have shown that the variations of Vint and Rint are
demonstrated that current ripple may impact the life span
not significant. For this reason, Vint and Rint have been esti-
and performance of fuel cell and electrolyzer during their
mated as constants, 4.32 V and 0.329 U respectively, see Figs. 9
and 10.
In the second adjusting, the estimated values found in the
first adjusting have been used, Vint ¼ 4:32V and Rint ¼ 0:329U.
The adjusting has shown the behavior of the parameters t1
and t2 as a Gaussian function, which is a result of the dy-
namics behavior of the voltages tests, Figs. 11 and 12, and also
emphasized a proportional relation between t1 and t2 . So, the
parameter t1 has been estimated as a current function:
" #
2
ðiel  4:2672Þ
t1 ¼ 1:1562 , exp þ 0:606 (19)
0:09487
Fig. 6 e Static voltage-current curve.
18824 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0

Fig. 7 e Behavior of the experimental data as a result of a rising electrical current.

Fig. 8 e Behavior of the experimental data as a result of a falling electrical current.

and the parameter t2 has been estimated like t2 ¼ 0:1, t1 . In

Figs. 5 and 6 can be appreciated the behaviors and estimations  
Fðiel ; t1 Þ ¼ 1:1 , t1 , 6:294  106 , exp½1:6943 , iel  þ 6:3204
of t1 and t2 , respectively.
 0:002 (21)
In the last adjusting, the estimated values obtained in the
first and second adjusting Vint , Rint , t1 and t2 have been reused. and
In this adjusting, the behavior of the parameter C1 has been
Hðiel Þ ¼ 1:7 , arctan½iel  2:5  0:7: (22)
studied, which has shown a dependency of iel and t1 . For this
reason, C1 has been estimated as a function of iel and t1 In Fig. 13 can be appreciated the behavior and estimations
of the parameter C1 .
C1 ðiel ; t1 Þ ¼ Fðiel ; t1 Þ þ Hðiel Þ (20)
In the next section, the parameter estimation of this sec-
where tion are used in the developed model from equivalent
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0 18825

Fig. 9 e Behavior and estimation of the parameter Vint . Fig. 12 e Behavior and estimation of the parameter t2 .

Fig. 10 e Behavior and estimation of the parameter Rint .

Fig. 13 e Behavior and estimation of the parameter C1 .

experimental measurement ½V, Vsim;k is the k simulation data

from the model ½V.
In Fig. 14, a comparison between the developed model and
the experiments is reported. In this comparison, an input
current from 0 to 5 A in a time from 0 to 50 s has been
considered. The comparison between the obtained simulation
data with the experimental data; for the input current from
Fig. 11 e Behavior and estimation of the parameter t1 .
0 to 1 A with a tc ¼ 10s has shown a Er ¼ 1:3% and Em ¼ 0:07V;
for the input current 1e2 A with a tc ¼ 10s has shown a
Er ¼ 0:95% and Em ¼ 0:05V; for the input current 2e3 A with a
tc ¼ 10s has shown a Er ¼ 0:66% and Em ¼ 0:04V; for the input
current 3e4 A with a tc ¼ 10 s has shown a Er ¼ 0:98% and Em ¼
electrical circuit and mathematical modeling to validate it by
0:06V; for the input current 4e5 A with a tc ¼ 10s has shown a
comparing with the experimental test.
Er ¼ 0:88% and Em ¼ 0:06V.
In Fig. 15, the results obtained by simulation and the
experimental data have been compared for a current range
Experimental validation of the model
from 5 to 7 A in a time from 0 to 50 s. The comparison between
the model with the experimental data; for the input current
Simulation and results
5e6 A with a tc ¼ 8s s has shown a Er ¼ 0:8% and Em ¼ 0:06V;
for the input current 6e7 A with a tc ¼ 1s has shown a
First of all, the developed model and its parameters have been
Er ¼ 0:18% and Em ¼ 0:013V.
tested by using the experimental data. From these data, the
On the other side, in Fig. 16, the model and the experiments
PEM electrolyzer stack voltage can be obtained by simulations.
have been compared with an input current from 8 to 5 A in a
Besides, to assess the effectiveness of the model in repro-
time between 0 and 50 s. The comparison between the model
ducing the real behavior of the electrolyzer, the relative error
and the experiments; for the input current 8 to 7 A with a tc ¼
Er and the mean error Em have been calculated for all different
17:5s has shown a Er ¼ 0:66% and Em ¼ 0:05V; for the input
experimental tests as follows:
current 7 to 6 A with a tc ¼ 12:5s has shown a Er ¼ 1:42% and

 Nd P Nd Em ¼ 0:1 V; for the input current 6 to 5 A with a tc ¼ 20s has
100 X Vexp;k  Vsim;k Vexp;k  Vsim;k
Er ¼ , ; Em ¼ k¼1 shown a Er ¼ 0:64% and Em ¼ 0:04 V.
Nd k¼1
Vexp;k Nd
In Fig. 17, the model has been compared with the experi-
(23)
ment data for a current range from 5 to 0 A in a time from 0 to
where Nd is the number of experimental data, Vexp;k is the k 50 s. The simulation comparation with the experimental data;
18826 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0

Fig. 14 e Comparison of the stack voltage between the experiments (solid line) and the model (dashed line) considering a
current range from 0 to 5 A.

Fig. 15 e Comparison of the stack voltage between the experiments (solid line) and the model (dashed line) considering a
current range from 5 to 7 A.

for the input current 5 to 4 A with a tc ¼ 15s has shown a Er ¼ In Fig. 18, simulations results are reported in a time from
1:29% and Em ¼ 0:08V; for the input current 4 to 3 A with a tc ¼ 0 to 50 s. The comparison between the model and the exper-
15s has shown a Er ¼ 1:64% and Em ¼ 0:1V; for the input cur- iments; for the input current 0e3 A with a tc ¼ 4:5s has shown
rent 3 to 2 A with a tc ¼ 15s has shown a Er ¼ 1:47% and Em ¼ a Er ¼ 4:2% and Em ¼ 0:2V; for the input current 0e6 A with a
0:08V; for the input current 2 to 1 A with a tc ¼ 15s has shown a tc ¼ 11s has shown a Er ¼ 3:95% and Em ¼ 0:25V; for the input
Er ¼ 1:93% and Em ¼ 0:1V; for the input current 1 to 0 A with a current 0e10 A with a tc ¼ 2s has shown a Er ¼ 3:9% and Em ¼
tc ¼ 20s has shown a Er ¼ 2:7% and Em ¼ 0:13V. 0:29V.
Finally, the mathematical model has been compared with In Fig. 19, the model is tested for the last current steps in a
the experiments obtained for different current steps from 0 to time from 0 to 50 s. The comparison of the model with the
3 A, 0e6 A, 0e10 A, 10 to 0 A, and 2e10 A. The results are re- experimental data; for the input current 10 to 0 A with a tc ¼
ported in Figs. 18 and 19. The aim is to estimate the accuracy 4s has shown a Er ¼ 2:52% and Em ¼ 0:12V; for the input
of the model in replicating the PEM stack voltage for different current 2e10 A with a tc ¼ 11s has shown a Er ¼ 4:19% and
current steps, different from those used to develop the model. Em ¼ 0:29V.
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0 18827

Fig. 16 e Comparison of the stack voltage between the experiments (solid line) and the model (dashed line) considering a
current range from 8 to 5 A.

Fig. 17 e Comparison of the stack voltage between the experiments (solid line) and the model (dashed line) considering a
current range from 5 to 0 A.

Discussion between the experimental data and the model can be

observed. Despite this, the model can provide a good
Based on the results reported in the previous subsection, in approximation of the real behavior of the electrolyzer. Be-
Table 2, a summary is provided to summarize the relative and sides, the obtained errors are particularly noticeable during
mean error of the model according to different tests carried the transient phenomena, which are complex to model ac-
out. From this table, it can be noted that the maximum mean cording to the input current. To conclude, these obtained re-
error obtained with the model is equal to 0.29 V (i.e. 4.19%) for sults demonstrate the effectiveness of the model to reproduce
a current step from 2 to 10 A; whereas for the other test (Figs. the real behavior of the PEM electrolyzer while reducing the
14e17), the mean error is very low (less than 0.1 V). The errors between the model and the experiments. Further
maximum errors are obtained for larger current steps (Figs. 18 improvement can be made to increase the reliability of the
and 19) since these experimental data have not been directly model in predicting the dynamic behavior of the electrolyzer
used to develop the adaptive model. Hence, the estimation of during dynamic operations.
the parameters of the model is not optimal, and errors
18828 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0

Fig. 18 e Comparison of the stack voltage between the experiments (solid line) and the model (dashed line) considering a
current range from 0 to 3 A, 0e6 A, and 0e10 A.

Fig. 19 e Comparison of the stack voltage between the experiments (solid line) and the model (dashed line) considering a
current range from 10 to 0, and 2e10 A.

as functions of the input current, which supplement a previ-

Conclusion ous work, where the authors have demonstrated that the
equivalent capacitor is equal to 37 F for a specific input
In this work, it has been investigated the dynamic behavior of current.
voltage in a PEM electrolyzer to develop an accurate static- The model accuracy has been validated through different
dynamic mathematical model. The experimental tests have experimental tests achieving an average relative error of
demonstrated the presence of dynamics for input current 1.45%. Besides, the model has been tested for other current
steps from 0 to 8 A and no dynamics for input currents higher steps to estimate accurately its performance. The obtained
than 8 A. The mathematical model has been developed using results have shown that the model offers good accuracy with a
the function input current step according to the experimental maximum error of around 4.19% in replicating the real
tests. To complete the model, the parameters have been behavior of a PEM electrolyzer. However, for future studies, it
estimated Vint and Rint as constants, and the other parameters
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0 18829

Academic Press; 2018. p. 17e62. https://doi.org/10.1016/B978-

Table 2 e Summary of the performance of the model. 0-12-811250-2.00003-0.
Input current tc Er Em Simulation [6] Krishnan S, Fairlie M, Andres P, de Groot T, Kramer GJ.
Chapter 10 - power to gas (h2): alkaline electrolysis. In:
0e1 A 10 s 1.3% 0.07 V Fig. 14
Junginger M, Louwen A, editors. Technological learning in
1e2 A 10 s 0.95% 0.05 V
the transition to a low-carbon energy system. Academic
2e3 A 10 s 0.66% 0.04 V
Press; 2020. p. 165e87. https://doi.org/10.1016/B978-0-12-
3e4 A 10 s 0.98% 0.06 V
818762-3.00010-8.
4e5 A 10 s 0.88% 0.06 V
[7] Ursúa A, Gandı́a LM, Sanchis P. Hydrogen production from
5e6 A 8s 0.8% 0.06 V Fig. 15
water electrolysis: current status and future trends. IEEE
6e7 A 1s 0.18% 0.01 V
2012;100(2):410e26. https://doi.org/10.1109/
8 to 7 A 17.5 s 0.66% 0.05 V Fig. 16
JPROC.2011.2156750.
7 to 6 A 12.5 s 1.42% 0.01 V
[8] Stempien JP, Sun Q, Chan SH. Solid oxide electrolyzer cell
6 to 5 A 20 s 0.64% 0.04 V
modeling: a review. Journal of Power Technologies
5 to 4 A 15 s 1.29% 0.08 V Fig. 17
2013;93(4):216e46.
4 to 3 A 15 s 1.64% 0.1 V
[9] Ruuskanen V, Koponen J, Sillanpa €a€ T, Huoman K, Kosonen A,
3 to 2 A 15 s 1.47% 0.08 V ~ M, Ahola J. Design and implementation of a power-
NiemelA¤
2 to 1 A 15 s 1.93% 0.1 V
hardware-in-loop simulator for water electrolysis emulation.
1 to 0 A 20 s 2.7% 0.13 V
Renew Energy 2018;119:106e15. https://doi.org/10.1016/
0e3 A 4.5 s 4.2% 0.2 V Fig. 18
j.renene.2017.11.088.
0e6 A 11 s 3.95% 0.25 V
[10] Sharifian S, Kolur NA, Harasek M. Transient simulation and
0e10 A 2s 3.9% 0.29 V
modeling of photovoltaic-pem water electrolysis. Energy
10 to 0 A 4s 2.52% 0.12 V Fig. 19
Sources, Part A Recovery, Util Environ Eff
2e10 A 11 s 4.19% 0.29 V
2020;42(9):1097e107. https://doi.org/10.1080/
15567036.2019.1602220.
[11] Sharifian S, Harasek M. Dynamic simulation of hydrogen
would be interesting to test the model under different input generation from renewable energy sources. Chemical
signals like straight, sine or cosine, and triangular (generally Engineering Transactions 2015;45:409e14. https://doi.org/
met with power electronics). 10.3303/CET1545069.
With the mathematical model developed in this work, [12] Millet P, Mbemba N, Grigoriev SA, Fateev VN, Aukauloo A,
Etie
vant C. Electrochemical performances of PEM water
different mathematical tools can be applied as control theory,
electrolysis cells and perspectives. Int J Hydrogen Energy
which is useful to improve the performance and optimize the
2011;36(6):4134e42. https://doi.org/10.1016/
electrolyzer. j.ijhydene.2010.06.105.
[13] Corengia M, Torres AI. Two-phase dynamic model for pem
electrolyzer. In: Eden MR, Ierapetritou MG, Towler GP,
Appendix A. Supplementary data editors. 13th international symposium on process systems
engineering (PSE 2018), vol. 44. of Computer Aided Chemical
Engineering, Elsevier; 2018. p. 1435e40. https://doi.org/
Supplementary data to this article can be found online at
10.1016/B978-0-444-64241-7.50234-2.
https://doi.org/10.1016/j.ijhydene.2020.04.182.
[14] Abe J, Popoola A, Ajenifuja E, Popoola O. Hydrogen energy,
economy and storage: review and recommendation. Int J
Hydrogen Energy 2019;44(29):15072e86. https://doi.org/
references 10.1016/j.ijhydene.2019.04.068.
[15] Nie J, Chen Y, Cohen S, Carter BD, Boehm RF. Numerical and
experimental study of three-dimensional fluid flow in the
[1] Bessarabov D, Millet P. Chapter 2 - brief historical bipolar plate of a PEM electrolysis cell. Int J Therm Sci
background of water electrolysis. In: Bessarabov D, Millet P, 2009;48(10):1914e22. https://doi.org/10.1016/
editors. PEM water electrolysis, hydrogen energy and fuel j.ijthermalsci.2009.02.017.
cells primers. Academic Press; 2018. p. 17e42. https://doi.org/ [16] Abdin Z, Webb CJ, Gray EM. Modelling and simulation of a
10.1016/B978-0-12-811145-1.00002-2. proton exchange membrane (PEM) electrolyser cell. Int J
[2] Coutanceau C, Baranton S, Audichon T. Chapter 1 - hydrogen Hydrogen Energy 2015;40(39):13243e57. https://doi.org/
electrochemical production. In: Coutanceau C, Baranton S, 10.1016/j.ijhydene.2015.07.129.
Audichon T, editors. Hydrogen electrochemical production, [17] Rahim AHA, Tijani AS, Kamarudin SK, Hanapi S. An overview
hydrogen energy and fuel cells primers. Academic Press; of polymer electrolyte membrane electrolyzer for hydrogen
2018. p. 1e6. https://doi.org/10.1016/B978-0-12-811250- production: modeling and mass transport. J Power Sources
2.00001-7. 2016;309:56e65. https://doi.org/10.1016/
[3] Abbasi T, Abbasi S. ’Renewable’ hydrogen: prospects and j.jpowsour.2016.01.012.
challenges. Renew Sustain Energy Rev 2011;15(6):3034e40. [18] Go € rgün H. Dynamic modelling of a proton exchange
https://doi.org/10.1016/j.rser.2011.02.026. membrane (PEM) electrolyzer. Int J Hydrogen Energy
[4] Mohammadi A, Mehrpooya M. A comprehensive review on 2006;31(1):29e38. https://doi.org/10.1016/
coupling different types of electrolyzer to renewable energy j.ijhydene.2005.04.001.
sources. Energy 2018;158:632e55. https://doi.org/10.1016/ [19] Marangio F, Santarelli M, Calı̀ M. Theoretical model and
j.energy.2018.06.073. experimental analysis of a high pressure PEM water
[5] Coutanceau C, Baranton S, Audichon T. Chapter 3 - hydrogen electrolyser for hydrogen production. Int J Hydrogen Energy
production from water electrolysis. In: Coutanceau C, 2009;34(3):1143e58. https://doi.org/10.1016/
Baranton S, Audichon T, editors. Hydrogen electrochemical j.ijhydene.2008.11.083.
production, hydrogen energy and fuel cells primers.
18830 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 1 8 8 1 7 e1 8 8 3 0

[20] Lebbal ME, Lecoeuche S. Identification and monitoring of a power-to-gas plants. Appl Energy 2018;218:192e8. https://
PEM electrolyser based on dynamical modelling. Int J doi.org/10.1016/j.apenergy.2018.02.155.
Hydrogen Energy 2009;34(14):5992e9. https://doi.org/10.1016/ [28] Schalenbach M, Carmo M, Fritz DL, Mergel J, Stolten D.
j.ijhydene.2009.02.003. Pressurized PEM water electrolysis: efficiency and gas
[21] Olivier P, Bourasseau C, Bouamama PB. Low-temperature crossover. Int J Hydrogen Energy 2013;38(35):14921e33.
electrolysis system modelling: a review. Renew Sustain https://doi.org/10.1016/j.ijhydene.2013.09.013.
Energy Rev 2017;78:280e300. https://doi.org/10.1016/ [29] Garcı́a-Valverde R, Espinosa N, Urbina A. Simple PEM water
j.rser.2017.03.099. electrolyser model and experimental validation. Int J
[22] Ursúa A, Marroyo L, Gubı́a E, Gandı́a LM, Die
guez PM, Hydrogen Energy 2012;37(2):1927e38. https://doi.org/10.1016/
Sanchis P. Influence of the power supply on the energy j.ijhydene.2011.09.027.
efficiency of an alkaline water electrolyser. Int J Hydrogen [30] Guilbert D, Vitale G. Dynamic emulation of a pem
Energy 2009;34(8):3221e33. https://doi.org/10.1016/ electrolyzer by time constant based exponential model.
j.ijhydene.2009.02.017. Energies 2019;12(4):750. https://doi.org/10.3390/en12040750.
[23] Dobo
Z, Palota
a

s Arp
d Bence. Impact of the voltage [31] ~ S, Noiying P, Nguyen DA, Davat B. An
Hinaje M, RaA«l
fluctuation of the power supply on the efficiency of alkaline equivalent electrical circuit model of proton exchange
water electrolysis. Int J Hydrogen Energy membrane fuel cells based on mathematical modelling.
2016;41(28):11849e56. https://doi.org/10.1016/ Energies 2012;5(8):2724e44. https://doi.org/10.3390/
j.ijhydene.2016.05.141. en5082724.
[24] Speckmann F-W, Bintz S, Birke KP. Influence of rectifiers on [32] Fontes G, Turpin C, Astier S, Meynard TA. Interactions
the energy demand and gas quality of alkaline electrolysis between fuel cells and power converters: influence of current
systems in dynamic operation. Appl Energy 2019;250:855e63. harmonics on a fuel cell stack. IEEE Trans Power Electron
https://doi.org/10.1016/j.apenergy.2019.05.014. 2007;22(2):670e8. https://doi.org/10.1109/TPEL.2006.890008.
[25] Guilbert D, Vitale G. Experimental validation of an equivalent [33] Wahdame B, Girardot L, Hissel D, Harel F, Francois X,
dynamic electrical model for a proton exchange membrane Candusso D, Pera MC, Dumercy L. Impact of power converter
electrolyzer. In: 2018 IEEE international conference on current ripple on the durability of a fuel cell stack. In: 2008
environment and electrical engineering and 2018 IEEE IEEE international symposium on industrial electronics;
industrial and commercial power systems europe (EEEIC/ 2008. p. 1495e500. https://doi.org/10.1109/ISIE.2008.4677206.
I&CPS europe); 2018. p. 1e6. https://doi.org/10.1109/ [34] Gemmen RS. Analysis for the effect of inverter ripple current
EEEIC.2018.8494523. on fuel cell operating condition. J Fluid Eng
[26] Carmo M, Fritz DL, Mergel J, Stolten D. A comprehensive 2003;125(3):576e85. https://doi.org/10.1115/1.1567307.
review on PEM water electrolysis. Int J Hydrogen Energy [35] Zhan Y, Guo Y, Zhu J, Liang B, Yang B. Comprehensive
2013;38(12):4901e34. https://doi.org/10.1016/ influences measurement and analysis of power converter
j.ijhydene.2013.01.151. low frequency current ripple on pem fuel cell. Int J Hydrogen
[27] Tjarks G, Gibelhaus A, Lanzerath F, Müller M, Bardow A, Energy 2019;44(59):31352e9. https://doi.org/10.1016/
Stolten D. Energetically-optimal PEM electrolyzer pressure in j.ijhydene.2019.09.231.