International Journal of Hydrogen Energy 28 (2003) 21 – 33

www.elsevier.com/locate/ijhydene

Modeling of advanced alkaline electrolyzers: a system

simulation approach
Iystein Ulleberg 1
Institute for Energy Technology, P.O. Box 40, N-2027 Kjeller, Norway

Abstract
A mathematical model for an advanced alkaline electrolyzer has been developed. The model is based on a combination
of fundamental thermodynamics, heat transfer theory, and empirical electrochemical relationships. A dynamic thermal model
has also been developed. Comparisons between predicted and measured data show that the model can be used to predict the
cell voltage, hydrogen production, e.ciencies, and operating temperature. The reference system used was the stand-alone
photovoltaic-hydrogen energy plant in J0ulich. The number of required parameters has been reduced to a minimum to make
the model suitable for use in integrated hydrogen energy system simulations. The model has been made compatible to a
transient system simulation program, which makes it possible to integrate hydrogen energy component models with a standard
library of thermal and electrical renewable energy components. Hence, the model can be particularly useful for (1) system
design (or redesign) and (2) optimization of control strategies. To illustrate the applicability of the model, a 1-year simulation
of a photovoltaic-hydrogen system was performed. The results show that improved electrolyzer operating strategies can be
identi2ed with the developed system simulation model. ? 2002 International Association for Hydrogen Energy. Published by
Elsevier Science Ltd. All rights reserved.

Keywords: Alkaline electrolyzer; Hydrogen systems; Stand-alone power; Renewable energy; Modeling; System simulation

1. Introduction project is the hydrogen refueling station at Munich Airport

[2]. Most of the previous RE=H2 -projects have been based
1.1. Background on solar energy from photovoltaics (PV). However, lately
also wind energy conversion systems (WECS) have been
Hydrogen is often referred to as the energy carrier of the considered to be a possible power source, particularly for
future because it can be used to store intermittent renewable weak-grid applications.
energy (RE) sources such as solar and wind energy. The idea In all of the cases mentioned above the electrolyzer is a
of creating sustainable energy systems lead over the past crucial component, and the technical challenge is to make
decade to several hydrogen energy demonstration projects it operate smoothly with intermittent power from renewable
around the world [1]. The main objectives of these hydrogen energy sources. Up until now most of the R&D on water
projects was to test and develop components, demonstrate electrolysis related to RE=H2 -projects have focused on alka-
technology, and perform system studies on two categories of line systems, although there have been some major research
systems: (1) stand-alone power systems and (2) hydrogen eAorts on proton exchange membrane (PEM) electrolyzers
refueling stations. In the latter category, the most notable as well, particularly within the Japanese WE-NET program
[3]. However, the costs associated with PEM-electrolysis are
still too high, and the market for small-scale H2 -production
1 Temporary address until 31.12.2002: Murdoch University, units is at present day still relatively small.
South Street, WA, Perth 6150, Australia. Institute for Energy Technology (IFE) has since the early
E-mail address: [email protected] (I. Ulleberg). 1990s been carrying out theoretical and practical research

0360-3199/02/$ 22.00 ? 2002 International Association for Hydrogen Energy. Published by Elsevier Science Ltd. All rights reserved.
PII: S 0 3 6 0 - 3 1 9 9 ( 0 2 ) 0 0 0 3 3 - 2
22 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33

Nomenclature
◦
Acronyms LMTD log mean temperature diAerence, C
AC alternating current nc number of cells in series per stack
DC direct current p pressure, bar
EES Engineering Equation Solver r parameter related to ohmic resistance of
FZJ ForschungsZentrum J0ulich electrolyte, M m2
HYSOLAR HYdrogen SOLAR Rt overall thermal resistance of electrolyzer,
IEA International Energy Agency W−1 K
IFE Institute for Energy Technology s coe.cient for overvoltage on electrodes, V
JANAF Joint Army-Navy-Air Force (database t coe.cient for overvoltage on electrodes,
for thermochemical properties) A−1 m2
KOH potassium hydroxide SOC state of charge (battery), 0 : : : 1
◦
MPPT maximum power point tracker T temperature, K or C
PEM proton exchange membrane U voltage, V
PHOEBUS PHOtovoltaik-Elektrolyse-BrennstoAzelle UAHX overall heat transfer coe.cient-area prod-
Und Systemtecknik uct for heat exchanger, W−1 K
PV photovoltaic OG change in Gibbs energy, J mol−1
R&D research and development OH change in enthalpy, J mol−1
RE renewable energy OS change in entropy, J K −1 mol−1
RMS root mean square ṅ molar Row rate, mol=s
SAPS stand-alone power system Q̇ heat transfer rate, W
SIMELINT SIMulation of Electrolyzers in INTer- Ot time interval, s
mittent operation
TRNSYS TRaNsient SYstem Simulation program Subscripts
WECS wind energy conversion system
WE-NET World Energy-Network a ambient
cool cooling (auxiliary)
Symbols cw cooling water
gen generated
A area of electrode, m2 H2 pure hydrogen
aq water based solution H2 O pure water
Ccw thermal capacity of cooling water, J K −1 i, o inlet, outlet
Ct overall thermal capacity of electrolyzer, ini initial
J K −1 loss loss to ambient
emf electromotive force, V O2 pure oxygen
f1 parameter related to Faraday e.ciency, rev reversible
mA2 cm−4 Constants
f2 parameter related to Faraday e.ciency
g gas F 96 485 C mol−1 or As mol−1 Faraday con-
hcond parameter related to conduction heat stant
transfer, W K −1 z 2 number of electrons transferred per reac-
hconv parameter related to convection heat tion
transfer, W K −1 A−1 R 8:315 J K −1 mol−1 universal gas constant
I current, A vstd 0:0224136 m3 mol−1 volume of an ideal
l liquid gas at standard conditions

in the area of stand-alone power systems (SAPS) based is being reported in this paper. However, it should be noted
on RE sources and H2 -technology [4 – 6], and joined in that IFE is currently in the process of acquiring a small-scale
1999 the IEA Hydrogen Program Annex 13 [7]. The elec- PEM-electrolyzer unit for testing in a laboratory setup,
trolyzer modeling eAorts performed in this context focused which hopefully will give valuable system performance data
on alkaline electrolysis, as this was the technology chosen over the next 2–3 years. The theory and modeling philoso-
for the relevant applications. It is this modeling eAort that phy presented here could be applied to the PEM-technology.
 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33 23

Fig. 2. Principle of a bipolar electrolyzer design.

Fig. 1. Principle of a monopolar electrolyzer design.

pressures of the bipolar electrolyzers require relatively so-

Over the past decade there has been an increasing interest phisticated and complex system designs, and consequently
in system analysis of integrated RE=H2 -systems, especially increases the manufacturing costs. The relatively simple and
among those energy and utility companies that are trying sturdy monopolar electrolyzers systems are in comparison
to position themselves in the future markets of distributed less costly to manufacture. Nevertheless, most commercial
power generation and alternative fuels. Therefore, there is a alkaline electrolyzers manufactured today are bipolar.
need for an electrolyzer model that is suitable for dynamic In new advanced alkaline electrolyzers the operational
simulation of such systems. There is also a need for a model cell voltage has been reduced and the current density in-
with a relatively high level of detail. One particularly im- creased compared to the more conventional electrolyzers.
portant requirement is that the technical model can be cou- Reducing the cell voltage reduces the unit cost of electrical
pled to economic models that account for both investment power and thereby the operation costs, while increasing the
and operational costs. current density reduces the investment costs [8]. However,
there is a conRict of interest here because the ohmic resis-
1.2. Technology tance in the electrolyte increases with increasing current due
to increasing gas bubbling. Increased current densities also
The electrolyte used in the conventional alkaline water lead to increased overpotentials at the anodes and cathodes.
electrolyzers has traditionally been aqueous potassium Three basic improvements can be implemented in the de-
hydroxide (KOH), mostly with solutions of 20 –30 wt% be- sign of advanced alkaline electrolyzers: (1) new cell con-
cause of the optimal conductivity and remarkable corrosion 2gurations to reduce the surface-speci2c cell resistance de-
resistance of stainless steel in this concentration range [8]. spite increased current densities (e.g., zero-gap cells and
The typical operating temperatures and pressures of these low-resistance diaphragms), (2) higher process tempera-
◦
◦
electrolyzers are 70 –100 C and 1–30 bar, respectively. tures (up to 160 C) to reduce the electric cell resistance in
Physically an electrolyzer stack consists of several cells order to increase the electric conductivity of the electrolyte,
linked in series. Two distinct cell designs does exist: and (3) new electrocatalysts to reduce anodic and cathodic
monopolar and bipolar [9]. In the monopolar design the overpotentials (e.g., mixed-metal coating containing cobalt
electrodes are either negative or positive with parallel elec- oxide at anode and Raney-nickel coatings at cathode). In
trical connection of the individual cells (Fig. 1), while in the zero-gap cell design the electrode materials are pressed
the bipolar design the individual cells are linked in series on either side of the diaphragm so that the hydrogen and
electrically and geometrically (Fig. 2). One advantage of oxygen gases are forced to leave the electrodes at the rear.
the bipolar electrolyzer stacks is that they are more compact Most manufacturers have adopted this design [9].
than monopolar systems. The advantage of the compactness
of the bipolar cell design is that it gives shorter current 1.3. Modeling
paths in the electrical wires and electrodes. This reduces the
losses due to internal ohmic resistance of the electrolyte, Most of the relevant electrolyzer modeling found in the
and therefore increases the electrolyzer e.ciency. How- literature is related to solar-hydrogen demonstration projects
ever, there are also some disadvantages with bipolar cells. from the past decade. The most detailed model to date
One example is the parasitic currents that can cause cor- is probably the SIMELINT-program, developed as part of
rosion problems. Furthermore, the compactness and high the Saudi Arabian–German HYSOLAR-project [10]. This
24 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33

program, which was validated against measured data, accu-

rately predicts the thermal behavior, cell voltage, gas puri-
ties, and e.ciencies for any given power or current pro2le.
Other empirical models have also been developed [11–15],
but these have either been less detailed or not tested and
veri2ed against experimental data.
The objective of the work described in this paper has
been to develop a model that accurately predicts the elec-
trochemical and thermal dynamic behavior of an advanced
alkaline electrolyzer. The model is primarily intended for
use in integrated renewable energy systems simulations stud-
ies that comprise subsystems such as PV-arrays, WECS,
electrolyzers, fuel cells, and hydrogen storage. A few key
requirements were placed upon the model; it needed to be
numerically robust, versatile and practical to use. Hence, the
model needed to be a trade-oA between simple and com-
plex modeling. For instance, empirical relations are used to
model the most complex electrochemical processes. At the Fig. 3. Operation principle of alkaline water electrolysis.
same time, a signi2cant eAort has been made to minimize
the number of required parameters required by the empirical
relations. In order to make the model as generic as possi- water is
ble, fundamental thermodynamics and heat transfer theory H2 O(l) + electrical energy → H2 (g) + 12 O2 (g): (1)
is used where appropriate.
The electrolyzer model presented is written as a FOR- For this reaction to occur a minimum electric voltage must
TRAN subroutine primarily designed to run with the simu- be applied to the two electrodes. This minimum voltage, or
lation programs TRNSYS and EES, but the model has also reversible voltage, can be determined from Gibbs energy
been designed so that it readily can be integrated into other for water splitting (described below). In an alkaline elec-
simulation programs (e.g., MATLABJ Simulink J ). TRN- trolyzer the electrolyte is usually aqueous potassium hydrox-
SYS is a transient systems simulation program with a mod- ide (KOH), where the potassium ion K + and hydroxide ion
ular structure [16]. The TRNSYS library includes many of OH− take care of the ionic transport. The anodic and ca-
the components commonly found in thermal and electrical thodic reactions taking place here are
renewable energy systems, as well as component routines to Anode : 2OH− (aq) → 12 O2 (g) + H2 O(l) + 2e− ; (2)
handle input of weather data or other time-dependent forcing
functions. The modular structure of TRNSYS gives the pro-
Cathode : 2H2 O(l) + 2e− → H2 (g) + 2OH− (aq): (3)
gram the desired Rexibility, as it facilitates for the addition
of mathematical models not included in the standard library. In an alkaline solution the electrodes must be resistant to
The program is well suited to perform detailed analyses of corrosion, and must have good electric conductivity and cat-
systems whose behavior is dependent on the passage of time. alytic properties, as well as good structural integrity, while
EES, an engineering equation solver, has built-in functions the diaphragm should have low electrical resistance. This
for thermodynamic and transport properties of many sub- can, for instance, be achieved by using anodes based on
stances, including steam, air, refrigerants, cryogenic Ruids, nickel, cobalt, and iron (Ni, Co, Fe), cathodes based on
JANAF table gases, hydrocarbons and psychrometrics [17]. nickel with a platinum activated carbon catalyst (Ni, C–Pt),
Additional property data can be added, and the program and nickel oxide (NiO) diaphragms. Fig. 3 illustrates the
allows user-written functions, procedures, modules, and tab- operation principle of alkaline water electrolysis.
ular data. In this study EES was used to perform parameter
sensitivity analyses and to test and verify the model against 2.1. Thermodynamic model
measured data, while TRNSYS was used to perform inte-
grated system simulations. Thermodynamics provides a framework for describing
reaction equilibrium and thermal eAects in electrochemical
reactors. It also gives a basis for the de2nition of the driving
2. Model description forces for transport phenomena in electrolytes and leads to
the description of the properties of the electrolyte solutions
The decomposition of water into hydrogen and oxygen [18]. Details on the fundamental equations for electrochemi-
can be achieved by passing an electric current (DC) between cal reactors, or electrolyzers, are found in the basic literature
two electrodes separated by an aqueous electrolyte with [19]. Below is a brief description of the thermodynamics
good ionic conductivity [9]. The total reaction for splitting of the low-temperature hydrogen–oxygen electrochemical
 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33 25

reactions used in the electrolyzer model. (A maximum elec- 2.2

◦
trolyzer temperature of 100 C was assumed in this study.) 2.0 T = 20°C
The following assumptions can be made about the water

Voltage, V/cell
splitting reaction: (a) hydrogen and oxygen are ideal gases, 1.8
(b) water is an incompressible Ruid, and (c) the gas and T = 80°C
1.6 Overvoltage
liquid phases are separate. Based on these assumptions the Utn@T=20-80°C
change in enthalpy OH , entropy OS, and Gibbs energy OG 1.4 Urev@T = 20°C
of the water splitting reaction can be calculated with refer-
ence to pure hydrogen (H2 ), oxygen (O2 ), and water (H2 O) 1.2
Urev@T = 80°C
◦
at standard temperature and pressure (25 C and 1 bar). The 1.0
total change in enthalpy for splitting water is the enthalpy 0 50 100 150 200 250 300 350
diAerence between the products (H2 and O2 ) and the re- Current Density, mA/cm2
actants (H2 O). The same applies for the total change in
entropy. The change in Gibbs energy is expressed by Fig. 4. Typical I –U curves for an electrolyzer cell at high and low
temperatures.
OG = OH − T OS: (4)
◦
At standard conditions (25 and 1 bar) the splitting of water used in this study is, for a given temperature
r
t 
is a non-spontaneous reaction, which means that the change U = Urev + I + s log I + 1 : (7)
in Gibbs energy is positive. The standard Gibbs energy for A A
◦
water splitting is OG = 237 kJ mol−1 . For an electrochem- Fig. 4 shows the cell voltage versus the current density at
ical process operating at constant pressure and temperature a high and low operation temperature for a typical alkaline
the maximum possible useful work (i.e., the reversible work) water electrolyzer. As seen, the diAerence between the two
is equal to the change in Gibbs energy OG. Faraday’s law I –U curves is mainly due to the temperature dependence of
relates the electrical energy (emf) needed to split water to the overvoltages.
the chemical conversion rate in molar quantities. The emf In order to properly model the temperature dependence
for a reversible electrochemical process, or the reversible of the overvoltages (Eq. (7)) can be modi2ed into a more
cell voltage, is expressed by detailed I –U model, which takes into account the tempera-
OG ture dependence of the ohmic resistance parameter r and the
Urev = : (5) overvoltage coe.cients s and t. A temperature dependent
zF
I –U model has been proposed by the author [21]:
The total amount of energy needed in water electrolysis r1 + r2 T
is equivalent to the change in enthalpy OH . From Eq. U = Urev + I
A
(4) it is seen that OG includes the thermal irreversibil-  
ity T OS, which for a reversible process is equal to the t1 + t2 =T + t3 =T 2
+ s log I +1 : (8)
heat demand. The standard enthalpy for splitting water is A
◦
OH = 286 kJ mol−1 . The total energy demand OH is
related to the thermoneutral cell voltage by the expression The Faraday e.ciency is de2ned as the ratio between the ac-
tual and theoretical maximum amount of hydrogen produced
OH in the electrolyzer. Since the Faraday e.ciency is caused
Utn = : (6)
zF by parasitic current losses along the gas ducts, it is often
called the current e9ciency. The parasitic currents increase
At standard conditions Urev = 1:229 V and Utn = 1:482,
with decreasing current densities due to an increasing share
but these will change with temperature and pressure. In the
of electrolyte and therefore also a lower electrical resistance
applicable temperature range Urev decreases slightly with
[20]. Furthermore, the parasitic current in a cell is linear to
increasing temperature (Urev@80◦ C; 1 bar = 1:184 V), while
the cell potential (Eq. (8)). Hence, the fraction of parasitic
Utn remains almost constant (Utn@80◦ C; 1 bar = 1:473 V). In-
currents to total current increases with decreasing current
creasing pressure increases Urev slightly (Urev@25◦ C; 30 bar =
densities. An increase in temperature leads to a lower re-
1:295 V), while Utn remains constant.
sistance, more parasitic currents losses, and lower Faraday
e.ciencies. An empirical expression that accurately depicts
2.2. Electrochemical model these phenomena for a given temperature is
(I=A)2
The electrode kinetics of an electrolyzer cell can be mod- F = f2 : (9)
f1 + (I=A)2
eled using empirical current–voltage (I –U ) relationships.
Several empirical I –U models for electrolyzers have been According to Faraday’s law, the production rate of hydro-
suggested [11,13,14,20]. The basic form of the I –U curve gen in an electrolyzer cell is directly proportional to the
26 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33

transfer rate of electrons at the electrodes, which in turn is balance can be expressed as
equivalent to the electrical current in the external circuit. dT
Hence, the total hydrogen production rate in an electrolyzer, Ct = Q̇gen − Q̇loss − Q̇cool ; (13)
dt
which consists of several cells connected in series, can be
expressed as where
nc I Q̇gen = c (U − Utn )I = nc UI (1 − e ); (14)
ṅH2 = F : (10)
zF
The water consumption and oxygen production rates are sim- 1
Q̇loss = (T − Ta ); (15)
ply found from stoichiometry (Eq. (1)), which on a molar Rt
basis 2 is
Q̇cool = Ccw (Tcw; i − Tcw; o ) = UAHX LMTD (16)
ṅH2 O = ṅH2 = 2ṅO2 : (11)
and
The generation of heat in an electrolyzer is mainly due to
electrical ine.ciencies. The energy e9ciency can be calcu- (T − Tcw; i ) − (T − Tcw; o )
LMTD = : (17)
lated from the thermoneutral voltage (Eq. (6)) and the cell ln[(T − Tcw; i )=(T − Tcw; o )]
voltage (Eq. (8)) by the expression
The 2rst term on the right-hand side of Eq. (13) is the in-
Utn ternal heat generation, the second term the total heat loss
e = : (12)
U to the ambient, and the third term the auxiliary cooling de-
mand. The overall thermal capacity Ct and resistance Rt for
For a given temperature, an increase in hydrogen production
the electrolyzer, and the UA-product for the cooling water
(i.e., an increase in current density) increases the cell voltage
heat exchanger are the constants that need to be determined
(Fig. 4), which consequently decreases the energy e.ciency.
analytically or empirically prior to solving the thermal equa-
For a given current density, the energy e.ciency increases
tions. It should be noted that the thermal model presented
with increasing cell temperature. It should be noted here that
here is on a per stack basis.
Eq. (12) is only valid for systems where no auxiliary heat
A simple method to calculate the electrolyzer tempera-
is added to the system. (If auxiliary heat is added, the volt-
ture is to assume constant heat generation and heat transfer
age may, at very low current densities, drop into the region
rates for a given time interval. If the time steps are chosen
between the reversible and thermoneutral voltage, and the
su.ciently small, the result is a quasi steady-state thermal
e.ciency would be greater than 100%). In low-temperature
model. Using Eq. (13) as the basis, a quasi steady-state ther-
electrolysis, the cell voltage will during normal operation
◦ mal model can be expressed as
(50 –80 C and 40 –300 mA cm−2 ) always be well above the
thermoneutral voltage, as observed in Fig. 4. However, some Ot
T = Tini + (Q̇ − Q̇loss − Q̇cool ): (18)
initial heating may be required during start-up if the elec- Ct gen
trolyzer has been allowed to cool down to ambient temper-
◦
ature (∼ 20 C). A more complex method is to solve the diAerential equation
In order to calculate the overall performance of an elec- analytically and calculate the temperature directly. How-
trolyzer system, information about the number of cells in ever, for this to be possible an expression for the outlet cool-
series and=or parallel per stack and the number of stacks per ing water must 2rst be found. If a constant temperature in
unit is needed. The rated voltage of an electrolyzer stack is the LMTD-expression above is assumed, Eq. (16) can be
found from the number of cells in series, while the number of rewritten to
  
cells in parallel yields the rated current (and H2 -production). UAHX
The total power is simply the product of the current and Tcw; o = Tcw; i + (T − Tcw; i ) 1 − exp − : (19)
Ccw
voltage.
Using Eq. (13) as the basis and inserting Eqs. (14) – (16)
and (19), it can be demonstrated that the overall ther-
2.3. Thermal model mal energy balance on the electrolyzer can be expressed
by the linear, 2rst-order, non-homogeneous diAerential
The temperature of the electrolyte of the electrolyzer can equation
be determined using simple or complex thermal models,
depending on the need for accuracy. Assuming a lumped dT
+ aT − b = 0 (20)
thermal capacitance model [22], the overall thermal energy dt
with solution
 
2 Gas Row rates are commonly given on a normal cubic meter b b
T (t) = Tini − exp(−at) + ; (21)
per hour (Nm3 h−1 ) basis (Eq. (A.1)). a a
 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33 27

where 2.2
  
1 Ccw UAHX Predicted
a= + 1 − exp − ; (22) 2.0
t Ct Ccw

Voltage, V/cell
1.8
nc UI (1 − e ) Ta
b= +
Ct t T = 30°C
1.6 T = 40°C
   T = 50°C
Ccw Tcw; i UAHX
+ 1 − exp − : (23) T = 60°C
Ct Ccw 1.4 T = 70°C
T = 80°C

One advantage of having an analytical expression for the 1.2

0 50 100 150 200 250 300 350
temperature is that it facilitates the determination of the ther- 2
mal time constant (t = Rt Ct ). It also provides a means to Current Density, mA/cm
double check a numerical solution of the diAerential equa-
Fig. 5. Predicted versus measured electrolyzer cell voltage.
tion.
In an alkaline electrolyzer with a stationary electrolyte,
it has been observed that the overall UA-product for the
cooling water heat exchanger is (indirectly) a function of 3.1. Electrochemical model
the electrical current required by the electrolyzer (Fig. 10).
Hence, an empirical expression that accounts for both con- In order to 2nd the six parameters needed in the proposed
duction and convection heat transfer phenomena is pro- empirical I –U relationship (Eq. (8)), a systematic strategy
posed: for obtaining the best possible curve 2t was developed (Ap-
pendix A). A comparison between simulated and measured
UAHX = hcond + hconv I: (24) values for current and voltage for various operation tem-
peratures are presented in Fig. 5. The current, voltage, and
The physical explanation for this behavior is that since the temperature data base (317 data points) used in Fig. 5 was
electrolyte is stationary, and no pump is being used, the derived from 3 months (May–July 1996) of operational data
convection heat transfer increases as a result of more mixing for the PHOEBUS electrolyzer.
of the electrolyte. An increase in mixing occurs because the The results show to which degree the ohmic resistance
volume of the gas bubbles in the electrolyte increases with parameter r is linearly dependent on temperature. Further-
increasing current density. Similarly, the ohmic resistance more, the results show that the overvoltage coe.cient s can
in the electrolyte increases with increasing currents due to be assumed constant, while the proposed expression for the
increasing gas bubbling. Hence, this behavior is accounted overvoltage coe.cient t can be used. That is, only six pa-
for in Eq. (24) and in the ohmic resistance term of Eq. (8). rameters are needed to model the I –U curve. Fig. 5 demon-
strates that the predictability of the proposed I –U model in
Eq. (8) is excellent; the RMS error is about 2:5 mV cell−1 .
3. Testing and verication of model Detailed measurements of the hydrogen production at
various current densities for the PHOEBUS electrolyzer
The alkaline electrolyzer analyzed in this study is the (26 kW; 7 bar) were only available for an operation temper-
◦
one installed at the PHOEBUS plant in J0ulich [23]. It is ature of 80 C. However, detailed experiments on the tem-
a so-called advanced alkaline electrolyzer that operates at perature sensitivity of the Faraday e.ciency were performed
◦
a pressure of 7 bar and at temperatures up to about 80 C. on a very similar electrolyzer (10 kW; 5 bar) installed at the
The cells are circular, bipolar (Fig. 1), have a zero spac- HYSOLAR test and research facility for solar hydrogen pro-
ing geometry, and consist of NiO diaphragms and activated duction in Stuttgart, Germany [20]. A comparison between
electrodes (Fig. 3), which make them highly e.cient. The these two electrolyzers is given in Fig. 6, which shows the
electrolyte is a stationary 30 wt% KOH solution. Each cell data points from the HYSOLAR experiments (performed at
◦ ◦ ◦
has an electrode area of 0:25 m2 and there are 21 cells con- temperatures of 40 C; 60 C, and 80 C), the data points for
◦
nected in series. This gives an operation voltage in the range PHOEBUS (80 C), and the corresponding curve 2ts. Fig. 6
30 –40 V. illustrate that the form of the Faraday e.ciency expression
The hydrogen production and water cooling Row rates for proposed in Eq. (9) is suitable. A more detailed analysis of
the PHOEBUS electrolyzer was not logged and collected on the results shows that the coe.cients f1 and f2 vary lin-
a regular basis, along with the minutely collected operational early with temperature (Appendix A).
data. However, an experiment, where this and other pertinent In a system simulation study it is important to re-
data was sampled for every 5 min, was performed on June duce the number of parameters required. Having this in
17, 1996 [24]. It is this 1-day experiment that forms the mind, a simpli2ed Faraday e.ciency expression with
basis for the comparisons between simulated and measured non-temperature-dependent coe.cients (Eq. (9)) was
data presented below. tested. The measured hydrogen production (recorded at
28 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33

4
100
Electrolyzer switched ON

H2 Flow Rate, Nm /hr

80 3
Efficiency, %

3
60 Predicted
HYSOLAR:
T = 40°C
2
40
PHOEBUS: T = 60°C Measured
T = 80°C T = 80°C 1 Predicted
20

0
0 50 100 150 200 0
0 4 8 12 16 20 24
2
Current Density, mA/cm Time, h

Fig. 6. Predicted versus measured faraday e.ciency for two ad- Fig. 8. Predicted versus measured H2 -production for a typical day
vanced alkaline electrolyzers: (1) PHOEBUS (26 kW; 7 bar) and with variable electrolyzer current input. (For clarity, only every
(2) HYSOLAR (10 kW; 5 bar). fourth measured data point was plotted).

4 4

Measured

H2 Flow Rate, Nm /hr

3 3 Predicted
Predicted, Nm /hr

3
3

2 2

1 1

3
RMS error = 0.053 Nm /hr
0 0
0 1 2 3 4 0 50 100 150 200 250 300 350 400

Measured, Nm /hr
3 Current, A

Fig. 7. Predicted versus measured hydrogen production. Fig. 9. Predicted and measured H2 -production as a function of
measured electrical current. (The same day as shown in Fig. 8, but
with all of the data points included).
5 min intervals over a 15-h time period) was compared
to the model. The results (Fig. 7) show that for system 500 50
simulations it su.ces to model the Faraday e.ciency with Electrolyzer switched ON
a simple non-temperature-dependent expression. This can
UA-product, W/°C
400 Measured Predicted 40
Predicted
Current, A

be explained by the fact that in an actual system the elec-

300 30
trolyzer is always operating above a minimum protective
(idling) current (Fig. 9). Hence, the electrolyzer is usu- 200
Current
20
ally operating in a region where the Faraday e.ciency is
not signi2cantly aAected by a change in temperature. A 100 10
closer look at the hydrogen production for a typical day
with variable electrical current (Figs. 8 and 9) show that 0
0 4 8 12 16 20 24
0

there is good agreement between the measured data and the Time, h
model.
Fig. 10. Overall UA-product for the heat exchanger between the
cooling water and the electrolyzer.
3.2. Thermal model

The cooling of the electrolyzer is crucial to prevent over- the cooling water and the electrolyzer was proposed in Eq.
heating. The most convenient cooling method is usually to (24). A comparison between the measured and predicted
use regular tap water, as was the case with the PHOEBUS UA-product is given in Fig. 10. The results clearly shows
electrolyzer. The average tap water Row rate and inlet tem- that there exists an indirect relationship between the elec-
perature for the speci2c day discussed here was 0:6 Nm3 h−1 trolyzer current and the UA-product, and that this can, to
◦
and 14:5 C, respectively. In order to estimate the cooling a good approximation, be accounted for by the proposed
eAect an overall UA-product for the heat transfer between empirical equation.
 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33 29

100 500 two distinct modeling areas: (1) system design (or redesign)
Electrolyzer switched ON Tmeas
and (2) optimization of control strategies.
400
Temperature, °C

80
The applicability of the model is best illustrated by show-

Current, A
60 300 ing some results from integrated system simulations. The
reference system (Fig. 12) used in the simulations pre-
Tpred
40 200 sented here is the PHOEBUS demonstration plant at the
Research Center in J0ulich, Germany [23]. At the time of
20 100
Current the study this system consisted of four diAerently oriented
0 0 PV-arrays with maximum power point trackers (MPPTs),
0 4 8 12 16 20 24 a pressurized advanced alkaline electrolyzer, hydrogen and
Time, h
oxygen storage pressure vessels, an alkaline fuel cell, power
Fig. 11. Predicted versus measured electrolyzer temperature.
conditioning equipment (two DC=DC-converters and one
DC=AC-inverter), and a lead acid battery bank.
The level of detail of the electrolyzer model makes it pos-
The electrolyzer temperature for a day with variable sible to investigate a number of important system perfor-
power input (solar energy minus user load) is depicted in mance parameters such as the number of electrolyzer starts,
Fig. 11. The initial temperature (at midnight) was 56:4 C
◦
H2 -production, operating time, and standby (idling) time.
and the temperature at start-up of the electrolyzer (04:00 Statistical data such as minimum, maximum, and average
◦
AM) was 51:7 C. This initial decrease in temperature is electrolyzer current, voltage, power, Row rates and temper-
only due to natural cooling to the ambient, with a tempera- atures can also be analyzed. Comparisons with the reference
◦
ture of about 20 C. The values for the thermal capacitance system show that the model has a suitable level of detail
and for the overall thermal resistance were found by inves- [5].
tigating the cooling pattern for electrolyzer for a number One of the key system control parameters is the opera-
◦ ◦
of diAerent days: Ct = 625 kJ C−1 and Rt = 0:167 C W−1 tional mode of the electrolyzer, which determines whether
(equal to t = 29 h). The heat generation was calculated the electrolyzer is to operate in a 2xed or variable current
from the energy e.ciency (Eq. (14)), where the electrical mode. In the constant current mode the battery is charged
current input was based on measurements depicted in Fig. during periods of excess current on the busbar and dis-
11, and the auxiliary cooling was based on measured tap charged during periods with de2cit current. The battery state
water conditions (0:6 Nm3 h−1 and 14:5 C).
◦
of charge (SOC) in this case will mainly depend on two
The result of this 1-day simulation shows that the model uncontrollable variables, the solar radiation and user load,
slightly underpredicts the temperature. There might be and one controllable variable, the 2xed current (or power)
several explanations for this. One possible reason is mea- setting of the electrolyzer.
surement error, where the main source of uncertainty is the In the variable current scenario only excess current avail-
measured temperature of the electrolyte. Another possible able on the busbar is fed to the electrolyzer, hence the
reason is simply the lack of detail in the thermal model, battery SOC remains constant. It is important to note that
where the main de2ciency of the model is that the tem- most alkaline electrolyzers, even advanced ones speci2cally
perature of the electrolyte is assumed to be homogeneous. designed to manage Ructuating input current, can only
However, in general, the thermal dynamic behavior of the operate down to about 20% of their rated power, and an
electrolyzer is predicted quite accurately. idling current needs to be maintained. Table 1 summarizes
The importance of modeling the electrolyzer temperature the result from an analysis made on the inRuence of alterna-
accurately depends on the purpose of the models. A com- tive electrolyzer control strategies on system performance.
parison between predicted and measured electrolyzer volt- Electrolyzer mode of operation ( :xed or variable): The
age and the corresponding power shows that the slight un- bene2t of operating the electrolyzer in a variable current
derprediction of the temperature has relatively little signi2- mode rather than at a 2xed current is illustrated by com-
cance from an energy system simulation point of view. For paring Sim A–Sim B. In Sim B the electrolyzer was oper-
instance, in the 1-day simulation presented above, the error ating at a 2xed current of 550 A (∼ 21 kW), which gave
between the total simulated and measured energy demand a much more frequent electrolyzer on=oA-switching than in
was less than 2%. Sim A. This is also reRected in the low average run time.
The net result is less hydrogen production, yielding a lower
2nal pressure level in the H2 -storage at the end of the year.
4. System simulation results In Sim A, on the other hand, a large fraction of the energy
from the PV-arrays was used directly to run the electrolyzer.
The intention of the electrolyzer model presented in this As a consequence the use of the battery was minimized. A
study is to integrate it with other renewable and hydrogen en- comparison between Sim A and Sim B shows that the bat-
ergy models, and perform system simulation studies. Thus, tery discharging energy increased by about 50%. Other 2xed
the usefulness of the model can essentially be divided into current set points have been investigated [5] and the trend
30 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33

Fig. 12. Reference system (PHOEBUS).

Table 1
InRuence of alternative electrolyzer control strategies on system performance

Component=system Units Electrolyzer variations

A B C

Control set points

Electrolyzer Fixed current set point (if any) A Variable 550 Variable
Battery SOC level for on-switching % 90 90 80
Battery SOC level for oA-switching % 80 80 70

Key performance indicators

H2 -storage Initial H2 storage pressure 0:::1 0.45 0.45 0.45
Final H2 storage pressure 0:::1 0.24 0.11 0.20
Performance parameters
Electrolyzer Energy consumption MW h 10.56 9.94 10.83
Average power kW 7.48 21.01 7.37
Number of starts — 156 273 162
Average run time h 9.0 1.7 9.1
H2 production Nm3 2,719 2,473 2,788
Battery Energy, discharging MW h 9.74 15.09 9.54
Energy, charging MW h 8.90 13.68 8.85
Italics = change in set point or majoreAect due to change in set point.

is the same: variable electrolyzer operation mode gives a were lowered by 10% compared to Sim A, which required
better system performance than 2xed current mode. the battery to operate much more frequently at medium high
Basic control strategy for electrolyzer: The inRu- SOC (60 –80%), and less frequently at high SOC levels.
ence of reducing the upper and lower thresholds for the Hence, the higher threshold settings used in Sim A gave a
on=oA-switching of the electrolyzer can be seen by compar- better utilization of the installed battery capacity. In general,
ing Sim A–Sim C. In Sim C the on=oA-switching set points there is a tradeoA between battery capacity utilization on
 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33 31

one hand, and the need to dump energy (at SOC ¿ 100%) goes to J0urgen Mergel (FZJ) for being so helpful. A humble
on the other. This needs to be incorporated into the control thought also goes to my late advisor Professor Odd Andreas
strategy. AsbjHrnsen.

5. Conclusions Appendix A.

A mathematical model for an advanced alkaline elec- A.1. Curve :tting

trolyzer has been developed based on a combination funda-
mental thermodynamics, heat transfer theory, and empirical The following systematic seven-step procedure that facil-
electrochemical relationships. A lumped capacitance ther- itates the curve 2tting of the six parameters needed in the
mal dynamic model with a special empirical relationship proposed empirical I –U relationship (Eq. (8)) is recom-
for the overall heat transfer between a stationary electrolyte mended:
and a cooling water loop has also been proposed. Data from
a reference system, the stand-alone photovoltaic-hydrogen (1) Collect experimental or operational data for current I ,
energy plant (PHOEBUS) in J0ulich, was collected. Compar- voltage U , and temperature T .
isons between predicted and measured data show that the (2) Organize the measured values for I and U in sets with
electrochemical part of the model accurately predicts the cell respect to constant values for T .
voltage, hydrogen production, and e.ciencies. The results (3) Perform individual curve :ts of the three coe.cients
also show that the thermal model can be used to predict the r; s, and t in Eq. (7).
transient behavior of the electrolyzer temperature. (4) Repeat step (3) for a few other temperatures (e.g., T =
◦ ◦
The number of required parameters has been reduced to a 20 C–80 C).
minimum so that the model is suitable for use in integrated (5) Perform intermediate curve :ts on the temperature-
hydrogen energy system simulations. The electrolyzer model dependent coe.cients r and t.
has, along with several other hydrogen energy models such (6) Verify that the temperature-dependent coe.cients in
as fuel cells and hydrogen storage, been made compatible Eq. (8) behave according to the expressions:
to a transient system simulation program (TRNSYS), which r(T ) = r1 + r2 T and t1 + t2 =T + t3 =T 2 :
makes it possible to integrate hydrogen energy component
model with a standard library of thermal and electrical re- (7) Perform an overall curve :t on the entire data set, using
newable energy components. The model can be particularly the values for the parameters ri ; s, and ti found from
useful for (1) system design (or redesign) and (2) optimiza- steps (1) to (6) as initial values for the regression.
tion of control strategies. To illustrate the applicability of
This systematic procedure is illustrated graphically in Fig.
the model, a 1-year simulation of a photovoltaic-hydrogen
13, which shows the results of individual curve 2ts at 2xed
system was made. The results show that several improved
temperatures (data points), intermediate curve 2ts of these
electrolyzer-operating strategies can be identi2ed with the
data points (thin lines), and 2nally the overall curve 2t (solid
developed system simulation model.
lines). The step-by-step strategy for 2nding the unknown
The technical electrolyzer model presented in this paper
parameters in Eq. (8) proves to be very robust. This indicates
is suitable for dynamic simulation of RE=H2 -systems. The
that the approach is not only limited to the curve 2tting of
level of detail in the model is relatively high, but not too
the I –U characteristics of an electrolyzer cell, but can also
high. This means that the model readily can be coupled to
be used in other situations where coe.cients in a model are
economic models that account for both investment and oper-
sensitive to inputs such as temperature, pressure, or other
ational costs. Detailed techno-economical system optimiza-
governing conditions.
tion of RE=H2 -systems is the next research topic that will
be studied by this author.
A.2. Parameters

Acknowledgements Tables 2– 4 list the parameters found and used in this

study.
The work presented in this paper is based a Ph.D.-study
carried out at the Institute for Energy Technology (IFE) A.3. Equations
in conjunction with the Norwegian University of Science
and Technology (NTNU). The author would like to thank Gas Row rates are commonly given on a normal cubic
the Norwegian Research Council (NFR) for the 2nancing meter per hour (Nm3 h−1 ) basis. For ideal gases the con-
(Ph.D. and post.doc.), IFE for providing the necessary re- version from mol s−1 to Nm3 h−1 is given by
search facilities, and the Research Center in J0ulich (FZJ) for
making the experimental data accessible. A special thanks V̇ = ṅvstd 3600; (A.1)
32 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33

10-4

Overvoltage Coeff. (t), m /A

1.0

2
Resistance Param. (r), Ω m
Resistance, r:
8x10-5 0.8 step 3
step 5
6x10-5 0.6 step 7

Overvoltage, t:
4x10-5 0.4
step 3
step 5
2x10-5 0.2
step 7

0x100 0.0
20 40 60 80 100
Temperature, °C

Fig. 13. Electrolyzer I –U curve 2tting; illustration of a step-by-step procedure to determine the temperature sensitivity of the resistance
parameter r and the overvoltage coe.cient t in Eq. (8).

Table 2 References
I –U curve parameters (Eq. (8))
[1] Schucan TH. Case studies of integrated hydrogen energy
r1 8:05e − 5 M m2
◦ systems. Report, IEA=H2=T11=FR1-2000, International
r2 −2:5e − 7 M m2 C−1
Energy Agency Hydrogen Implementing Agreement Task
s 0:185 V
11—Integrated Systems. Operating agent: National
t1 −1:002 A−1 m2
◦ Renewable Energy Laboratory, Golden, Colorado, 1999.
t2 8:424 A−1 m2 C
◦ [2] The Hydrogen project at munich international airport.
t3 247:3 A−1 m2 C2
http://www.hydrogen.org/h2muc/, October 2001.
[3] Yamaguchi M, Shinohara T, Taniguchi H, Nakanori T,
Okisawa K. Development of 2500 cm2 solid polymer
electrolyte water electrolyser in WE-NET. Proceedings of the
Table 3 12th World Hydrogen Energy Congress, Buenos Aires, 21–26
Faraday e.ciency parameters (Eq. (9)) June 1998. p. 747–55.
[4] Galli S, Stefanoni M, Borg P, Brocke WA, Mergel
PHOEBUS HYSOLAR J. Development and testing of a stand-alone small-size
T 80 40 60 80
◦
C solar photovoltaic-hydrogen power system (SAPHYS).
f1 250 150 200 250 mA2 cm−4 Report, JOU2-CT94-0428, JOULE II-Programme, Directorate
f2 0.96 0.990 0.985 0.980 0:::1 General XII: Science, Research and Development, European
Commission, Brussels, 1997.
[5] Ulleberg I. Stand-alone power systems for the future: optimal
design, operation and control of solar-hydrogen energy
systems. Ph.D. thesis, Norwegian University of Science and
Table 4 Technology, Trondheim, 1998.
UAHX parameters (Eq. (24))
[6] Eriksen J, Aaberg RJ, Ulleberg I, Ingebretsen F. System
hcond
◦
7 W C−1 analysis of a PEMFC-based stand alone power system
◦ (SAPS). First European PEFC Forum, Lucerne, Switzerland,
hconv 0:02 W C−1 per A
3– 6 July 2001. p. 447–58.
[7] International energy agency (IEA) hydrogen program.
Annex 13—design and optimization of integrated systems.
http://www.eren.doe.gov/hydrogen/iea/, October 2001.
where V̇ and ṅ are the volumetric and molar Row rates, [8] Wendt H, Plzak H. Hydrogen production by water electrolysis.
respectively, and vstd is the volume of an ideal gas at standard Kerntechnik 1991;56(1):22–8.
◦
conditions (0 C and 1 bar). [9] Divisek J. Water electrolysis in low- and medium-temperature
The root mean square (RMS) error is calculated according regime. In: Wendt H, editor. Electrochemical hydrogen
to the equation: technologies-electrochemical production and combustion of
hydrogen. Oxford: Elsevier, 1990.

n [10] Hug W, Bussmann H, Brinner A. Intermittent operation and
i=1 (ŷ i − yi )
2
RMS error = ; (A.2) operation modeling of an alkaline electrolyzer. Int J Hydrogen
n−1 Energy 1993;18(12):973–7.
[11] Griesshaber W, Sick F. Simulation of Hydrogen–Oxygen–
where ŷ i and yi are the predicted and measured values, Systems with PV for the Self-Su.cient Solar House. FhG-ISE,
respectively, and n is the number of data points evaluated. Freiburg im Breisgau, 1991 (in German).
 (. Ulleberg / International Journal of Hydrogen Energy 28 (2003) 21 – 33 33

[12] Ulleberg I, Morner SO. TRNSYS simulation models for [18] Rousar I. Fundamentals of electrochemical reactors. In:
solar-hydrogen systems. Solar Energy 1997;59(4 – 6):271–9. Ismail MI, editor. Electrochemical reactors: their science and
[13] Havre K, Borg P, Tommerberg K. Modeling and control of technology part A. Amsterdam: Elsevier, 1989.
pressurized electrolyzer for operation in stand alone power [19] Pickett DJ. Electrochemical reactor design, 2nd ed. New York:
systems. Second Nordic Symposium on Hydrogen and Fuel Elsevier, 1979.
Cells for Energy Storage, Helsinki, January 19 –20, 1995. p. [20] Hug W, Divisek J, Mergel J, Seeger W, Steeb H. Highly
63–78. e.cient advanced alkaline electrolyzer for solar operation. Int
[14] Vanhanen J. On the performance improvements of small-scale J Hydrogen Energy 1992;17(9):699–705.
photovoltaic-hydrogen energy systems. PhD thesis, Helsinki [21] Ulleberg I. Simulation of autonomous PV-H2 systems:
University of Technology, Espoo, Finland, 1996. analysis of the PHOEBUS plant design, operation and energy
[15] Meurer C, Barthels H, Brocke WA, Emonts B, Groehn HG. management. ISES 1997 Solar World Congress, Taejon,
PHOEBUS—an autonomous supply system with renewable August 24 –30, 1997.
energy: six years of operational experience and advanced [22] Incropera FP, DeWitt DP. Fundamentals of heat and mass
concepts. Solar Energy 1999;67(1–3):131–8. transfer, 3rd ed. New York: Wiley, 1990.
[16] Klein SA, Beckman WA, Mitchell JW, Du.e JA, Du.e NA, [23] Barthels H, Brocke WA, BonhoA K, Groehn HG, Heuts
Freeman TL, Mitchell JC, Braun JE, Evans BL, Kummer G, Lennartz M, Mai H, Mergel J, Schmid L, RitzenhoA
JP, Urban RE, Fiksel A, Thornton JW, Blair NJ, Williams P. PHOEBUS-J0ulich: an autonomous energy supply system
PM, Bradley DE. TRNSYS—a transient system simulation comprising photovoltaics, electrolytic hydrogen, fuel cell. Int
program. Manual v15, Solar Energy Laboratory, University J Hydrogen Energy 1998;23(4):295–301.
of Wisconsin, Madison, 2000. [24] Mergel J. Personal communication, PHOEBUS electrolyzer:
[17] Klein SA, Alvarado FL. EES—engineering equation solver. detailed operational data. FZJ—Forschungszentrum, J0ulich,
Manual v6.315, F-Chart Software. Middleton: Wisconsin, February 1997.
2001.