Development of a performance model for a semi-industrial PEM electrolyzer: impact assessment of operational intermittency

EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476247 A1803-A1808 Page 1/10 A1803=A1808 Development of a performance model for a semiindustrial PEM electrolyzer: impact assessment of operational intermittency Emma Nguyen* (1,2), Pierre Olivier (1), Marie-Cécile Péra (2), Elodie Pahon (2), Robin Roche (2), Oscar Frappereau (1) (1) Engie Lab CRIGEN, Stains/France; (2) Université Marie et Louis Pasteur, UTBM, CNRS, institut FEMTO-ST, FCLAB, F-90000 Belfort/France; *Contact corresponding authors: www.EFCF.com/ContactRequest Abstract The transition to a low-carbon economy relies heavily on the deployment of renewable hydrogen as a key energy vector. Electrolyzers, particularly PEM technology, play a central role in this transition, yet their intermittent operation presents significant challenges. Understanding the impact of intermittency on electrolyzer performance is essential to ensuring the reliability and cost-effectiveness of large-scale hydrogen production projects. This knowledge is critical not only for mitigating risks but also for accelerating the deployment of low-carbon hydrogen at scale. Despite its importance, the current literature reveals a lack of quantitative studies addressing the specific effects of intermittency on electrolyzer performance, alongside an absence of dedicated methodologies for its characterization [1]. Previous experimental studies on a 55 kW semi-industrial pressurized PEM electrolyzer have suggested that dynamic electrical load fluctuations over hourly timescales do not significantly affect system performance compared to the average operating point. These results indicate that the electrolyzer can withstand certain types of intermittency without notable losses in efficiency or performance [2]. However, other critical aspects of intermittency remain underexplored, particularly the influence of fluctuation in operational states on system efficiency and overall lifespan. Modeling offers a unique advantage in simulating scenarios that are impractical to study experimentally. In this research, a performance model was created in MATLAB/Simulink to specifically analyze the impact of operational state intermittency. The model was implemented on the 55 kW semi-industrial pressurized PEM electrolyzer and validated against experimental operational data. This study delivers key insights into optimizing electrolyzer performance under intermittent conditions, contributing to the design of resilient and efficient hydrogen production systems integrated with renewable energy sources. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476247 A1803-A1808 Page 2/10 Introduction Understanding the impact of intermittency on industrial electrolyzers is essential for optimizing their performance and durability. Yet, experimental investigations alone face practical limitations in duration, cost, and operational flexibility. To overcome these constraints, modeling offers a powerful alternative, enabling the extrapolation of short-term experimental results and the exploration of complex operating scenarios that are otherwise difficult to reproduce under real-world conditions. Although a number of models have been proposed in the literature, they do not always provide a macroscopic representation of industrial-scale systems. This is primarily due to their frequent focus on cell or stack-level behavior, the use of empirical formulations tailored to specific configurations, and limited consideration of the complete Balance-of-Plant. As a result, these models often fail to provide access to performance indicators at an appropriate level of granularity for comprehensive system-level assessments. Additionally, such models are often not suited for long-term performance assessment or for exploring a wide range of operating scenarios representative of real-world industrial constraints. Other recent models [3], [4], [5], [6] stand out by adopting a dynamic system approach, providing a detailed description of the system components, from the stack to the Balance-of-Plant. While these models provide valuable insights into dynamic behavior and their influence on system performance, they may not be suitable for long-term extrapolation as their extended computation time poses a limitation — particularly in industrial settings where long-term testing is costly and complex. Additionally, their high reliance on detailed system parameters or tailored boundary conditions, may limit their adaptability to other systems, which poses a challenge for both scientific community and industrials. Having a reusable foundation would allow the model to be easily adapted to different configurations, thereby reducing the costs and time associated with modeling new systems. In light of the current gap in the literature — where few approaches effectively address the need for reliable long-term extrapolation — the main objective of the proposed research is to develop a model able to i) assess the system key performance indicators, including hydrogen production, hydrogen losses, power consumption, specific consumption, and ii) extend short-term experimental results to longer timeframes, potentially representative of the actual operational lifetime of electrolyzers. This responds to both academic interest and industrial requirements for predictive and cost-effective aging assessments. Moreover, the model is designed to be generic, in order to facilitate optimization studies related to system design and operation, while also allowing for adaptation to other industrial systems. Its development was initially adapted to a semi-industrial one-stack 55 kW PEM electrolyzer, which served as the foundation for its implementation and validation. Baseline specifications of the electrolyzer are summarized in Table 1 and a simplified process flow diagram of the system is provided in Figure 1. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476247 A1803-A1808 Page 3/10 Figure 1: Simplified process flow diagram of the 55 kW PEM electrolyzer [2]. Table 1: Baseline specifications of the PEM electrolyzer [2]. Specifications Value Nominal power (kW) 55 Number of cells per stack 47 Active surface of cells (cm²) 600 Hydrogen production (Nm3/h) 1 - 10 Nominal temperature (°C) 65 Stack outlet pressure (bar) 30 Hydrogen purity (%) 99.999 In this paper, the modeling approach and the associated validation results are presented. 2. Methodology A 0D dynamic approach was adopted due to its computational efficiency and suitability for system-level analysis. Additionally, the model relies on semi-empirical equations grounded in fundamental physical principles, with certain parameters requiring calibration on experimental data. Finally, specific assumptions and modeling choices were made to align with the defined objectives. 2.1. Assumptions A set of assumptions, spanning various operational scales of the system, were established to meet the modeling objectives. First, the model exclusively focuses on fluid and thermal dynamics, while electrical and electrochemical behaviors were excluded from the modeling scope. This is justified by their much faster time constants (on the order of milliseconds), which are negligible given the employed simulation time step [7]. Besides, this assumption was supported by prior experimental campaigns carried out on the electrolyzer [2]. Similarly, diffusion overpotentials were omitted, as they only become relevant at current densities above 2 A/cm², well beyond the system’s operational range, which peaks at approximately 0.85 A/cm². EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476247 A1803-A1808 Page 4/10 Hydrogen losses due to crossover mechanisms were not explicitly modeled. Instead, they were globally captured through an empirical loss coefficient, referenced to as “UH₂C”, which aggregates all losses at both stack and system levels. This includes Faradaic losses, crossover mechanisms, and fugitive leaks. Finally, all Balance-of-Plant components, as cooling units, gas separators, purification systems, tanks, and pumps, were grouped into a single variable. Although not individually modeled, their total power consumption was included based on data from previous test campaigns. This aggregate value also accounts for the power needs of the rectifier, sensors, air conditioning, and lighting in the system utilities, simplifying the model while still enabling good estimation of overall energy consumption. 2.2. Model development 2.2.1. Mass balance The molar production flows are determined following the Faraday law incorporating a correction factor referred to as “UH₂C”. This coefficient broadly accounts for crossover losses at the stack level and fugitive leaks at the system level. Therefore, these variables can be estimated following Equation (1) and Equation (2). n󰇗H2,actual =IDC 2F∗Ncells ∗UH2C (1) n󰇗O2,actual =IDC 4F∗Ncells ∗UH2C (2) Meanwhile, the implemented mass balance is described in Equation (3). dpH2 dt =R∗Tsys Vsys ∗(n󰇗H2 actual −(n 󰇗H2 prod+n󰇗H2 vent)) (3) Experimental data collected during the system pressurization over several days enabled the estimation of the electrolyzer dead volume, Vsys , which amounts to the total internal gasfilled volume of the system, as defined by Equation (4). Based on a average pressurization period of 600 seconds during the system cold start-up, this volume was estimated to be approximately 0.018 m³. Vsys =R∗Tsys pH2∗∫ n󰇗H2 actual(t)dt ∆t=600 s 0 (4) Pressure regulation during state transitions was managed in the model through the control of two single valves: a valve releasing gas to the vent, and another valve connected to the storage line. The mass balance was implemented in accordance with the flow diagram depicted in Figure 2. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476247 A1803-A1808 Page 5/10 Figure 2: Mass balance - model flow diagram. 2.2.2. Electrochemical model The electrochemical model was developed based on the analytical expression of the polarization curve. The cell voltage is defined as the sum of the reversible potential and the overpotentials generated during the electrochemical reaction. This equation is provided in Equation (5): Ucell =Urev + ∑ μact,k k=an,cat +μohm (5) UDC=Ucell∗Ncells (6) where Urev is the reversible potential, and μact,k and μohm are the activation and ohmic overpotentials, respectively. Based on literature, the reversible potential and standard reversible potential were respectively modelled following Equations (7) and (8) [7]. Urev =Urev°+RTsys 2F ∗ln(pH2 ∗pO2 0,5) (7) Urev°=1,5184−1,5421∗10−3 ∗Tsys+9,526∗10−5∗Tsys ∗ln(Tsys)+9,84∗10−8∗Tsys 2 (8) The ohmic overpotentials were determined according to Equation (9) : μohm =Req ∗ i (9) where Req is the equivalent ohmic resistance. Finally, the activation overpotentials were described by the Butler-Volmer equation. To simplify its definition, the symmetry factor is assumed to be 0.5. This widely adopted assumption implies an equal distribution of the energy barrier between the anode and the cathode [8]. With this simplification, the Butler-Volmer equation could be expressed as an inverse hyperbolic sine function, according to Equation (10) : EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476247 A1803-A1808 Page 6/10 μact,k =RTsys 2αkF∗sinh−1(i 2i0,k) (10) where αk is the charge transfer coefficient of the electrode k. The polarization curve reconstruction involved three technology-specific parameters that need to be calibrated on experimental data. These parameters include the anodic and cathodic exchange current densities (i0,an and i0,cat), and the equivalent ohmic resistance Req. 2.2.3. Thermal balance A control volume encompassing the stack, the two separators, the water circuits, and the circulation pumps was defined to model the system thermal behavior. The thermal balance was implemented following the flow diagram illustrated in Figure 3. Figure 3: Thermal balance - model flow diagram. The differential equation governing the system thermal balance is described in Equation (11). Cth∗dTsys dt =Qgen+Qpumps−Qloss (11) This equation highlights three terms related to heat exchange: two heat sources, Qgen and Qpumps, associated with the heat produced by the stack irreversibilities and the operation of the pumps, respectively, and one dissipation source, Qloss, which quantifies the heat lost by the system to the ambient. First, the heat generation term Qgen, originating from the electrochemical reaction, is related to the stack voltage UDC and the thermoneutral voltage Utn under given operating conditions. It is defined by Equation (12): Qgen =(UDC−Utn)∗IDC (12) Next, the pumps, which consume a fixed amount of electrical energy in AC current, generate a constant amount of heat, Qpumps, regardless of the system operating conditions. Furthermore, their activation depends on the system operational state and is only effective in production and stand-by conditions. Finally, the heat dissipation is expressed by the following equation: EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476247 A1803-A1808 Page 7/10 Qloss = kloss ∗(Tsys−Tamb) (13) where kloss is the heat transfer coefficient between the system and the surrounding ambient environment. 2.3. System characterization and model validation To ensure the model representativeness, operational data were used to compare simulated results with real-world data. Model calibration is performed via nonlinear optimization algorithms that minimize the mean relative error between simulation and experiment. Acceptability criteria for each performance indicator were based on typical standards reported in the literature [3], [7], [9]. This section details the parameter identification steps, which support model validation under both steady-state and dynamic conditions. 2.3.1. Electrochemical performance A nonlinear optimization algorithm was then applied to calibrate the electrochemical model by minimizing the root mean square error between the analytical equation and recorded experimental data. This method was selected for its ability to enforce realistic bounds on the three parameters - ohmic resistance and the anodic and cathodic exchange current densities. The calibration results, detailed in Table 2, yielded parameter values consistent with those typically reported for PEM electrolyzers operating under similar conditions. Table 2: Fitted parameters from identification process – electrochemical model. Parameter Value i0,an (A/cm²) 2.54*10-4 i0,cat (A/cm²) 1.00*10-2 Req (Ω.cm²) 0.42 MSE 6.13*10-5 Figure 4 provides a comparison between the experimental data and the simulated voltage values at the same current points, with a relative error of less than 1.25 % in absolute value. Figure 4: Polarization curve. Comparison between experimental and simulation results. Additionally, the efficiency curve was plotted based on the actual hydrogen production and the total system power consumption. Its reconstruction from simulation data was validated using the experimental data previously employed to calibrate the electrochemical model. As depicted in Figure 5, the relative deviation between the simulation and experimental results remains low across the system’s operating range. On average, the model slightly EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476247 A1803-A1808 Page 8/10 underestimates the specific consumption. However, its overall performance remains satisfactory. Figure 5: System efficiency curve. Comparison between experimental and simulation results. 2.3.2. Thermal behavior The system thermal balance involves three parameters to be calibrated: the thermal capacity Cth within the control volume, the heat generated by the circulation pumps Qpumps, and the heat transfer coefficient kloss. An optimization algorithm based on an explicit Euler numerical integration method was applied to fit the differential equation to the experimental data collected during the system cold start-up. The optimization of the temperature curve delivered calibrated parameters that fall within the range of values commonly reported in the literature [8]. These parameters are summarized in Table 3. Table 3: Fitted parameters from identification process - thermal balance. Parameter Value Cth (J/K) 326695.1 Qpumps (J) 4642.5 kloss (J/K) 64.71 Figure 6 depicts the system thermal response during a cold start. Results demonstrate a very good agreement between the simulation and the experimental measurements, indicating the model ability to accurately reproduce the system transient behavior. A correlation factor of 0.997 and a root mean square error of 1.11 K were achieved. Figure 6: System thermal response. Comparison between experimental and simulation results during the system cold start-up. EFCF 2025: Low-Temp. Fuel Cells, Electrolysers & H2 Processing 1 – 4 July 2025, Lucerne Switzerland https://doi.org/10.5281/zenodo.17476247 A1803-A1808 Page 9/10 To demonstrate the model validity, an intermittent profile from previous test campaign, presented in Figure 7, conducted on the electrolyzer was selected and simulated [2]. Simulation results were then compared to the experimental data obtained during the test. Key performance indicators — including voltage, power, hydrogen production, and specific consumption — show strong agreement between the model and measurements, with high coefficients of determination and low mean relative deviations, as described in Table 4. Figure 7: Random intermittent test selected for model validation. Random intermittent test (T75) from [2]. Table 4: Performance indicators obtained through experiments and simulation. Results from random intermittent test (T75) from [2]. T75 Simulation Experimental R² RD (%) RMSE DC voltage (V) 88.27 89.02 0.90 - 0.85 0.96 DC power (kW) 33.76 34.05 0.97 - 0.87 0.52 System power (kW) 46.25 46.63 0.93 - 0.83 0.79 Hydrogen production (Nm3/h) 6.67 6.72 0.08 - 0.75 1.13 System specific consumption (kWh/kgH2) 77.91 77.17 - 0.95 - Altogether, these results confirm the model ability to accurately replicate the system behavior under both steady-state and dynamic conditions. Therefore, the model is considered validated and can be confidently used for performance assessment and optimization, and operational scenario analysis. 4. Conclusion This paper introduced a generic performance model dedicated to assessing the behavior of industrial electrolyzers under intermittent operation, specifically designed for long-term simulations. The model was initially developed and calibrated for a semi-industrial PEM electrolyzer. A dynamic, 0D, semi-empirical approach was adopted. Implemented in Matlab/Simulink and calibrated with operational data, the model demonstrated good accuracy under both static and dynamic conditions, including critical phases such as cold start and shutdown. A major strength of this approach lies in its generic design, making the model adaptable to a broad range of industrial systems beyond the initial PEM case study. This flexibility can pave the way for advanced multi-module strategies as well as design and control optimization studies. By enabling efficient, cost-effective, and reliable performance assessments across various electrolyzer technologies over long time horizons and under complex, intermittently