Droop-Based Hierarchical Power Management of Hybrid Hydrogen Electrolyzers Under Grid-Frequency Disturbances

Hydrogen electrolyzers, traditionally employed for hydrogen production, are increasingly being considered as dynamic assets for grid support. Hybridization of proton exchange membrane electrolyzers (PEME) and alkaline electrolyzers (AE) offers a balance between fast response and lower cost. However, droop-based power management for such hybrid electrolyzer systems (HES) remains largely underexplored. This paper proposes a novel droop-based hierarchical power management strategy for a grid-scale HES comprising PEME and AE units during grid-frequency disturbances. The strategy assigns distinct frequency-deviation dead-bands and droop slopes to each technology. The PEME, characterized by smaller dead-bands and steeper droop characteristics, is designated for primary frequency support. When PEME stack’s input power limits are exceeded under larger frequency deviations, AE units characterized by larger deadbands and gentler droop slopes are activated to extend dynamic support within their limits. Both electrolyzers maintain constant-power operation for hydrogen production outside of frequency disturbance events. This hierarchical scheme enhances the dynamic range of grid-frequency support while preserving hydrogen output. Results demonstrate effective power sharing between the two electrolyzer types, leveraging their complementary strengths. Unlike prior hybrid approaches that broadly classify fast and slow components, the proposed hierarchical droop-based scheme explicitly links electrolyzer response to grid frequency using distinct dead-bands and droop slopes, ensuring coordinated participation of PEME and AE. This control-oriented approach is simpler, grid-compatible, and better suited for real-time ancillary service provision while maintaining hydrogen production.

Introduction

In recent years, due to global warming, air pollution, and climate change, the use of renewable energy has been increasing at an accelerated pace. Renewable energy sources are highly dependent on nature, time variability, resource availability, geographical location, and some other factors that make the power generated from these sources intermittent. To mitigate the impacts of intermittency, energy storage is necessary [1]. Energy storage systems have better ramping characteristics than traditional generators, which helps to maintain supply and demand more effectively. Batteries are the most common storage device, but they have limitations due to limited charge–discharge cycles and frequent replacement require-ments. Other storage technologies, such as supercapacitors, superconducting magnetic energy storage, flywheels, etc., are also available; however, they are constrained by drawbacks such as low energy density, high cost of cryogenic systems, significant self-discharge, etc. Nowadays, hydrogen storage is gaining popularity because of its long-life storage capacity, high energy density, and scalability. For the production of green hydrogen, electrolyzers powered by renewable energy sources are gaining popularity in many countries like Sweden, Australia, USA, etc. [2], [3], which work on the principle of converting electrical energy into hydrogen, which can be used for industrial purposes as well as producing electricity from hydrogen via hydrogen fuel cell. Moreover, these electrolyzers can be used as a controllable load and offer grid frequency and other ancillary services [2]–[8].

Modeling of hydrogen electrolyzer is one of the highly researched topics in the literature. Typical modeling techniques of an electrolyzer system adopted by researchers involve physical and electrical modeling [3], [8]–[11]. Electrical modeling is preferred because it is computationally simpler, easier to integrate into energy systems, scalable, and accurate enough for most practical applications, while physical models are more detailed but complex and time-consuming [11].

Several studies have investigated the application of electrolyzers for grid-frequency regulation. Among various technologies of electrolyzers, the proton exchange membrane electrolyzer (PEME) and alkaline electrolyzer (AE) are the most common types. Authors of [5] and [6] presented one of the first works to provide frequency support by grid-scale AE and PEME, respectively. However, the authors did not consider the hybrid electrolyzer scheme’s ability to provide frequency services. The studies in [3], [8], and [12] modeled PEME and AE using electrical equivalent circuits for grid integration, and evaluated their ability to provide frequency ancillary services and virtual inertia. These works proposed active power–frequency characteristics for the reference power of a single electrolyzer type. However, they considered each technology in isolation, without exploring a hybrid configuration or coordinated droop characteristics. Authors in [2] and [13] adopted resistance droop control, which emulates a virtual resistor at the output of dc-dc converters, enabling decentralized current sharing but introducing a voltage drop at the dc bus. It is also sensitive to line impedance mismatches, causing inaccurate power sharing, and suffers from a trade-off between voltage regulation and load sharing accuracy. These drawbacks limit its suitability in applications requiring both tight voltage control and precise power allocation. Filtration-based methods, such as in [14], decompose renewable fluctuations into low- and high-frequency components allocated to PEME and AE. Their focus was primarily on hydrogen production efficiency, volatility reduction, and lifetime extension, rather than direct provision of grid frequency ancillary services.

This paper presents a hybrid electrolyzer configuration integrated with the power grid, targeting both hydrogen production and frequency support. The hybrid electrolyzer system (HES) combines PEME and AE to exploit the fast response of the PEME along with the reliable support of the AE. Both electrolyzers operate to produce hydrogen at a constant rate during normal operation, improving the utilization efficiency of the costly technologies. To address limitations in existing literature, this paper proposes a droop-based hierarchical power management strategy that efficiently allocates power between the PEME and AE during grid frequency fluctuations. The hierarchical droop characteristics incorporate frequency-deviation dead-bands, with a larger dead-band assigned to the AE compared to the PEME, enabling the PEME stack to act as the primary dynamic load during frequency deviations. Furthermore, the droop design accounts for the slower dynamics of the AE, which provides supplementary support once the PEME reaches its operational limits in both over-frequency and under-frequency conditions.

The remainder of this paper is organized as follows: Section 2 presents the circuit models of both PEME and AE, Section 3 describes the proposed characteristics for dynamic power sharing, Section 4 outlines the system under consideration and discusses the results, and Section 5 summarizes the key findings.

Circuit Modeling of Electrolyzers

Several electrolyzer technologies exist, including AE, PEME, solid oxide electrolyzer (SOE), and anion exchange membrane electrolyzer (AEME). Among them, AE and PEME are the most commercially mature and widely adopted due to their readiness and compatibility with renewables, whereas SOE and AEME remain at early research stages [11]. AEs are low-cost, durable, and suitable for large-scale hydrogen production but have slow dynamics, high minimum load, and long start-up times, limiting grid-support applications. PEMEs, on the other hand, offer rapid response, high current density, compactness, and flexible partial-load operation, though their higher cost and reliance on precious metals restrict large-scale deployment [2]. These complementary features motivate hybrid AE–PEME systems for hydrogen production and ancillary services. Electrical equivalent circuit models of electrolyzers are essential for studies involving integration into the power system. Such models capture the nonlinear voltage–current characteristics, ohmic losses, and dynamic response of the electrolyzer in a simplified form that is compatible with standard simulation tools [5]. By representing the electrochemical process as an electrical network, these models enable accurate analysis of control strategies, grid interactions, and ancillary service provision without requiring complex multiphysics simulations. The electrical equivalent circuit models for the AE and PEME are shown in Fig. 1a and 1b, respectively. The electrical circuit models of the electrolyzers consist of a dc voltage source representing the minimum voltage requirement for the operation of a hydrogen electrolyzer known as the reversible voltage. The ohmic losses are modeled by a resistor representing the ohmic overvoltage, which accounts for the combined resistances of the catalyst layers, electronic current collectors, electrolyte membrane, and contact interfaces [11]. From an electrical perspective, activation overvoltage accounts for two effects: the charge transfer resistance, which represents the energy required to release electrons from electrode surfaces, and the time delay caused by charge accumulation across the polymeric membrane. These are typically modeled using a parallel RC branch, where the resistor emulates the energy loss in overcoming charge transfer barriers and the capacitor represents the double-layer capacitance. Such RC branches can be applied separately at the anode and cathode to capture their differing reaction dynamics [8], [11]. The cathode dynamics can be ignored in PEME circuit modeling due to the faster cathode dynamics as shown in Fig. 1b [2], [11], [12]. The parameters for 250 kW PEME and 250 kW AE are listed in Table I [2].

Fig. 1. Circuit models for (a) 250 kW AE stack and (b) 250 kW PEME stack.

Table I. Specifications of PEME and AE

Particulars	PEME	AE
Rated power (kW)	250	250
Reversible voltage (V)	132	114
Ohmic resistance (mΩ)	5.79	30.40
Anodic resistance (mΩ)	17.40	1.93
Cathodic resistance (mΩ)	—	2.90
Anodic capacitance (F)	2.13	14.50
Cathodic capacitance (mF)	—	407

Proposed Hierarchical Droop-based Active Power Frequency Characteristics

In this work, a hierarchical active power–frequency characteristic for a hybrid hydrogen electrolyzer system is proposed. The concept is illustrated in Fig. 2, where the red and the blue lines correspond to the characteristics for the PEME stack and AE stack, respectively. Dead-bands of frequency deviation are introduced to prevent unnecessary cycling of electrolyzers under minor frequency deviations, thereby reducing lifetime stress and efficiency losses. They also ensure coordinated activation between PEME and AE. The dead-band for the PEME stack is $[-f_d^{PEME},+f_d^{PEME}]$, which is less than the dead-band for the AE stack, which is $[-f_d^{AE},+f_d^{AE}]$. This is because the PEME has faster dynamics compared to AE, allowing it to handle smaller frequency deviations and mitigate the frequent triggering of the AE stack's dynamic changes in input power. Outside the frequency dead-bands, linear droop characteristics are adopted for both the electrolyzer stacks to support both over-frequency and under-frequency grid conditions. However, in selecting the droop constants, the slowness of the AE technology is taken into consideration, such that $|k_d^{AE}|<|k_d^{PEME}|$, where $|k_d^{AE}|$ and $|k_d^{PEME}|$ are the droop constants for AE and PEME, respectively. This hierarchical coordination ensures that PEME primarily contributes to fast frequency stabilization while AE provides extended support under larger deviations. By explicitly shaping the active power–frequency response of each technology, the scheme leverages their complementary strengths.

Fig 2. Proposed hierarchical droop-based active power-frequency characteristics for the hybrid PEME and AE system.

Consequently, the overall hybrid system achieves improved frequency resiliency without compromising hydrogen production efficiency. For an upward frequency deviation, $\delta f>f_d^{PEME}$, the PEME will start increasing its power input following the positive droop characteristic, $+k_d^{PEME}$ up to its maximum capacity $P_{max}$. For further higher frequency deviations, $\delta f>f_{max}^{PEME}$, the AE's dynamic power input changes following its droop characteristics, $+k_d^{AE}$ up to a frequency deviation of $f_{max}^{AE}$, when the AE's input power saturates to the maximum rating. Similarly, for downward frequency deviations, $\delta f<-f_d^{PEME}$, the PEME stack's dynamic power input reduces following the negative linear droop, $-k_d^{PEME}$ up to $-f_{max}^{PEME}$ and beyond that, the AE stack will reduce its power consumption at a rate $-k_d^{AE}$up to $-f_{max}^{AE}$. Note that under normal operating conditions and within the dead-bands, both the electrolyzers consume a constant power $P_0$ for hydrogen production. Moreover, the condition $|f_{max}^{PEME}|=|f_d^{AE}|$ indicates that the AE provides dynamic support once the PEME reaches its operating limits. Mathematically, the characteristics can be expressed in equations (1) and (2) as follows:

For PEME:

(1) $$\begin{array}{r}{P}_{ref}^{PEME}=\{\begin{array}{c}{P}_{min};\delta f<-f_{max}^{PEME}\\ -k_d^{PEME}\cdot \delta f;-f_{max}^{PEME}<\delta f<-f_d^{PEME}\\ P_0;-f_d^{PEME}<\delta f<f_d^{PEME}\\ k_d^{PEME}\cdot \delta f;f_d^{PEME}<\delta f<f_{max}^{PEME}\\ P_{max};f_{max}^{PEME}<\delta f\end{array}\end{array}$$

For AE:

(2) $$\begin{array}{r}{P}_{ref}^{AE}=\{\begin{array}{c}{P}_{min};\delta f<-f_{max}^{AE}\\ -k_d^{AE}\cdot \delta f;-f_{max}^{AE}<\delta f<-f_d^{AE}\\ P_0;-f_d^{AE}<\delta f<f_d^{AE}\\ k_d^{AE}\cdot \delta f;f_d^{AE}<\delta f<f_{max}^{AE}\\ P_{max};f_{max}^{AE}<\delta f\end{array}\end{array}$$

System Description and Results

This section illustrates the test system with the proposed power management strategy and the results of the proposed hierarchical droop-based control of the HES. The system under study, illustrated in Fig. 3, implements a power management strategy for a HES where both electrolyzer technologies are integrated with the grid. Each electrolyzer stack is rated at 750 kW, consisting of three 250 kW units. Since an electrolyzer is a low-voltage device, integration of a hydrogen electrolyzer requires dc-dc converter that steps the available higher dc voltage to a suitable value. The stacks are supplied from the grid through a transformer, filters, a rectifier, and two separate buck converters. The rectifier is controlled using proportional–integral (PI)–based synchronous reference frame control. The decentralized PI controllers in Fig. 3 generate the required control signals for the buck converters. The active power–frequency characteristics proposed in Section III are employed to generate the reference power for both PEME and AE stacks. Grid frequency deviation, measured via a PLL, determines the reference power for each electrolyzer technology. Finally, the reference powers are compared with the actual powers of the PEME and AE stacks, and the resulting error signals are fed into the PI controllers. An electrolyzer connected to the grid can act as a dynamic load, capable of adjusting power input dynamically and hence providing frequency support. This mitigates the need for load shedding and generation shedding in grid-frequency deviation scenarios. The electrolyzer is expected to generate hydrogen continuously to supply the processes that require a continuous hydrogen supply. Being a costly technology, electrolyzers should not only support frequency services occasionally but also maintain a constant hydrogen generation. Therefore, during normal operating conditions, both electrolyzers consume 400 kW of power, maintaining reserves of 350 kW for supports during upward frequency deviations and 400 kW for downward frequency deviations. The studies are performed in the MATLAB/Simulink platform for both over-frequency and under-frequency scenarios. The frequency deviation dead-bands for the PEME and AE are chosen to be $[-0.1Hz,+0.1Hz]$ and $[-0.3Hz,+0.3Hz]$, respectively [8].

Fig. 3. Hybrid electrolyzer system with the proposed power management strategy.

Fig. 4 presents the dynamic responses of the HES when subjected to an under-frequency disturbance. These results are evaluated against the proposed hierarchical power-frequency droop control characteristics illustrated in Fig. 2. At $t\approx 2.0s$, the system experiences a sudden frequency drop of $-0.2Hz$ from the nominal value of 60 Hz as shown in Fig. 4a. In accordance with the proposed characteristics, the PEME, which is assigned a narrower frequency dead-band and a steeper droop slope, is the first to respond. This is clearly observed in Fig. 4b, where PEME active power input decreases from its pre-disturbance value to closely follow the new reference trajectory. The transient mismatch between the PEME power and its reference is minimal, highlighting the fast dynamic capability of the PEME technology. In contrast, the AE power output, as shown in Fig. 4c, remains essentially constant throughout the event. This behavior is consistent with the wider frequency dead-band assigned to AE in the proposed scheme in Fig. 2. Since the observed frequency deviation does not exceed the PEME stack’s range, AE support is not activated. This selective engagement ensures that the slower-responding AE is not unnecessarily cycled for smaller disturbances, thereby improving its operational lifespan and reducing control effort. The voltage and current profiles as illustrated in Fig. 4d and 4e further confirm this staged engagement strategy. The PEME terminal voltage and current show significant variation during the disturbance, corresponding to its active participation in frequency support. Meanwhile, AE voltage and current remain largely unchanged, indicating that it remained in a steady operating state. The dynamic response of the hybrid system for a larger frequency deviation of $\delta f=-0.4Hz$ at $t\approx 2.0s$ is shown in Fig. 5. The frequency deviation sustain from $t\approx 2.0s$ to $t\approx 3.0s$ as depicted in Fig. 5a. According to the proposed characteristics, for such frequency deviation of −0.4 Hz, the PEME stack’s input power saturates to the minimum value following the reference power, and the AE stack’s input power now drops following its droop characteristics. The comparisons of the voltage and current responses of the PEME and the AE stacks are depicted in Fig. 5d and 5e, support the dynamic response of the two electrolyzer technologies. In this case, both the PEME and AE stacks’ voltage and current drops, and the current through the PEME drops to zero. The performance of the proposed power management strategy is also validated during grid over-frequency conditions, as shown in Figs. 6 and 7. For an upward frequency deviation of $+0.2Hz$ at $t\approx 2.0s$ as shown in Fig. 6, the dynamic input power of the PEME increases following the reference trajectory. However, the input power to the AE stack remains unchanged since the frequency deviation is well below the dead-band assigned for the AE stack. Both the voltage and current responses support such dynamic response of the PEME stack and steady response of the AE stack, as shown in Fig. 6d and 6e, respectively. Moreover, the responses with the proposed power management strategy for a $+0.4Hz$ frequency deviation are explained in Fig. 7a–e. According to the proposed droop characteristics, the PEME stack’s input power saturates to its maximum rated value, as shown in Fig. 7b, which is supported by the increase in voltage and current of the PEME. Moreover, since the dead-band of AE is now being exceeded, AE’s input power also changes following its reference value as shown in Fig. 7c. The corresponding voltages and currents of the PEME and AE, as shown in Fig. 7d and 7e, also validate such responses.

Fig. 4. Performance of the proposed hierarchical droop-based strategy: (a) grid frequency deviation of −0.2 Hz, (b) first-layer dynamic power response by the PEME stack, (c) steady power response by the AE stack, (d) voltages across the PEME and AE stacks, and (e) currents through the PEME and AE stacks.

Fig. 5. Performance of the proposed hierarchical droop-based strategy: (a) grid frequency deviation of −0.4 Hz, (b) second-layer dynamic power response by the PEME stack, (c) first-layer dynamic power response by the AE stack, (d) voltages across the PEME and AE stacks, and (e) currents through the PEME and AE stacks.

Fig. 6. Performance of the proposed hierarchical droop-based strategy: (a) grid frequency deviation of +0.2 Hz, (b) first-layer dynamic power response by the PEME stack, (c) steady power response by the AE stack, (d) voltages across the PEME and AE stacks, and (e) currents through the PEME and AE stacks.

Fig. 7. Performance of the proposed hierarchical droop-based strategy: (a) grid frequency deviation of +0.4 Hz, (b) second-layer dynamic power response by the PEME stack, (c) first-layer dynamic power response by the AE stack, (d) voltages across the PEME and AE stacks, and (e) currents through the PEME and AE stacks.

Conclusions

This paper proposes a power management strategy for a grid-connected hybrid electrolyzer system comprising of a proton exchange membrane electrolyzer stack of 750 kW and an alkaline electrolyzer stack of 750 kW. The hybrid configuration is intended to provide dynamic responses during grid-frequency disturbances. A droop-based hierarchical power management strategy for the hybrid technology is proposed in this paper, aiming to prioritize PEME for supporting smaller and more frequent grid-frequency disruptions due to its faster dynamics as compared to AE. Higher frequency deviations, which are less frequent in power system, are handled by the AE stack, provided that the limits of the PEME stack are exceeded. Such layering of the dynamic support has been accomplished by separating the frequency deviation dead-bands of the two electrolyzer technologies. Results show that when the frequency deviation is within the dead-band, the electrolyzer technologies maintain steady operation. In case of frequency deviations exceeding the corresponding dead-band of each technology, both the PEME and AE can provide dynamic power response according to the proposed droop characteristics. With the adoption of such a power management strategy, it is possible to link the dynamic responses of the electrolyzer technologies with the grid-frequency deviations and thus overcome the limitations of each technology, trading off between faster response and cost. Future work will focus on evaluating the proposed scheme within a benchmark system to enhance frequency resiliency.

Conflict of Interest

The authors declare that there are no known financial or personal conflicts of interest that could have influenced the work reported in this paper.