Optimization of Renewable Energy Hydrogen Production Systems Using Volatility Improved Multi-Objective Particle Swarm Algorithm

Abstract

Optimizing the energy structure to effectively enhance the integration level of renewable energy is an important pathway for achieving dual carbon goals. This study utilizes an improved multi-objective particle swarm optimization algorithm based on load fluctuation rates to optimize the architecture and unit capacity of hydrogen production systems. It investigates the optimal configuration methods for the architectural model of new energy hydrogen production systems in Xining City, Qinghai Province, as well as the internal storage battery, ALK hydrogen production equipment, and PEM hydrogen production equipment, aiming at various scenarios of power sources such as wind, solar, wind–solar complementary, and wind–solar–storage complementary, as well as intermittent hydrogen production scenarios such as hydrogen stations, hydrogen metallurgy, and continuous hydrogen production scenarios such as hydrogen methanol production. The results indicate that the fluctuation of hydrogen load scenarios has a significant impact on the installed capacity and initial investment of the system. Compared with the single-channel photovoltaic hydrogen production scheme, the dual-channel hydrogen production scheme still reduces equipment capacity by 6.04% and initial investment by 6.16% in the chemical hydrogen scenario with the least load fluctuation.

1. Introduction

With societal development, energy crises are becoming increasingly severe, prompting scientists to actively promote the utilization of renewable energy technologies [1]. Photovoltaic and wind power generation can be considered as complex energy flows in energy systems [2]. Influenced by the environment, the supply processes of these two forms of electricity exhibit instability and demonstrate different characteristics on different time scales; hence, wind and solar resources exhibit a certain degree of complementarity in time sequence [3]. The characteristics of wind and solar power generation exhibit nonlinear development over time, and although storage batteries are installed in the system, the supply of green energy is still significantly influenced by the local environment [4]. Owing to the variability and uncertainty of wind and solar power generation, many regions are beginning to require system operators to install a certain capacity of storage batteries to smooth out electricity output and meet grid consumption demands [5]. Therefore, enhancing the consumption capacity of green electricity requires improving the energy quality of renewable energy. It is imperative to upgrade relevant technologies and industries. Optimizing the configuration and coordinating control of different equipment in the key links of generation, load, and storage is crucial for reducing overall system investment and improving performance and returns.

Hydrogen is a zero-carbon fuel, and hydrogen-based fuels have the potential for large-scale and long-term cyclic energy storage [6]. Zhang et al. argue that the production of green hydrogen using renewable energy sources, such as wind and solar power, has become an important approach for clean electricity consumption, storage, and utilization [7]. Li et al. suggest that the technology for converting green electricity to hydrogen is considered a feasible route to achieve carbon neutrality, enabling partial or complete decarbonization in the chemical, steel, and transportation sectors [8]. Yu et al. demonstrated that optimization of the capacity configuration and operation mode of hydrogen production equipment can effectively enhance the hydrogen production capacity and economic benefits of the system [9,10]. Alkaline (ALK) water electrolysis has the advantages of large single-cell capacity and relatively low investment per unit of hydrogen production capacity, while proton exchange membrane (PEM) water electrolysis exhibits characteristics such as fast load response, wide adjustment range, and strong resilience to fluctuations [11,12]. With the increasing scale of wind and solar power generation, it is necessary to conduct optimization research for different hydrogen energy scenarios.

Particle swarm optimization (PSO) has been applied in various fields, including grid operation and the integration of renewable energy systems [13]. Fukuyama et al. compared the original PSO, evolutionary PSO, and adaptive PSO, all of which can be used to plan problems in energy systems [14]. Zhang et al. employed a multimodal optimization algorithm to optimize different types of energy systems, enhancing diversity in the solution space, which helps reduce solution time. However, this method struggles to converge for discrete problems [15]. Aghajani et al. utilized a multi-objective particle swarm algorithm based on fuzzy control to solve small-scale renewable energy systems; however, the load-leveling process has not been addressed [16]. Arezoo et al. optimized the configuration of hybrid systems, including wind, solar, and fuel cells, using the QPSO algorithm. Although it can achieve the goal of minimizing energy costs to the greatest extent possible, the optimization process did not consider the randomness of natural resources, indicating that the robustness of energy hubs still needs improvement [17].

In summary, hydrogen energy is a promising medium for the sustainable utilization of wind and solar power. Current research efforts aimed at enhancing the integration of wind and solar renewable energy have mainly focused on improving new energy generation or hydrogen production technologies. However, specific optimization techniques for hybrid wind–solar–hydrogen energy systems have not yet been clearly defined. Therefore, this study incorporates the volatility of hydrogen energy loads from different scenarios into a multi-objective particle swarm optimization scheme. The optimization objectives include the capacity of equipment in different energy sectors and energy system investments, with energy–electricity and energy–hydrogen balance as constraints. Through an improved multi-objective optimization algorithm, the diversity of energy systems is reflected by incorporating volatility data from different scenarios into the position update process of the particle swarm. This approach enhances the specificity of system configuration and the smoothing effect on energy fluctuations from a data structural perspective, thereby achieving specialized system configuration and attenuating energy fluctuations.

2. Analysis of Natural Resources and Hydrogen Energy Loads

This study takes Xining, Qinghai Province, as an example to calculate the minimum time scale of the process as 1 h. Figure 1 shows the installed capacity (1 kW) of wind and solar power generation. It is evident that solar resources in Qinghai are superior to wind resources. Currently, commercialized hydrogen load scenarios mainly include industrial transport vehicle refueling stations (HS1L, Hydrogen Station 1 Load in Figure 2), passenger vehicle refueling stations (HS2L, Hydrogen Station 2 Load in Figure 2), chemical industry hydrogen load (CIHL, Figure 2), and electricity generation hydrogen load (EGHL, Figure 2), with typical energy characteristic curves [18,19,20]. The load characteristic data are shown in

Table 1. Hydrogen Load in Different Scenarios.

	Hydrogen Station 1 Load	Hydrogen Station 2 Load	Chemical Industry Hydrogen Load	Electricity Generation Hydrogen Load
Daily Load (kg/d)	898	484	1154	997
Peak Load (kg/h)	54	43	59	80
Volatility	0.0467	0.0723	0.016	0.0684

. Industrial transport vehicles typically bear a large workload when transporting or performing heavy industrial tasks, resulting in higher energy consumption of their power systems and requiring more hydrogen to meet their operational needs. In the time series, the overall load demand of industrial transport vehicle refueling stations is higher during the day, with small peaks in hydrogen energy load at 8:00 in the morning and 8:00 in the evening. Due to the nature of heavy industrial transport work, some vehicles still operate at night. The load characteristics of passenger vehicle refueling stations significantly fluctuate day and night, presenting an overall step-shaped curve. During the day, owing to the high demand for power performance and cruising range of passenger vehicles, combined with user habits, there is a peak period of refueling from 2:00 to 6:00 p.m.

Hydrogen energy loads in chemical processes such as methanol production and metallurgy exhibit relatively low volatility. Chemical production typically employs highly optimized process flows, which are meticulously adjusted during the design and operation stages to achieve efficient and stable production. Compared to other industries, the variations in reaction conditions and process parameters during chemical production processes are relatively small, thereby reducing the volatility of hydrogen energy loads. As seen in Figure 2, there is a parabolic characteristic at 7:00 a.m. and 9:00 p.m.; however, there is no significant fluctuation during these time periods. Such continuous production load characteristics, compared to intermittent characteristics, are more conducive to the overall energy system stability. The hydrogen energy load characteristics for electricity generation (Figure 2) exhibit a double-peak curve, reflecting the fluctuation of hydrogen demand with electricity demand. The peak–valley difference reflects the power demand differences between high and low load periods, which are crucial for hydrogen load planning and operation. The lowest hydrogen load in the figure, from 3:00 to 4:00 a.m., is 12 kg, while the hydrogen load reaches 80 kg at 9:00 p.m. Additionally, a sustained high-load state occurs from 11:00 a.m. to 3:00 p.m. Such significant fluctuations pose significant challenges to the stable supply of energy.

3. Different Power Source Scenarios for Hydrogen Production System Architecture and Mathematical Models

3.1. Mathematical Model of Wind Power Generation

The system includes a model of wind power generation, where the output power of the wind turbine is represented as:

$$P_{wind}=\frac{1}{2}\times \rho \times A\times v^3\times C_p$$

(1)

where P_wind represents the output power of the wind turbine, ρ represents the air density, A represents the blade area of the wind turbine, v represents the wind speed, and C_p represents the power coefficient of the wind turbine.

The wind energy utilization coefficient is closely related to the wind turbine startup parameters, as shown in the following equation:

$$C_p=0.5176(\frac{116}{{\lambda }_i}-0.4\beta -5)e^{-21/{\lambda }_i}+0.0068\lambda$$

(2)

$$\frac{1}{{\lambda }_i}=\frac{1}{\lambda +0.08\beta }-\frac{0.0035}{{\beta }^3+1}$$

(3)

$$\lambda =\frac{\omega R}{v}$$

(4)

where λ represents the tip-speed ratio, ω represents the rotor angular velocity in radians per second (rad/s), and β represents the pitch angle of the rotor blade in radians (rad).

3.2. Photovoltaic Power Generation Mathematical Model

Due to the nonlinear characteristic curve of photovoltaic (PV) panels, the maximum output power varies under different radiation intensities and temperatures. Maximum power point tracking (MPPT) is adopted to ensure that the PV panel output remains at the optimal power point. According to the principle of the current-voltage (IV) characteristic, the output power P_s of the PV system is expressed as follows:

$$P_s=I_h{m}_b{U}_s-I_r{m}_b{U}_s-I_r{e}^{\frac{q{U}_s}{m_{c}AK{T}_s}}{m}_b{U}_s$$

(5)

where m_b and m_c represent the number of photovoltaic cells in parallel and series, I_h and I_r represent the photocurrent and reverse saturation current of the diode (A); U_s stands for the output voltage of the photovoltaic cells (V), T_s represents the operating temperature of the photovoltaic cells (°C), q represents the elementary charge of an electron, which is 1.6 × 10⁻²⁹ C, A is the ideality factor of the PN junction, ranging from 1 to 5, K denotes the Boltzmann constant, which is 1.381 × 10⁻²³ J/K.

The photocurrent model I_h is as follows:

$$I_h=G\times A_{pv}\times \eta$$

(6)

where G represents the irradiance, A_pv represents the surface area of the photovoltaic cell, and η represents the photovoltaic conversion efficiency.

The model I_r for the reverse saturation current, considering the temperature effect, is represented as follows:

$$I_r \left(T\right)=I_0\times (\frac{T}{T_{ref}}{)}^{3/q}$$

(7)

where I_r represents the reverse saturation current considering the temperature effect, T represents the temperature, T₀ represents the reference temperature (usually 25 °C), and q represents the elementary charge.

3.3. Energy Storage Battery and Hydrogen Production System Model

The mathematical model of the battery energy storage system is as follows:

$$P_{bat}={\eta }_{bat}\times (P_{wind}+P_{PV})$$

(8)

where P_bat represents the power of the battery energy storage system, and η_bat represents the efficiency of the battery energy storage system.

The rate of change of electrical energy in the battery energy storage system is as follows:

$$\frac{d{E}_{bat}}{dt}=P_{bat}-P_{load}$$

(9)

where P_load represents the system load power.

The mathematical models for ALK and PEM electrolysis hydrogen production systems are as follows:

$$\mathrm{ALK}: P_{H2\_ALK}={\eta }_{elec\_ALK}\times V_{ALK}\times I_{ALK}$$

(10)

where P_{H2_ALK} represents the power of ALK electrolysis hydrogen production, η_{elec_ALK} represents the efficiency of ALK electrolysis, V_ALK represents the voltage of the ALK electrolysis cell, and I_ALK represents the current of the ALK electrolysis cell.

$$\mathrm{PEM}: P_{H2\_PEM}={\eta }_{elec\_PEM}\times V_{PEM}\times I_{PEM}$$

(11)

where P_{H2_PEM} represents the power of PEM electrolysis hydrogen production, η_{elec_PEM} represents the efficiency of PEM electrolysis, V_PEM represents the voltage of the PEM electrolysis cell, and I_PEM represents the current of the PEM electrolysis cell.

3.4. System Architecture Design

This subsection presents the architectures of single-channel and dual-channel hydrogen energy systems. Figure 3a–c depict the single-channel ALK hydrogen production, while Figure 3d shows the hybrid ALK + PEM hydrogen production. Each system incorporates different modules such as wind power generation, photovoltaic power generation, and battery energy storage. The hybrid hydrogen production module in Figure 3d is designed to be “grid-connected without grid dependence”, integrating wind and photovoltaic renewable energy forms and utilizing battery energy storage. The study considers a 3% network loss for different energy flows, a rated charge–discharge efficiency of 95% for battery energy storage, a discharge capacity lower limit of 10%, a load adjustment range of 20–100% for ALK electrolysis cells, and a load adjustment range of 0–120% for PEM electrolysis cells. The system utilizes wind and solar power synergistically for power supply, battery storage, and hydrogen production. A key feature is the ability to enhance hydrogen production stability through two different electrolysis methods (ALK and PEM) under “grid-connected without grid dependence”. The synergistic generation of multiple energy sources improves overall energy utilization efficiency and enhances system robustness by enabling both independent operation and grid connection.

3.5. Mathematical Optimization Modeling of System Capacity

By defining decision variables and objective functions, a particle swarm multi-objective optimization problem was established to solve the capacity of different modules in the system. Considering equal hydrogen production capacities, the system objectives were defined in two ways. The first objective is to minimize the system equipment capacity to improve overall efficiency. The optimization process considers an hourly time scale with a yearly time series of 8760 h. The second objective is to minimize the initial investment of equipment to reduce the economic cost of the system. The constraints of the system include energy balance and equipment capacity limitations which play a critical role in the feasibility and stability of the system.

The decision variables are defined as follows:

X_wind, Wind power generator unit capacity;

X_pv, Photovoltaic (PV) generator unit capacity;

X_battery, Battery energy storage system capacity;

X_alk, Alkaline electrolysis (ALK) hydrogen production unit capacity;

X_pem, Proton exchange membrane (PEM) electrolysis hydrogen production unit capacity.

The objective function for total investment cost is as follows, where the unit cost of equipment refers to

Table 2. Equipment Cost Reference Table.

Item Name	Unit Price (CNY)
Wind power generation	6500/kW
Photovoltaic generation	4000/kW
Battery	1500/kW
ALK hydrogen production units	4000/kW
PEM hydrogen production units	12,000/kW

$$f_1\left(X\right)=c_{wind}·X_{wind}+c_{PV}·X_{PV}+c_{battery}·X_{battery}+c_{alk}·X_{alk}+c_{pem}·X_{pem}$$

(12)

where c_wind, c_pv, c_battery, c_alk, and c_pem represent the unit capacity prices of wind power generation, photovoltaic power generation, battery energy storage, ALK hydrogen production units, and PEM hydrogen production units (CNY/k·Wh), respectively.

The objective function of the system equipment capacity is as follows:

$$f_2\left(X\right)=X_{wind}+X_{PV}$$

(13)

$$f_3\left(X\right)=X_{battery}$$

(14)

$$f_4\left(X\right)=X_{alk}+X_{pem}$$

(15)

Electricity balance constraint:

$$\sum _{i=1}^{8760}{ (E}_{wind \left(i\right)}+E_{PV \left(i\right)}+E_{battery \left(i\right)}+E_{grid \left(i\right)}=E_{alk \left(i\right)}+E_{pem \left(i\right)}+E_{demand \left(i\right)})$$

(16)

where E_wind represents the wind power generation, E_PV represents the photovoltaic power generation, E_battery represents the battery charging and discharging, E_grid represents the grid electricity consumption, E_alk represents the electricity demand of the ALK electrolyzer, E_pem represents the electricity demand of the PEM electrolyzer, E_demand represents the self-consumption of the system, and i represents the dynamic time series.

Hydrogen production capacity constraint:

$$\sum _{i=1}^{8760}{ (X}_{alk \left(i\right)}+X_{pem \left(i\right)}\ge H_{demand\left(i\right)})$$

(17)

where H_demand represents the hydrogen demand load.

The available capacity constraints for wind and photovoltaic power generation are as follows:

$$\sum _{i=1}^{8760} (X_{wind\left(i\right)}\le X_{wind, \max \left(i\right)})$$

(18)

$$\sum _{i=1}^{8760}{ (X}_{PV \left(i\right)}\le X_{PV, \max \left(i\right)})$$

(19)

The constraint condition for battery energy storage capacity is as follows:

$$\sum _{i=1}^{8760}{ (X}_{battery \left(i\right)}\le X_{battery, \mathrm{m}\mathrm{a}\mathrm{x} \left(i\right)})$$

(20)

where

$$X_{wind},X_{PV},X_{battery},X_{alk},X_{pem}\ge 0$$

(21)

$$Optimization objective: Minimize, \left\{f_1\left(X\right)\right., f_2\left(X\right), f_3\left(X\right), \left.f_4\left(X\right)\right\}$$

(22)

3.6. Optimization Process

The optimization process employs a multi-objective particle swarm optimization algorithm, incorporating the fluctuation rate data of different load scenarios into the optimization process (Figure 4a) to enhance the specificity of the optimization process. By utilizing the load fluctuation rate to adjust the learning factors, the iteration process of individuals and groups is intervened separately, thus involving the load characteristics of different scenarios in the mathematical optimization stage of particle position updating. Additionally, by adjusting the velocity constraint factor, the features of load fluctuation rate are introduced into the particle velocity updating process, thereby integrating into the particle swarm optimization from different dimensions. Initially, all particles are initialized, and {f₁ (X), f₂ (X), f₃ (X), f₄ (X)} are computed to form the Pareto initial solution set. The particle position updating process is formulated as Equation (23), while the load fluctuation rate calculation is represented by Equation (24).

$$x_{i, d}\left(t+1\right)=w·x_{i, d}\left(t\right)+c_1·r_1·\left(p_{i, d}-x_{i, d}\left(t\right)\right)+c_2·r_2\left(p_{gbest, d}-x_{i, d}\left(t\right)\right)$$

(23)

In this context, $x_{i, d}\left(t+1\right)$ represents the next-generation update of particle i in dimension d of the position; w represents the inertia weight. The coefficients c₁ and c₂ are acceleration coefficients that, respectively, regulate the learning behavior of individuals and the group. They intervene in the position updating through $c_1=c_1·R; c_2=c_2·R$, altering the learning factors based on the load volatility, thus influencing the individual and group iteration processes differently. r₁ and r₂ are random numbers uniformly distributed within the range [0, 1]; p_i,d represents the individual best position of particle i in dimension d; p_gbest,d represents the global best position in dimension d within the particle swarm.

The fluctuations generated by photovoltaic and wind power units, as well as hydrogen load, are controlled and allocated by the system to AEL and PEM for electrolytic hydrogen production. To analyze the impact of wind and solar power generation and load fluctuations on the characteristics of different unit capacity configurations, the average fluctuation rate of unit output is defined as follows [18]:

$$R_{f, \Delta t}^{ave}=\frac{1}{t}{\sum }_{i=1}^t{R}_{f, \Delta t}^i=\frac{1}{t}{\sum }_{i=1}^t\frac{P_{\mathit{max}, \Delta t}^i-P_{\mathit{min}, \Delta t}^i}{P_{m, \Delta t}^i}$$

(24)

where Rⁱ_f,Δt represents the power fluctuation rate at time i with a time scale of Δt units; Pⁱ_max,Δt, Pⁱ_min,Δt, and Pⁱ_m,Δt, respectively, represent the maximum, minimum, and average power values within the time scale Δt at that moment.

The process of updating particle velocities is as follows:

$$v_{i, d}\left(t+1\right)=\chi \left[v_{i, d}\left(t\right)+c_1·r_1\left(p_{i, d}-x_{i, d}\left(t\right)\right)+c_2·r_2\left(p_{gbest, d}-x_{i, d}\left(t\right)\right)\right]$$

(25)

where $v_{i, d}\left(t+1\right)$ represents the update of particle i in dimension d velocity; χ is the velocity constraint factor, used to control the range of velocities. Considering the constraint of energy fluctuation rate R added in the position updating, to balance the convergence speed of particles, $\chi =\frac{\chi }{R}$ is employed to drive the velocity updating process again. Thus, the updating range of particles in the time dimension is similar to the fluctuation rate of the load, which can enhance the convergence speed.

The fitness of the particles is calculated as follows:

$$I\left(X_i\right)=\frac{\left|\left[f_1\left(X_j\right)-f_1\left(X_k\right)\right]\right|}{f_{1max}}+\frac{\left|\left[f_2\left(X_j\right)-f_2\left(X_k\right)\right]\right|}{f_{2max}}+\frac{\left|\left[f_3\left(X_j\right)-f_3\left(X_k\right)\right]\right|}{f_{3max}}+\frac{\left|\left[f_4\left(X_j\right)-f_4\left(X_k\right)\right]\right|}{f_{4max}}$$

(26)

where X_j and X_k are the nearest particles to X_i; f_m (X_j) represents the value of the m-th objective for particle X_j; f_m,max is the maximum value of the m-th objective.

Using dynamic density distance, the Pareto set is updated, retaining distant solutions while eliminating closer ones. If the stopping conditions (preset computational accuracy or number of iterations) are satisfied, the search stops. When capacity requirements and economic indicators are both set as decision variables, constructors tend to favor the most economically optimal solution. Therefore, in this study, the final output results are the economically optimal solutions among the Pareto frontiers.

After improving the particle swarm algorithm based on load volatility, three models were defined: rR-MOPSO, which only intervenes in particle position updates; χR-MOPSO, which only intervenes in particle velocity updates; and RR-MOPSO, which intervenes in both particle position and velocity updates simultaneously. As shown in Figure 4(bB), the enhanced particle swarm scheme achieves good results after 90 iterations, and convergence is essentially achieved after 100 iterations. In the mid-term process approaching convergence, the velocity in the rR-MOPSO model decreases significantly compared to the other two models since it only modifies particle position variables. In Figure 4(bA), it can be observed that the initial fitness of the χR-MOPSO model is consistently lower, benefiting from the significant amplification of particle movement speed due to load volatility, resulting in relatively good results in the early stage of convergence. Although the initial fitness of RR-MOPSO is relatively high, it converges rapidly. If larger-scale unit capacity configurations are performed, its compatibility superiority will be demonstrated, enhancing computational efficiency.

4. Results and Discussion

4.1. Optimization Results of Wind Power Generation–ALK Electrolytic Hydrogen Production

The capacity allocation and initial investment for the combination of wind power generation and ALK under different hydrogen energy scenarios are illustrated in Figure 5. Combined with the load fluctuation conditions of different scenarios, as depicted in Figure 2 and Table 1, it is evident that the load volatility has a significant impact on the capacity allocation of systems without energy storage. For instance, the average daily load of HS1L is 85% higher than that of HS2L, whereas the installed wind power capacity is only 26% higher. Similarly, the average daily load for CIHL is 1154 kg, while for power generation hydrogen energy, it is 997 kg. However, on an hourly basis, the load characteristics of CIHL are relatively stable compared to those of EGHL. Consequently, the system requires a smaller power adjustment range during peak and off-peak periods, resulting in a smaller total installed capacity.

4.2. Optimization Results of Photovoltaic Generation-ALK Electrolytic Hydrogen Production

The combination of photovoltaic generation and ALK exhibits different capacity configurations and initial investment scenarios across various hydrogen energy scenarios, as depicted in Figure 6. Compared to the wind power generation–ALK electrolytic hydrogen production scheme, there is a proportionate reduction in the capacity of electricity generation equipment in each scenario. Furthermore, the initial cost of the photovoltaic generation–hydrogen production system is significantly lower than that of the wind power generation–hydrogen production system. Particularly in scenarios with relatively stable hydrogen energy loads, the investment in electricity generation equipment is reduced by up to 62%.

4.3. Optimization Results of Photovoltaic Generation–ALK Electrolytic Hydrogen Production

In this section, wind power and photovoltaic power are utilized with batteries as backup power sources, while ALK is employed as hydrogen production equipment. The capacity configuration and initial investment for the wind power–ALK–energy storage scenario are shown in Figure 7, while those for the photovoltaic power–ALK–energy storage scenario are displayed in Figure 8. In the same hydrogen energy load scenario, the implementation of energy storage systems effectively reduces the capacity of the power generation units. The energy storage systems enhance the comprehensive energy utilization efficiency of the system by enabling short-term energy shifting. In the four hydrogen energy scenarios, the installed capacity of wind power generation units increases by 90% to 104% relative to photovoltaic power generation units. By comparing Figure 7 and Figure 8 with Figure 5 and Figure 6, it is observed that in the four scenarios with energy storage modes, the capacity of the power generation units decreases by 29% to 46% compared to scenarios without energy storage. The equipment investment in different wind power scenarios decreases by 23% to 39%, and that in different photovoltaic power scenarios decreases by 14% to 24%. Particularly in the scenario with significant fluctuations in hydrogen energy load, such as the scenario with HS2L, the system capacity decreases by 48% and the system investment decreases by 38% compared to the non-energy storage mode. However, in the scenario of CIHL with smaller fluctuations in load, the capacity of the energy storage mode system decreases by 12%, and the system investment decreases by 9%, indicating a significant impact of hydrogen energy load fluctuations on system capacity and economic feasibility.

4.4. Optimization Results of Wind Power Generation–Photovoltaic Generation–Battery Energy Storage–Grid-Connected without Grid Dependence–ALK–PEM Electrolytic Hydrogen Production

The “grid-connected without grid dependence” hybrid energy system adopts a mode of wind power generation, photovoltaic power generation, battery energy storage, and ALK–PEM dual-channel electrolytic hydrogen production. The unit capacity allocation and initial investment scheme are illustrated in Figure 9. The total investment of the hybrid system is lower in all scenarios compared to the wind power–ALK and photovoltaic–ALK schemes. The main reason for this is to adapt the system to the fluctuation of hydrogen load, which changes the capacity of hydrogen production and energy storage equipment. The short-term shifting characteristic of the energy storage equipment further reduces the system capacity and investment indicators. As shown in the comparison table of the reduction ratio between dual-channel and single-channel hydrogen production in different scenarios (

Table 3. Comparison of capacity and investment between dual-channel hydrogen production and single-channel hydrogen production in different scenarios.

System Scheme	Hydrogen Station 1 Load	Hydrogen Station 2 Load	Chemical Industry Hydrogen Load	Electricity Generation Hydrogen Load
Hybrid system and wind–ALK hydrogen production system (Generator capacity)	55.69%	65.14%	53.02%	64.03%
Hybrid system and photovoltaic–ALK hydrogen production system (Generator capacity)	11.39%	30.28%	6.04%	28.05%
Hybrid hydrogen production mode and the ALK hydrogen production mode (Hydrogen production unit capacity)	27.57%	48.69%	12.77%	46.27%
Hybrid system and wind–ALK hydrogen production system (Investment cost)	65.53%	66.89%	64.59%	66.11%
Hybrid system and photovoltaic–ALK hydrogen production system (Investment cost)	8.63%	12.25%	6.16%	10.18%

), the hybrid system with dual-channel hydrogen production mode has greater advantages in terms of system capacity and initial investment compared to the wind power–ALK single-channel hydrogen production. In the case of hydrogen production from electricity, the system capacity reduction ratio of the hybrid system is 64.03%, and the reduction ratio of the initial investment in the scenario of hydrogenation station 2 reaches 66.89%, indicating that in scenarios with large load fluctuations, the system capacity is designed to be larger and the initial investment is higher to meet the load requirements. The comparison of indicators in the CIHL scenario shows that the hybrid system with dual-channel hydrogen production reduces the equipment capacity by 6.04% and the initial investment by 6.16% compared to the photovoltaic–ALK hydrogen production equipment, indicating that the hybrid system still has a significant advantage.

5. Conclusions

This paper reviews the system configuration optimization process with the load characteristics of different hydrogen energy scenarios in the form of volatility. It analyzes the wind and solar power resources and data characteristics in Xining, Qinghai Province, through an improved multi-objective particle swarm optimization algorithm. The main research conclusions are as follows.

The utilization of load fluctuation parameters from various renewable energy scenarios has enhanced the particle swarm optimization process, thereby improving the specificity of the optimization model and significantly enhancing the convergence performance of the optimization algorithm. If there is a significant disparity between the hydrogen load and the local renewable energy volatility, it will result in a substantial increase in the initial investment of the entire project. In such scenarios, configuring suitable energy storage batteries can actually make the system more economical.

The hybrid system with dual-channel hydrogen production exhibits significant advantages in both the level of renewable energy integration and initial investment. Even in scenarios with minimal load fluctuations, such as the chemical hydrogen energy scene, the dual-channel hydrogen production scheme of hybrid energy shows a reduction in equipment capacity of 6.04% compared to the single hydrogen production scheme of photovoltaic energy, with an associated decrease in initial investment of 6.16%.