Optimal scheduling method of photo-electric-thermal-hydrogen microgrid considering degradation of electrolyzer

Abstract

The present application relates to the technical field of off-grid park optimization scheduling, and discloses a light-electricity-heat-hydrogen microgrid DRL optimization scheduling method considering electrolytic cell degradation, which takes the minimum total operation cost of the system as the target, establishes a refined physical model for the off-grid light-heat-electricity-hydrogen coupling microgrid, and includes photovoltaic, photo-thermal, double molten salt heat storage, energy storage battery, electric heater and high-temperature SOEC, decomposes the SOEC operation apparent resistance into irreversible aging term and reversible dynamic response term, and overcomes the technical barrier that the degradation state in the existing single degradation model is unobservable in real-time scheduling; a Markov decision process is established, the state space introduces the electrolytic cell operating current and apparent resistance as the observable variables of degradation, and the action space introduces the electrolytic cell operating temperature as the degradation management control variable; and the TD3 algorithm is used to solve the day-ahead optimal scheduling strategy. The present application realizes effective coordination between the short-term operation economy of the off-grid microgrid and the long-term life protection of the SOEC.

Description

Technical Field

[0001] This invention relates to the field of off-grid park optimization scheduling technology, specifically to a photovoltaic-electrical-thermal-hydrogen microgrid DRL optimization scheduling method that takes into account the degradation of electrolyzers. Background Technology

[0002] Off-grid industrial park microgrids typically integrate photovoltaic power generation, energy storage batteries, molten salt tower solar thermal power generation, and solid oxide electrolysis cell (SOEC) high-temperature hydrogen production equipment. Since the system lacks external grid support, its internal electricity, heat, and hydrogen loads must be entirely provided by local renewable energy sources, placing extremely high demands on multi-energy flow coupling and real-time supply-demand balance. To improve renewable energy consumption and system regulation capabilities, current mainstream technical approaches fall into two main categories: one is to configure energy storage to achieve peak-shifting of electricity use and smooth power fluctuations; the other is to apply surplus renewable energy power to scenarios such as hydrogen electrolysis and clean heating, thus broadening the consumption channels.

[0003] Under long-term, frequent, high-frequency adjustments and high-load fluctuations, the electrolyzer materials within the microgrid undergo irreversible performance degradation. This degradation not only directly reduces the active power conversion efficiency of each device but also shortens the overall lifespan of the equipment, significantly increasing the long-term reset and maintenance costs of off-grid microgrids. Existing research largely focuses on short-term economic optimization and conventional quantity balancing scheduling of park microgrids, lacking a perspective on the life-cycle degradation characteristics of core equipment to construct scheduling models and ignoring the long-term dynamic degradation characteristics of equipment during time-series control. On the other hand, the high-proportion renewable energy optimization scheduling problem, due to its high dimensionality, strong coupling, multiple uncertainties, and strongly nonlinear equipment degradation characteristics, places higher demands on solution methods. Traditional optimization algorithms are computationally inefficient when dealing with complex dynamic constraints, while heuristic algorithms are prone to getting trapped in local optima and have limited generalization ability.

[0004] Furthermore, in the few existing studies involving SOEC degradation modeling, degradation models generally employ a single empirical formula to directly express the apparent resistance of operation as a function of operating temperature, humidity, and cumulative operating time. In such single-formula models, the resistance increase caused by irreversible cumulative material damage and the reversible resistance fluctuations caused by changes in operating conditions (temperature rise and fall, current density fluctuations) are tied together in the same expression and cannot be separated. In offline planning scenarios, this aliasing does not pose an obstacle—planning only needs to estimate the total resistance evolution over the entire lifespan of the equipment to determine the capacity configuration scheme, without needing to distinguish the specific causes of each resistance change. However, in the real-time scheduling scenario of off-grid microgrids, this aliasing will lead to the unobservable degradation state of SOEC during operation: the scheduling agent cannot determine whether the currently observed resistance increase is due to actual irreversible degradation of the equipment (requiring the triggering of life extension protection strategies) or merely a reversible transient response caused by a normal increase in operating temperature (requiring no adjustment to scheduling decisions). The unobservability of degradation information means that any degradation-aware scheduling decision is based on fuzzy information, thereby undermining the effectiveness of the trade-off between short-term operating economics and long-term equipment life protection. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to provide a Deep Reinforcement Learning (DRL) optimization scheduling method for a photoelectric-thermal-hydrogen microgrid that considers electrolyzer degradation. This method establishes a refined scheduling model framework for the microgrid within the industrial park and incorporates real-time performance degradation and lifetime decay models for each electrolyzer into the environmental modeling. By introducing deep reinforcement learning, the scheduling problem is transformed into a Markov Decision Process (MDP), which is solved using the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm. By introducing equipment degradation states into the state space and considering the performance degradation penalty and lifetime cost caused by degradation in the reward function, effective synergy between high-frequency dynamic adjustment and long-term anti-degradation and lifespan extension is achieved. The technical solution is as follows:

[0006] A DRL optimization scheduling method for a photoelectric-thermal-hydrogen microgrid considering electrolyzer degradation includes the following steps:

[0007] Step 1: Construct a physical and mathematical model of the off-grid industrial park microgrid system architecture and core equipment. The off-grid industrial park microgrid has no energy or information interaction with the external power grid and heating network. The physical and mathematical model deeply embeds a performance degradation model caused by SOEC high-temperature operation polarization and material deterioration. The performance degradation model divides the apparent resistance of SOEC operation into irreversible aging terms and reversible dynamic response terms, and embeds the operating temperature limit and temperature change slope limit as constraints.

[0008] Step 2: Establish the Markov decision process of the off-grid industrial park microgrid, including the state space, action space, and reward function;

[0009] The state space includes at least: photovoltaic power generation, SOEC hydrogen production thermal power, electric heater thermal power, thermal storage capacity, SOEC hydrogen production rate, battery energy storage capacity, electrical load demand, hydrogen load demand, thermal load demand, heat obtained by the heat collection field, SOEC real-time operating current, and SOEC operating apparent resistance.

[0010] The operating space includes at least: SOEC hydrogen production power, thermal storage charging and discharging power, battery charging and discharging power, electric heater power, hydrogen storage tank charging and discharging rate, SOEC operating temperature, and power module output power.

[0011] The reward function is constructed based on the overall system operating cost and the penalty for violating constraints.

[0012] Step 3: The agent is interactively trained using the standard TD3 algorithm. The objective function and constraints are determined with the lowest overall system operating cost. The optimal scheduling strategy is obtained by alternately updating the parameters of the Actor network and the Critic network and then applied online.

[0013] The beneficial effects of this invention are:

[0014] 1) This invention overcomes the technical barrier of the unobservable degradation state in real-time scheduling in the existing single formula model by decomposing the apparent resistance of SOEC operation into irreversible aging terms and reversible dynamic response terms. This dual-channel decomposition enables the scheduling agent to separate the irreversible aging term that characterizes the cumulative damage of the material and the reversible dynamic response term that characterizes the instantaneous impact of the operating condition from the apparent resistance l at each time step, thereby obtaining accurate degradation state perception capability. This allows the trade-off decision between short-term operating economy and long-term equipment life protection to be based on reliable information, fundamentally avoiding misjudgment caused by the overlap of operating condition fluctuations and actual aging.

[0015] 2) This invention achieves decoupling and synergy between day-to-day economic balance and long-cycle equipment degradation prevention and life extension: By incorporating penalty terms based on electrolytic cell degradation and curtailment of light, electricity, and load into the design of deep reinforcement learning, it changes the limitation of traditional scheduling that ignores SOEC degradation, which leads to scheduling decisions deviating from the actual situation. In order to verify the effectiveness of the proposed model in long-cycle operation and accurately characterize the real physical characteristics of the system under complex operating conditions, dynamic load and thermal stress changes under variable current density and variable temperature conditions are simulated based on the system lifetime evolution path, and typical stage dynamic degradation assessment examples of SOEC are designed.

[0016] 3) This invention effectively overcomes the shortcomings of traditional optimization algorithms in handling strong nonlinear degradation constraints: by utilizing the Twin Critic mechanism and delayed policy update mechanism of the TD3 algorithm, it significantly alleviates the Q-value overestimation problem that is common in the traditional DDPG (Deep Deterministic Policy Gradient) algorithm, making the scheduling strategy more stable and efficient when the agent faces a complex and strongly coupled uncertain environment of electricity, heat and hydrogen. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the interaction architecture between a deep reinforcement learning agent and the multi-energy flow environment of the campus microgrid.

[0018] Figure 2 This is a diagram showing the optimization results of day-to-day power balance for a typical microgrid in the park.

[0019] Figure 3 This is a diagram showing the optimization results of the day-to-day heat power balance of a typical microgrid in the park.

[0020] Figure 4 This is a diagram showing the daily hydrogen energy balance results of a typical microgrid in the park. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0022] This invention studies an off-grid industrial park microgrid system for equipment degradation, comprising a photovoltaic power generation unit, a concentrated solar power generation unit, a power conversion module, an energy storage power station, a dual-tank molten salt thermal storage unit, a hydrogen storage unit, a high-temperature solid oxide electrolysis water unit, an electric heater, a steam generator, a heat exchanger, and a heat recovery unit. The electricity generated by the photovoltaic and concentrated solar power generation units is connected to the power system via the power conversion module to meet electrical load demands and supply power to the energy storage power station, electric heater, and high-temperature solid oxide electrolysis water unit. Specifically, the electric heater converts electrical energy into heat energy, the steam generator produces steam required for system operation, the dual-tank molten salt thermal storage unit stores and releases heat energy, and the heat exchanger and heat recovery unit recover and utilize waste heat from the system to meet heat load demands. The high-temperature solid oxide electrolysis water unit uses electrical and thermal energy to produce hydrogen, which can be transported to the hydrogen storage unit for storage or directly supplied to hydrogen and ammonia loads. Through the connection of the above equipment, the mutual conversion, storage, and coordinated supply of electrical, thermal, and hydrogen energy are achieved.

[0023] The present invention provides an optimized scheduling method for the DRL (Dielectric-Heat-Hydrogen) microgrid of an electrolyzer that considers electrolyzer degradation, comprising the following steps:

[0024] Step 1: Construct physical and mathematical models of the off-grid industrial park microgrid system architecture and core equipment, including the receiver heat collection power model, photovoltaic power generation model, dual-tank molten salt thermal storage system model, power module model, SOEC high-temperature operation polarization and material degradation performance degradation model, energy storage power station model, hydrogen storage tank model, and electric heater model.

[0025] The performance degradation model divides the apparent resistance of SOEC operation into irreversible aging terms and reversible dynamic response terms, and embeds operating temperature limits and temperature change slope limits as constraints.

[0026] (1) Heat collector power model of the receiver:

[0027] The solar field (SF), composed of numerous heliostats, serves as the center for energy capture and conversion. Its function is to concentrate low-energy-density direct normal irradiance (DNI) and convert it into high-grade thermal energy. The solar field mainly consists of heliostats, a solar tracking system, and receivers, forming a physical process of "concentration-heat absorption-heat transfer." The receiver's thermal power model is expressed as follows:

[0028] ;

[0029] ;

[0030] In the formula, This indicates the heat energy entering the absorber. This indicates the usable thermal energy obtained from SF concentrated solar power. This represents the global normal irradiance of the photovoltaic module surface. Indicates the effective heat collection area. Indicates the optical efficiency factor. This represents the efficiency factor of the heat absorber; This represents the heat loss coefficient generated in the absorber, and the subscript t indicates time.

[0031] (2) Photovoltaic power generation model:

[0032] Power generation of a single photovoltaic module Calculated using the following formula:

[0033] ;

[0034] In the formula, The area of ​​a single photovoltaic module ( ); This refers to the number of photovoltaic modules; For inverter efficiency; The depreciation factor is due to the cleanliness of the mirror surface and shadows. For photovoltaic power generation efficiency;

[0035] The efficiency of a photovoltaic module is expressed as:

[0036] ;

[0037] In the formula, Temperature coefficient ( ); and These are the operating temperatures under actual operating conditions and standard conditions, respectively. This is the reference efficiency of the photovoltaic module under standard conditions.

[0038] (3) Model of a dual-tank molten salt thermal storage system:

[0039] Dual-tank direct thermal energy storage technology has become the mainstream configuration for current thermal energy storage systems due to its high efficiency, high reliability, and good commercial maturity. It consists of two independent tanks, a thermal storage medium, a heat exchanger, and corresponding pumps, valves, and control systems. To further characterize the variable operating conditions of solar thermal energy storage, the TES (Thermal Energy Storage, dual-tank molten salt thermal energy storage system) model in this invention considers a steady-state energy flow model with electrical coupling.

[0040] ;

[0041] In the formula, This represents the energy stored in the TES; This indicates the hourly self-heating loss rate of the TES.

[0042] The charge and discharge heat constraints of a dual-tank molten salt thermal storage system can be expressed as:

[0043] ;

[0044] ;

[0045] In the formula, The heating power supplied by the heat source during time period t; and They are respectively The charging and releasing power of the thermal storage tank during a given period; and These are the charging and releasing efficiencies of the thermal storage tank, respectively.

[0046] The dual-tank molten salt thermal storage system cannot simultaneously charge and release heat.

[0047] ;

[0048] In the formula, and They represent The state variables of the thermal storage tank during the time period during heat charging and heat dissipation.

[0049] To ensure the system's scheduling needs for the thermal storage tanks are met in the next scheduling cycle, the heat storage capacity of the thermal storage system at the end of the scheduling cycle will be allocated accordingly. Restore heat storage to the initial period :

[0050] ;

[0051] Thermal storage tanks also have a minimum capacity limit, which is expressed as follows:

[0052] ;

[0053] In the formula, This indicates the minimum heat storage capacity of the thermal storage tank; This indicates the maximum heat storage capacity of the thermal storage tank.

[0054] The mass balance constraint of a dual-tank molten salt thermal storage system is expressed as:

[0055] ;

[0056] ;

[0057] In the formula, the subscripts HT and CT represent hot tank and cold tank, respectively; the subscripts in and out represent molten salt flowing into and out of the thermal storage tank, respectively. This refers to the amount of molten salt in the hot tank; The amount of molten salt flowing into the hot tank; The amount of molten salt flowing out of the hot tank; This refers to the amount of molten salt in the cold tank; The amount of molten salt flowing into the cold tank; This represents the amount of molten salt flowing out of the cold tank; the subscript t indicates the time period.

[0058] (4) Power module model:

[0059] The power generation stage uses steam turbines for power generation. Similar to the operating constraints of conventional thermal power units, this invention uses the electro-thermal energy conversion efficiency, which is easy to embed into the optimization model, as the input parameter. When performing system scheduling optimization, it considers operating constraints such as output limit constraints, unit ramp power limit constraints, and unit start-up and shutdown constraints.

[0060] ;

[0061] ;

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] ;

[0068] In the formula, This represents the power of PB (Power Block) during time period t; This represents the thermal power required for PB to generate electricity; This indicates the conversion efficiency of PB; Represents the operating state variables of a solar thermal power plant; and These represent the minimum and maximum active power output of the power module, respectively. and These represent the upward and downward climbing capabilities of a solar thermal power plant, respectively. This represents the startup status variable of the power module during time period t. This represents the shutdown status variable of the power module during time period t; This indicates the minimum continuous running time of PB; Indicates the minimum continuous downtime of PB; This represents the total number of segments included in the current optimized scheduling cycle; This represents the traversal index of the scheduling period.

[0069] (5) SOEC model considering dynamic response and cumulative degradation under operating conditions (i.e., performance degradation model caused by high-temperature operation polarization and material deterioration in SOEC):

[0070] Under high temperature and high load operating conditions, the performance of SOEC electrodes and electrolytes degrades with the accumulation of temperature, current density, and operating time. Its internal resistance exhibits complex dynamic characteristics that change with operating time, current density, and operating temperature. To accurately characterize the impedance evolution of SOEC, this invention divides the apparent resistance into irreversible aging terms and reversible dynamic response terms, embedding operating temperature limits and temperature change slope limits as constraints. While ensuring the physical feasibility and operational safety of the model, it simulates the performance degradation caused by electrothermal stress during long-term operation and changes in operating conditions, thereby characterizing the dynamic evolution of the aging state from the initial operating stage to the long-term stable stage.

[0071] This invention selects the degradation mechanism and simulation approach under SOEC mode. In reversible operation, it is necessary to consider the different contributions of dynamic behavior of operating condition response and cumulative degradation effects, and construct a performance degradation model caused by polarization and material deterioration during high-temperature operation of SOEC.

[0072] ;

[0073] ;

[0074] ;

[0075] ;

[0076] ;

[0077] In the formula, This is the open-circuit voltage of SOEC; The apparent resistance at time t; This is the irreversible aging term of the resistance, used to characterize the cumulative damage to materials and interfaces, exhibiting a monotonic and unchanging characteristic. It is a reversible operating term used to characterize the instantaneous correction of impedance by changes in temperature and current density, allowing for fluctuations in impedance. The initial area ratio resistivity (mΩ•cm) 2 ), Operating temperature (°C) Current density (A / cm) 2 ), Resistivity degradation rate (mΩ•cm) 2 / kh); This refers to the number of operating hours. The number of operating hours within the scheduling period; The temperature sensitivity coefficient is taken as -5 × 10⁻⁵. -4 Ω•cm 2 / K, The current sensitivity coefficient is taken as 0.05 Ω•cm. 2 / (A / cm 2 ); The irreversible resistance of the s-th SOEC electrolyzer increases due to cumulative aging during time period t; This is the reference current density for the SOEC electrolyzer; The voltage of the tSOEC electrolyzer during the specified time period is 1.

[0078] (6) Energy storage power station model:

[0079] Energy storage power stations are an important component of microgrids, and can alleviate the fluctuations in renewable energy generation and the uncertainty of load in microgrids to a certain extent. In this invention, a storage model is established using a battery as an example. The state of charge (SOC) of the battery in time period t is related to the SOC and charging / discharging power of the previous time period, and its dynamic changes can be described as follows:

[0080] ;

[0081] In the formula, This indicates the state of charge of the battery during time period t. and These represent the charging power and discharging power of the battery during that period, respectively. This refers to the rated capacity of the battery. This represents the length of the time period.

[0082] The battery's energy storage state and charge / discharge power satisfy the state of charge and charge / discharge power constraints:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] In the formula, and These represent the maximum and minimum charge coefficients of the battery, respectively. and These are the maximum charge rate and discharge rate of the battery, respectively. and A Boolean variable, representing Charging and discharging cannot occur simultaneously; if so, set to 1, otherwise set to 0.

[0088] (7) Hydrogen storage tank model:

[0089] Hydrogen storage tanks are used to safely and stably store high-pressure or cryogenic liquid hydrogen under various operating conditions. Their internal pressure, temperature, and hydrogen density dynamically change during the charging and discharging process, significantly impacting storage efficiency and system safety. The hydrogen storage capacity and the charging / discharging capacity can be expressed as:

[0090] ;

[0091] ;

[0092] ;

[0093] ;

[0094] ;

[0095] ;

[0096] In the formula, For hydrogen storage tanks in Hydrogen storage capacity over a given period of time; and These represent the hydrogen filling and discharging efficiencies of the hydrogen storage tank, respectively. This represents the actual amount of hydrogen charged. This represents the actual amount of hydrogen released. and They represent The state variables of the hydrogen storage tank during the time period for charging and releasing heat; This indicates the amount of hydrogen added to the hydrogen storage tank during time period t. This indicates the amount of hydrogen released from the hydrogen storage tank during time period t; This indicates the hydrogen filling efficiency of the hydrogen storage tank; This indicates the hydrogen release efficiency of the hydrogen storage tank; This indicates the amount of hydrogen stored in the hydrogen storage tank during time period t; This indicates the minimum allowable hydrogen storage capacity of the hydrogen storage tank during time period t. This indicates the maximum allowable hydrogen storage capacity of the hydrogen storage tank during time period t. This indicates the amount of hydrogen stored in the hydrogen storage tank at the beginning of the scheduling cycle; This indicates the amount of hydrogen stored in the hydrogen storage tank at the end of the scheduling cycle.

[0097] (8) Electric heater model:

[0098] The main function of an electric heater is to convert electrical energy into heat energy. According to the law of conservation of energy, given the heat generated by the electric heater... And the efficiency of the electric heater In this case, the power of the electric heater can be calculated. :

[0099] ;

[0100] Power balance: In off-grid industrial microgrids, electricity is mainly supplied by photovoltaic power generation, solar thermal power module generation, and energy storage discharge. Simultaneously, it needs to meet the electrical load demand and the energy consumption of various equipment operations. The system must satisfy the following power balance constraints:

[0101] ;

[0102] In the formula, This represents the power generation of the photovoltaic system during time period t; This represents the discharge power of the energy storage battery during time period t. This represents the power generation capacity of the power module during time period t; This indicates the power demand of the park's electrical load during time period t; This represents the electrical power consumed during the hydrogen production process in time period t. This represents the charging power of the energy storage battery during time period t. This represents the electrical power consumed by the electric heater during time period t.

[0103] Thermal energy balance: The thermal energy mainly comes from the heat collection field, the heat released by the thermal storage system, and a small amount of heat generated by the electric heaters. At the same time, it needs to meet the thermal energy requirements of the SOEC electrolysis process and the industrial heat load of the park. The system thermal energy needs to meet the following balance relationship:

[0104] ;

[0105] In the formula, Indicates the location of the heat collection point during the time period t; and These represent the charging and releasing power of the thermal storage tank during time period t, respectively. This represents the thermal energy required to drive the steam turbine to generate electricity in PB. This represents the heat power generated by the electric heater during time period t; The supplementary heat power of the auxiliary heat source for time period t; This represents the thermal power required for hydrogen production in a solid oxide electrolyzer during time period t. This represents the heat load demand during the production process in the industrial park during time period t.

[0106] Step 2: Model the multi-energy flow coordination and scheduling problem of off-grid microgrids as a Markov Decision Process (MDP) involving multi-stage continuous decision-making in time. The deep reinforcement learning agent interacts with the microgrid multi-energy flow simulation environment by constructing the following state space, action space, and core reward function: ,in, Let be the possible state space of each electrical device in the light-heat-electricity-hydrogen process. The operating space includes the electrolyzer hydrogen production power (SOEC), the thermal storage charging / discharging power, the battery charging / discharging power, the electric heater (EH), the hydrogen charging / discharging rate of the hydrogen storage tank, the electrolyzer operating temperature, and the power of the Rankine cycle power module (PB). The state transition probability describes the state transition probability. Take a certain action Then, the probability distribution of the environment transitioning from one state to another. For immediate rewards, representing the state Take a certain action The reward that the intelligent agent receives afterwards. It is a discount factor, indicating the degree of importance attached to future rewards.

[0107] (1) State space S;

[0108] Each time period State variables of the internal light-electricity-thermal-hydrogen microgrid system It can be represented as:

[0109] ;

[0110] In the formula, This refers to the photovoltaic power generation capacity predicted recently. The high-temperature thermal power required for hydrogen production in an electrolyzer; This represents the current output thermal power of the backup electric heater. This represents the current capacity of the dual molten salt thermal storage system. This represents the current real-time hydrogen production rate of the solid oxide electrolyzer (SOEC). This refers to the state of charge of the battery. , as well as These are the current electricity load demand, hydrogen load demand, and industrial heat load demand of the park, respectively. The actual heat absorbed by the heat collection field from the outside environment; This represents the real-time electrolytic current during the operation of the electrolytic cell. This refers to the apparent resistance during the operation of the electrolytic cell.

[0111] (2) Action space A;

[0112] The action space of the intelligent agent consists of control decision variables that directly affect the multi-energy flow interaction regulation of the microgrid system. A seven-dimensional continuous action vector is designed with the following mathematical configuration:

[0113] ;

[0114] In the formula, The electrolyzer consumes electrical power to produce hydrogen. This represents the heat charging / heat dissipation power of the thermal storage system (positive value for heat charging, negative value for heat dissipation). The charging / discharging power of the battery; The electrical power consumed by the backup electric heater; The rate control values ​​for charging and discharging hydrogen from the hydrogen storage tank; This refers to the real-time physical operating temperature of the electrolytic cell; This refers to the output electrical power of the Ken cycle power module.

[0115] (3) Instant rewards ;

[0116] The reward is the benefit an agent gains from its current action based on operational goals. For the optimized scheduling in the photovoltaic-electricity-thermal-hydrogen microgrid system of this invention, the reward is a combination of operating costs and constraint violation penalties. The reward is calculated for each time period t. for:

[0117] ;

[0118] In the formula, This represents the overall operating cost of the system. This indicates a penalty for violating the rules.

[0119] (4) State transition probability ;

[0120] ;

[0121] In the formula, For time period The state space, For time period The action space.

[0122] Step 3: The standard TD3 algorithm is used to interactively train the agent. Taking into account the operation and maintenance costs of photovoltaic and solar thermal power, the life reset costs of batteries and electrolyzers, the penalty for curtailment of solar and solar power, and the load shedding penalty triggered by the imbalance of supply and demand of multiple energy flows, the objective function and constraints are determined with the lowest system cost. The day-ahead optimal scheduling strategy is obtained by alternately updating the parameters of the Actor network and the Critic network and applied online.

[0123] The TD3 algorithm structure is as follows: Figure 1 As shown, it consists of one Actor network, two Critic networks, a corresponding target network, and an experience pool. TD3 reduces Q-value overestimation and improves the stability of continuous action scheduling by using "the smaller value in the dual-value network, target policy smoothing, and delayed policy network updates." Here, t represents a time period; t+h represents the h-th time period starting from t. This indicates the environmental state at that time step; This indicates the state after h time steps following the execution of the action; Indicates that the intelligent agent is in The following action; This represents the target action generated by the target policy network after h time steps. Indicate execution The instant reward obtained afterward; Indicates the current policy network; Represents the target policy network; This indicates the action output by the current policy network; This represents the action generated by the target policy network; Indicates the current state of the policy network. The action to be output; These represent the parameters of the current policy network; These represent the parameters of the target policy network; and This represents two independent action value functions; This represents an estimated value of the cumulative discount reward after an action is performed starting from this state. and These represent the parameters of the current value network 1 and value network 2; and The parameters represent target value network 1 and target value network 2; This represents random noise added to the target's action; Loss function of value network; This represents the optimization objective of the policy network; It is a conversion factor; This represents the gradient operator; Represents the target network soft update coefficient; Indicates the length of the empirical sequence; and These represent the Q-values ​​calculated by value network 1 based on the current state and actions, and the Q-values ​​calculated by value network 2 based on the current state and actions, respectively.

[0124] Considering the combined operating and maintenance costs of photovoltaic (PV), solar thermal (CSP), high-temperature solid oxide electrolysis water production devices, energy storage power stations, and hydrogen storage tanks, as well as the costs of curtailed solar and solar power, and aiming to minimize the overall system cost, a system scheduling optimization cost model is established. The objective function is:

[0125] ;

[0126] In the formula, For system operating costs; Costs associated with the system's waste of light and heat; Penalty for system load abandonment.

[0127] (1) System operating costs: Operating costs include photovoltaic power generation costs Cost of concentrated solar power High-temperature solid oxide water electrolysis hydrogen production operating cost Energy storage costs Hydrogen storage cost Net revenue from electricity and hydrogen sales by the system .

[0128] ;

[0129] ;

[0130] ;

[0131] ;

[0132] ;

[0133] ;

[0134] ;

[0135] In the formula, The photovoltaic power generation cost coefficient is 0.0331 yuan / kW; The cost coefficient for concentrated solar power (CSP) is 0.01 yuan / kW. The cost coefficient for hydrogen production via water electrolysis is 0.01 yuan / kW; The hydrogen storage cost coefficient for the hydrogen buffer tank is 0.01 yuan / kW; The energy storage cost coefficient is 0.2 yuan / kW; and These represent the revenue from electricity sales and hydrogen sales, respectively. and The system's electricity and hydrogen sales prices; and Let be the system's power output and hydrogen output during time period t; This represents the operating volume corresponding to the SOEC hydrogen production process during time period t; This indicates the amount of hydrogen stored in the hydrogen storage tank during time period t; This represents the discharge power of the energy storage system during time period t; This represents the charging power of the energy storage system during time period t.

[0136] (2) Costs of curtailing heat and light:

[0137] The waste of heat and light energy is transformed into part of the economic cost through a penalty factor.

[0138] ;

[0139] ;

[0140] ;

[0141] In the formula, and These are the costs of wasted solar power and wasted heat, respectively. , Each is the original power output of the wind and solar power system; and The power outputs are respectively provided by the photovoltaic and solar thermal power units of the system; and The penalty factors for wasted light and wasted heat are 2.0, respectively.

[0142] Example Selection: The electricity demand of industrial parks is characterized by its year-round continuity. Production load typically accounts for about 75% of the total load demand, and the corresponding load curves are mostly unimodal, with daytime load rates generally higher than nighttime load rates. However, the fluctuation range within each phase is relatively limited, exhibiting a segmented characteristic of "high production during the day and low production at night." Based on these characteristics, this invention uses a microgrid in an industrial park in Northwest China as the scheduling object. To simulate the new energy access scenario in areas rich in solar energy resources, an off-grid solar-thermal-electricity-hydrogen coupling system was established. This station includes a 6MW solar thermal generator (CSP), a 4MW high-temperature water electrolysis hydrogen production unit (SOEC), and a 10MW photovoltaic array (PV), supported by a 150MWh thermal energy storage system (TES) and a 1MW electric heater (EH).

[0143] Scheduling result analysis:

[0144] (1) Power supply analysis: Figure 2 The power balance result of this invention is that in an off-grid independent system without external grid support, the TD3 agent must maintain a rigid balance of internal power at all times. The power scheduling of the system in a typical 24-hour day exhibits a strong "day-night alternation, source-load coordination" characteristic, and the specific sub-period analysis is as follows.

[0145] Nighttime off-peak period (00:00–07:00): During this period, there is no sunlight, and the 10MW photovoltaic array output is 0. The park's electrical load is mainly supported by the 6MW solar thermal generator unit, supplemented by the stable discharge of energy storage batteries. Due to the relatively stable electrical load at night, in order to mitigate the resistance degradation caused by frequent start-ups and shutdowns of the electrolyzers or extreme temperature fluctuations, the intelligent system controls the high-temperature SOEC hydrogen production unit in a "hot standby" state with extremely low power. During this period, there is no excess electrical energy, and the electric heater remains off (output is 0).

[0146] During peak daytime hours (08:00–16:00): As solar irradiance rapidly increases in Northwest China, the 10MW photovoltaic (PV) system reaches its peak output. This large-scale PV output not only fully covers the park's basic electrical load but also generates significant surplus electricity. The TD3 intelligent agent precisely guides the 4MW high-temperature electrolytic water generator (SOEC) to gradually reach full capacity, pushing its power output to a high level (approaching its 4MW rated limit), converting surplus green electricity into hydrogen energy. During the peak PV output period at midday (11:00–14:00), even with the SOEC operating at full capacity, some surplus electricity remains. At this time, the 1MW electric heater activates, converting this surplus electricity into high-grade heat energy injected into the thermal storage system, achieving "zero waste" of day-ahead solar power. During this period, the intelligent agent decisively reduces or even shuts down the electrical output of the solar thermal generator, transferring all the heat captured by the collector field and the heat generated by the EH (Electric Heater) into the thermal storage tank for storage, preparing for nighttime power generation.

[0147] During the peak load period from evening to night (17:00-24:00): After 17:00, sunlight begins to disappear, and photovoltaic output drops sharply and reaches zero, while the industrial park is still in its peak electricity and heat consumption period. Faced with a huge instantaneous power shortage, the intelligent system seamlessly switches to "thermal energy storage and power generation" mode. The huge amount of heat energy stored in the TES during the day is released, driving the 6MW solar thermal power unit to quickly switch to high-power generation, operating at full capacity to stably support the electricity load until late at night, fully demonstrating the backup advantages of solar thermal power generation: "built-in energy storage and stable and controllable output".

[0148] (2) Analysis of heating supply: Figure 3This is the core manifestation of the "SOEC degradation" of this invention. Thermal energy is the key link connecting the heat collection field, the thermal storage system, the solar thermal power generation (Rankine cycle), and the high-temperature SOEC electrolyzer. The system exhibits complex "heat-electricity-hydrogen" strong coupling characteristics at high temperatures (typically 750℃-950℃). For SOEC, its core advantage lies in "using heat instead of electricity," that is, by absorbing external high-temperature thermal energy, the active electrical input required for the water electrolysis reaction can be effectively reduced.

[0149] Figure 3 This demonstrates the thermal energy capture of the collector field and the time shift of the dual molten salt tanks. During the day (08:00–16:00), the collector field begins to heat up dramatically under abundant normal direct irradiation from the northwest, resulting in a rise in the thermal power curve. Simultaneously, the hydrogen production capacity of the SOEC increases significantly during the day, with its required high-temperature thermal power reaching its peak. When the SOEC operates at insufficient temperatures, its internal ion conductivity irreversibly decreases, leading to a rapid increase in polarization and ohmic losses, which in turn triggers a surge in irreversible aging terms and a significant increase in apparent internal resistance. The intelligent agent, through multi-energy flow coordination, directly directs some of the thermal energy from the collector field and the system to the SOEC. This maximizes its high energy conversion efficiency ("heat-for-electricity") while ensuring the isotropic thermal balance of the high-temperature electrolyzer, fundamentally preventing the increase in apparent internal resistance and accelerated material degradation caused by insufficient or drastic temperature fluctuations. Apart from the heat energy consumed by SOEC, most of the excess heat energy captured by the collector field is continuously injected into the 150MWh thermal storage system for storage using molten salt as a medium, resulting in a consistently high TES charging power.

[0150] During the night (17:00-7:00 the next day), there is no solar radiation, the output of the thermal collector is reduced to zero, and the electric heaters are shut down. The only available heat source for the system becomes the thermal storage system. The thermal storage system switches to high-power heat release mode. The released heat energy is used as the main input and directly injected into the power module to drive the steam turbine for Rankine cycle power generation. At night, the SOEC is in low-power thermal standby, and its thermal power demand is extremely low. However, the system still needs to allocate a small amount of heat energy from the TES to keep it warm, maintaining its temperature within a reasonable range to suppress internal resistance degradation, thereby avoiding severe thermal stress damage and irreversible aging degradation caused by frequent cold starts.

[0151] (3) Analysis of hydrogen supply: The hydrogen energy dispatch results reflect how the off-grid system uses hydrogen storage tanks to break the constraints of real-time supply and demand and achieve a long-term optimization process of stable all-weather external supply of green hydrogen.

[0152] In this embodiment, the industrial hydrogen load of the park exhibits a continuous, stable, rigid, and constant demand curve. Since the off-grid system cannot purchase hydrogen externally, the total daily hydrogen production must be greater than or equal to the total hydrogen load. Figure 4During the daytime intensive hydrogen charging phase (8:00-16:00), excess photovoltaic power is plentiful, and the SOEC water electrolysis hydrogen production unit operates at high power and high efficiency, with a hydrogen production rate far exceeding the park's constant basic hydrogen load demand. The excess green hydrogen is then pressurized into the hydrogen storage tank, which then switches to charging mode. The corresponding state of capacity (SOC / pressure) curve of the hydrogen storage tank shows a near-linear, steep upward trend during this period, completing a large-scale, efficient conversion of electrical energy into hydrogen energy and its time-shifted storage.

[0153] During the nighttime hydrogen consumption and peak shaving phase (17:00-7:00 the next day), due to the lack of low-cost photovoltaic green electricity at night and the priority given to supplying the park's electricity load with electricity generated by solar thermal power generation, the intelligent agent, considering the economics of the entire life cycle, reduced the hydrogen production power of SOEC. Since the hydrogen release rate remained constant throughout the night, the hydrogen storage tanks did not undergo any unnecessary back-and-forth charging and discharging operations. Its state-of-the-art capacity (SOC / pressure) curve showed an extremely smooth, uniform, and monotonically decreasing trend starting at 17:00, seamlessly filling the hydrogen supply-demand gap that lasted for more than ten hours at night, until it dropped to a baseline low level the following morning. This low-frequency, long-cycle steady-state scheduling mode highlights the current self-sufficiency in hydrogen production.

[0154] Figures 2-4 The invention clearly demonstrates that the TD3 agent successfully learned the optimal scheduling mechanism in a strongly nonlinear, non-convex constraint environment. In decision-making, the agent did not blindly pursue short-term hydrogen production maximization or frequently adjust battery charging and discharging. Instead, it automatically adjusted the thermal stress of the SOEC electrolyzer during day-ahead scheduling and avoided deep discharge of the energy storage batteries at low SOC (State of Charge), sacrificing a minimal amount of day-ahead operational efficiency in exchange for a significant extension of the microgrid's core equipment lifespan. In extreme off-grid scenarios without external power grid or heating network support, the TD3 agent, through interaction with the environment alone, achieved perfect multi-dimensional, cross-time-period, and cross-energy-substance decoupling and coordination among 10MW photovoltaic, 6MW solar thermal, 4MW solid oxide electrolysis hydrogen production, 150MWh thermal storage, and electric heaters. This ensured a safe and stable supply and demand balance for the park's day-ahead electricity, heat, and hydrogen, fully validating the high efficiency of this method in the intelligent scheduling application of microgrids.

Claims

1. A DRL optimization scheduling method for a photoelectric-thermal-hydrogen microgrid considering electrolyzer degradation, characterized in that, Includes the following steps: Step 1: Construct a physical and mathematical model of the off-grid industrial park microgrid system architecture and core equipment. The off-grid industrial park microgrid has no energy or information interaction with the external power grid and heating network. The physical and mathematical model deeply embeds a performance degradation model caused by SOEC high-temperature operation polarization and material deterioration. The performance degradation model divides the apparent resistance of SOEC operation into irreversible aging terms and reversible dynamic response terms, and embeds the operating temperature limit and temperature change slope limit as constraints. Step 2: Establish the Markov decision process of the off-grid industrial park microgrid, including the state space, action space, and reward function; The state space includes at least: photovoltaic power generation, SOEC hydrogen production thermal power, electric heater thermal power, thermal storage capacity, SOEC hydrogen production rate, battery energy storage capacity, electrical load demand, hydrogen load demand, thermal load demand, heat obtained by the heat collection field, SOEC real-time operating current, and SOEC operating apparent resistance. The operating space includes at least: SOEC hydrogen production power, thermal storage charging and discharging power, battery charging and discharging power, electric heater power, hydrogen storage tank charging and discharging rate, SOEC operating temperature, and power module output power. The reward function is constructed based on the overall system operating cost and the penalty for violating constraints. Step 3: The agent is interactively trained using the standard TD3 algorithm. The objective function and constraints are determined with the lowest overall system operating cost. The optimal scheduling strategy is obtained by alternately updating the parameters of the Actor network and the Critic network and then applied online.

2. The photoelectric-thermal-hydrogen microgrid DRL optimization scheduling method considering electrolyzer degradation according to claim 1, characterized in that, The performance degradation model mentioned in step 1 is as follows: ； ； ； ； ； In the formula, This is the open-circuit voltage of SOEC; The apparent resistance during time period t; This is the irreversible aging term of the resistance, used to characterize the cumulative damage to materials and interfaces, exhibiting a monotonic and unchanging characteristic. It is a reversible operating term used to characterize the instantaneous correction of impedance by temperature and current density, allowing for fluctuations in impedance. The initial area ratio resistance. Operating temperature; Current density; The resistance degradation rate; This refers to the number of operating hours. The number of operating hours within the scheduling period; For temperature sensitivity coefficient, The current sensitivity coefficient; The irreversible resistance of the s-th SOEC electrolyzer increases due to cumulative aging during time period t; This is the reference current density for the SOEC electrolyzer; The operating voltage of the SOEC electrolyzer during time period t; This is a reference temperature.

3. The photoelectric-thermal-hydrogen microgrid DRL optimization scheduling method considering electrolyzer degradation according to claim 1, characterized in that, In step 1, the physical mathematical model also includes a heat collector power model for the heat absorber, expressed as: ； ； In the formula, This indicates the heat energy entering the absorber. This indicates the usable thermal energy obtained from SF concentrated solar power. This represents the global normal irradiance of the photovoltaic module surface. Indicates the effective heat collection area. Indicates the optical efficiency factor. This represents the efficiency factor of the heat absorber; This represents the heat loss coefficient generated in the absorber, and the subscript t indicates the time period.

4. The photoelectric-thermal-hydrogen microgrid DRL optimization scheduling method considering electrolyzer degradation according to claim 1, characterized in that, In step 1, the physical mathematical model also includes a photovoltaic power generation model; Power generation of a single photovoltaic module Calculated using the following formula: ； In the formula, The area of ​​a single photovoltaic module; This refers to the number of photovoltaic modules; The global normal irradiance of the photovoltaic module surface; For inverter efficiency; The depreciation factor is due to the cleanliness of the mirror surface and shadows. For photovoltaic power generation efficiency; The efficiency of a photovoltaic module is expressed as: ； In the formula, Temperature coefficient; and These are the operating temperatures under actual operating conditions and standard conditions, respectively. This is the reference efficiency of the photovoltaic module under standard conditions.

5. The photoelectric-thermal-hydrogen microgrid DRL optimization scheduling method considering electrolyzer degradation according to claim 1, characterized in that, In step 1, the physical mathematical model also includes a dual-tank molten salt thermal storage system model. This dual-tank molten salt thermal storage system model adopts an electrically coupled steady-state energy flow model, expressed as: ； In the formula, This represents the energy stored in the thermal storage tank; This indicates the hourly self-heating loss rate of the thermal storage tank; and They are respectively The charging and releasing power of the thermal storage tank during a given period; The charge and discharge heat constraints of the dual-tank molten salt thermal storage system are expressed as follows: ； ； In the formula, The heating power supplied by the heat source during time period t; and These are the heat storage tank's charging and releasing efficiencies, respectively. The constraint that the dual-tank molten salt thermal storage system cannot simultaneously charge and release heat is expressed as follows: ； In the formula, and They represent The state variables of the thermal storage tank during the time period during which it is filled and released heat; To ensure the system's scheduling needs for the thermal storage tanks are met in the next scheduling cycle, the heat storage capacity of the thermal storage system at the end of the scheduling cycle will be allocated accordingly. Restore heat storage to the initial period , represented as: ； Thermal storage tanks have a minimum capacity limit, which is expressed as: ； In the formula, This indicates the minimum heat storage capacity of the thermal storage tank; This indicates the maximum heat storage capacity of the thermal storage tank; The mass balance constraint of a dual-tank molten salt thermal storage system is expressed as: ； ； In the formula, the subscripts HT and CT represent hot tank and cold tank, respectively; the subscripts in and out represent molten salt flowing into and out of the thermal storage tank, respectively. This refers to the amount of molten salt in the hot tank; The amount of molten salt flowing into the hot tank; The amount of molten salt flowing out of the hot tank; This refers to the amount of molten salt in the cold tank; The amount of molten salt flowing into the cold tank; This represents the amount of molten salt flowing out of the cold tank; the subscript t indicates the time period.

6. The photoelectric-thermal-hydrogen microgrid DRL optimization scheduling method considering electrolyzer degradation according to claim 1, characterized in that, In step 1, the physical mathematical model also includes a power module model, represented as follows: ； ； ； ； ； ； ； ； In the formula, This represents the power of PB during time period t; This represents the thermal power required for PB to generate electricity; This indicates the conversion efficiency of PB; The variable representing the operating state of the solar thermal power plant during time period t; and These represent the minimum and maximum active power output of the power module, respectively. and These represent the upward and downward climbing capabilities of a solar thermal power plant, respectively. This represents the startup status variable of the power module during time period t. This represents the shutdown status variable of the power module during time period t; This indicates the minimum continuous running time of PB; Indicates the minimum continuous downtime of PB; This represents the total number of segments included in the current optimized scheduling cycle; Indicates the time period of the solar thermal power plant The running state variables; This represents the traversal index of the scheduling period.

7. The photoelectric-thermal-hydrogen microgrid DRL optimization scheduling method considering electrolyzer degradation according to claim 1, characterized in that, In step 1, the physical mathematical model also includes an energy storage power station model and a hydrogen storage tank model; In the energy storage power station model, the dynamic changes of the battery are described as follows: ； In the formula, This indicates the state of charge of the battery during time period t. and These represent the charging power and discharging power of the battery during time period t, respectively. This refers to the rated capacity of the battery. The duration of the time period; The battery's state of charge and charge / discharge power satisfy the constraints of state of charge and charge / discharge power: ； ； ； ； In the formula, and These represent the maximum and minimum charge coefficients of the battery, respectively. and These are the maximum charging rate and discharge rate of the battery, respectively. and A Boolean variable, representing Charging and discharging cannot occur simultaneously during a given period; if so, set the value to 1, otherwise set it to 0. In the hydrogen storage tank model, the hydrogen storage capacity and the hydrogen filling / discharging capacity are expressed as follows: ； ； ； ； ； ； In the formula, For hydrogen storage tanks in Hydrogen storage capacity over a given period of time; and These represent the hydrogen filling and discharging efficiencies of the hydrogen storage tank, respectively. This represents the actual amount of hydrogen charged. This represents the actual amount of hydrogen released. and They represent The state variables of the hydrogen storage tank during the time period for charging and releasing heat; This indicates the amount of hydrogen added to the hydrogen storage tank during time period t. This indicates the amount of hydrogen released from the hydrogen storage tank during time period t; This indicates the hydrogen filling efficiency of the hydrogen storage tank; This indicates the hydrogen release efficiency of the hydrogen storage tank; This indicates the amount of hydrogen stored in the hydrogen storage tank during time period t; This indicates the minimum allowable hydrogen storage capacity of the hydrogen storage tank during time period t. This indicates the maximum allowable hydrogen storage capacity of the hydrogen storage tank during time period t. This indicates the amount of hydrogen stored in the hydrogen storage tank at the beginning of the scheduling cycle; This indicates the amount of hydrogen stored in the hydrogen storage tank at the end of the scheduling cycle.

8. The photoelectric-thermal-hydrogen microgrid DRL optimization scheduling method considering electrolyzer degradation according to claim 1, characterized in that, In step 1, the physical mathematical model also includes an electric heater model; specifically as follows: According to the law of conservation of energy, given the heat generated by the electric heater... And the efficiency of the electric heater Calculate the power of the electric heater under the given conditions. : ； Satisfy the power balance constraint: ； In the formula, This represents the power generation of the photovoltaic system during time period t; This represents the discharge power of the energy storage battery during time period t. This represents the power generation capacity of the power module during time period t; This indicates the power demand of the park's electrical load during time period t; This represents the electrical power consumed during the hydrogen production process in time period t. This represents the charging power of the energy storage battery during time period t. This represents the electrical power consumed by the electric heater during time period t. Satisfying thermal energy balance constraints: ； In the formula, Indicates the location of the heat collection point during the time period t; and These represent the charging and releasing power of the thermal storage tank during time period t, respectively. This represents the heat power generated by the electric heater during time period t; The supplementary heat power of the auxiliary heat source for time period t; This represents the thermal power required for hydrogen production in a solid oxide electrolyzer during time period t. This represents the heat load demand during the production process in the industrial park at time period t. This represents the thermal energy required to drive the steam turbine to generate electricity in the power module.

9. The photoelectric-thermal-hydrogen microgrid DRL optimization scheduling method considering electrolyzer degradation according to claim 1, characterized in that, In step 2, the Markov decision process for the off-grid industrial park microgrid is established, as follows: Deep reinforcement learning agents construct the following state space Action space The core reward function interacts with the microgrid multi-energy flow simulation environment, as shown below: ; Among them, the state space This represents the possible state space and action space of each power component in the photovoltaic-electrical-thermal-hydrogen microgrid system. This includes the electrolyzer's hydrogen production power, the thermal storage charging / discharging power, the battery charging / discharging power, the electric heater's power, the hydrogen storage tank's charging / discharging rate, the electrolyzer's operating temperature, and the power of the power module PB. Let be the state transition probability, describing the state transition probability. Take a certain action Then, the probability distribution of the environment transitioning from one state to another; For immediate rewards, representing the state Take a certain action The reward that the agent receives afterwards; It is a discount factor, indicating the degree of importance attached to future rewards; (1) State space ; Each time period State variables of the internal light-electricity-thermal-hydrogen microgrid system Represented as: ； in, This refers to the photovoltaic power generation capacity predicted recently. The high-temperature thermal power required for hydrogen production in an electrolyzer; This represents the current output thermal power of the backup electric heater. This represents the current capacity of the dual-tank molten salt thermal storage system. This represents the current real-time hydrogen production rate of SOEC. This represents the current remaining percentage of battery charge. , as well as These are the current electricity load demand, hydrogen load demand, and industrial heat load demand of the park, respectively. The actual heat absorbed by the heat collection field from the outside environment; and These are the real-time electrolytic current magnitude and the apparent operating resistance of the electrolytic cell during operation, respectively. (2) Action space A; The action space of the intelligent agent consists of control decision variables that directly affect the multi-energy flow interaction regulation of the microgrid system. A seven-dimensional continuous action vector is designed with the following mathematical configuration: ； in, Decision value for electrical power consumption in hydrogen production by electrolyzer; The heat charging / heat dissipation power of the thermal storage system; The charging / discharging power of the battery; The electrical power consumed by the backup electric heater; The rate control values ​​for charging and discharging hydrogen from the hydrogen storage tank; This refers to the real-time physical operating temperature of the electrolytic cell; Decision on the output power of the Rankine cycle power module; (3) Instant rewards ; The reward for each time period t is : ； in, This represents the overall operating cost of the system. This indicates the penalty for violating the constraint; (4) State transition probability ; ； in, for The state space of a time period For time period The action space.

10. The photoelectric-thermal-hydrogen microgrid DRL optimization scheduling method considering electrolyzer degradation according to claim 1, characterized in that, In step 3, when using the standard TD3 algorithm to interactively train the agent, a system scheduling optimization cost model is established by comprehensively considering the operating and maintenance costs of photovoltaic power generation, solar thermal power generation, high-temperature solid oxide water electrolysis hydrogen production devices, energy storage power stations, and hydrogen storage tanks, as well as the costs of curtailed solar power and curtailed thermal power. The objective function is: ； In the formula, For system operating costs; Costs associated with the system's waste of light and heat; Penalty for system load abandonment; System operating costs Including the cost of photovoltaic power generation Cost of solar thermal power generation High-temperature solid oxide water electrolysis hydrogen production operating costs Energy storage costs Hydrogen storage cost and the net revenue from the sale of electricity and hydrogen by the system ; ； ； ； ； ； ； ； in, This represents the cost coefficient for photovoltaic power generation. This represents the cost coefficient for concentrated solar power (CSP). This is the cost coefficient for hydrogen production via water electrolysis; The hydrogen storage cost coefficient for hydrogen buffer tanks; This is the energy storage cost coefficient; and These represent the revenue from electricity sales and hydrogen sales, respectively. and The system's electricity and hydrogen sales prices; and The system's power output and hydrogen output during time period t; This represents the operating volume corresponding to the SOEC hydrogen production process during time period t; This indicates the amount of hydrogen stored in the hydrogen storage tank during time period t; This represents the discharge power of the energy storage system during time period t; This represents the charging power of the energy storage system during time period t; Costs of waste heat and light As a part of the economic cost through the penalty factor, it is expressed as: ； ； ； in, and These are the costs of wasted solar power and wasted heat, respectively. , Each is the original power output of the wind and solar power system; and The power outputs are respectively for the photovoltaic and solar thermal power units of the system; and These are the penalty factors for wasted light and wasted heat, respectively.

Images