Photovoltaic-solid oxide fuel cell hybrid energy system control method, system and equipment

By applying TD3 algorithm and deep neural network in fuel cell system, an intelligent control strategy was constructed, which solved the limitations of fuel cell control strategy in microgrids, achieved more efficient and stable power management, and improved the microgrid's Performance and reliability.

CN118659447BActive Publication Date: 2024-12-03STATE GRID TIANJIN ELECTRIC POWER COMPANY +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411134456.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-12-03
Estimated Expiration
2044-08-19

AI Technical Summary

Technical Problem

The existing control strategies of fuel cells in microgrids have limitations, especially the poor performance of control for highly nonlinear systems, which makes it difficult to accurately manage dynamic power needs.

Method used

The TD3 algorithm in deep reinforcement learning and combined with deep neural networks, an intelligent control strategy for the photovoltaic-solid oxide fuel cell hybrid energy system is constructed, and the power output is optimized through the policy network and the value network to meet user needs and ensure the stable operation of the power grid.

Benefits of technology

It significantly improves the efficiency and response speed of fuel cells, performs better than traditional PID control methods, realizes more efficient power system control, and enhances the performance and reliability of the microgrid.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118659447B_ABST
    Figure CN118659447B_ABST
Patent Text Reader

Abstract

The present invention discloses a photovoltaic-solid oxide fuel cell hybrid energy system control method, system and equipment, including: using a PV-SOFC hybrid energy system as a simulation model; studying the control strategy of the simulation model in a networked operation mode and an island operation mode; the simulation model adaptively adjusts the continuous and stable power output to be incorporated into the power grid based on the TD3 algorithm to ensure the stable operation of the power grid and meet user needs. The present invention adopts the TD3 algorithm, which is precisely adjusted to adapt to the specific needs of fuel cells, such as state and action space definition and reward function design, thereby significantly improving the effect of the control strategy and verifying its potential value in power system control.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of hybrid energy systems, and in particular to a photovoltaic-solid oxide fuel cell hybrid energy system control method, system and equipment. Background Art

[0002] With the focus on renewable energy transformation, microgrid and fuel cell technology have gradually become a hot topic of research, not only because of their potential value in reducing carbon emissions and achieving sustainable energy supply, but also because of the technical challenges and opportunities they present in practical applications.

[0003] At present, relevant research abroad has deeply explored the flexibility and stability of microgrids, emphasizing the important position of microgrids in future energy systems. Others have focused on the specific application of fuel cells in microgrids and pointed out their key role in meeting peak loads.

[0004] Domestically, a new microgrid management strategy has been proposed, which effectively integrates multiple energy resources and ensures a stable supply of electricity through advanced algorithms. In addition, related research has further optimized the control strategy of fuel cells, thereby improving their efficiency and stability.

[0005] These studies not only show that microgrid and fuel cell technology have received widespread attention both at home and abroad, but more importantly, they reveal important technical difficulties and challenges in current research. Although existing research has provided valuable experience and knowledge for the development of microgrids, how to further improve the performance and reliability of fuel cells in microgrids is still an urgent problem to be solved.

[0006] As an energy conversion technology with high efficiency and low emission, the core competitiveness of fuel cells lies in efficient control strategies. Although traditional PID control has its application in the early stage, it has limitations in controlling highly nonlinear systems such as fuel cells. Summary of the invention

[0007] The object of the present invention is to provide a photovoltaic-solid oxide fuel cell hybrid energy system control method, system and device to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A photovoltaic-solid oxide fuel cell hybrid energy system control method, comprising:

[0010] The PV-SOFC hybrid energy system is used as a simulation model;

[0011] Study the control strategies of the simulation model in networked operation mode and island operation mode;

[0012] The simulation model adaptively adjusts the continuous and stable power output based on the TD3 algorithm and integrates it into the power grid to ensure the stable operation of the power grid and meet user needs.

[0013] Furthermore, the PV-SOFC hybrid energy system specifically includes:

[0014] Photovoltaic modules, which are used to convert solar radiation directly into electricity;

[0015] SOFCs, which convert hydrogen and oxygen directly into electricity;

[0016] Energy storage devices, which store excess electrical energy and release it when needed;

[0017] and a control system for controlling the release of electrical energy.

[0018] Furthermore, the PV-SOFC hybrid energy system meets the following internal operating condition constraints: temperature constraints of the PV modules, temperature constraints of the SOFC, and power and load matching constraints.

[0019] Furthermore, the temperature constraint of the PV module is expressed as the following formula:

[0020] ;

[0021] in, is the efficiency of the PV module, is the efficiency at standard temperature, is the temperature coefficient, is the actual operating temperature, is the reference temperature.

[0022] Furthermore, the temperature constraint of SOFC is expressed as:

[0023] ;

[0024] is the operating temperature of SOFC, and are respectively the minimum and maximum operating temperatures allowed for SOFC.

[0025] Furthermore, the power and load matching constraints are expressed as:

[0026] ;

[0027] is the total power generated by the system, is the power generated by the photovoltaic module, is the power generated by SOFC, is the load demand of the system.

[0028] Furthermore, the control strategy of the simulation model in the networked operation mode includes optimization objectives and rewards.

[0029] Function construction.

[0030] Furthermore, the optimization goal is to minimize the long-term operating cost, and the optimization objective function is as follows:

[0031] ;

[0032] and Respectively represent the real-time electricity purchase price and the real-time electricity sales price. represents the unit operating cost of SOFC, , , Respectively represent the real-time main grid power, electricity sales power and SOFC power;

[0033] Constraints:

[0034] ;

[0035] ;

[0036] in, is the real-time electrolyzer power, is the maximum electrolyzer power, is the real-time hydrogen storage tank pressure, and They are the minimum hydrogen storage tank pressure and the maximum hydrogen storage tank pressure respectively.

[0037] Furthermore, the reward function is constructed based on the system optimization objectives and constraints to express the reward function:

[0038] ;

[0039] ;

[0040] Pick , They are and The trade-off coefficient between the two parts, It is the penalty term when the remaining hydrogen storage is lower than the lower limit of the constraint condition.

[0041] Furthermore, the control strategy of the simulation model in the island operation mode includes defining the optimization target by minimizing the probability of power failure and power abandonment. The optimization target is expressed as follows:

[0042] ;

[0043] in, is the load demand of the system, is the actual operating temperature of the PV module, represents the maximum excess power that the system can absorb, is the main network power.

[0044] Furthermore, the simulation model adaptively adjusts the continuous and stable power output into the grid based on the TD3 algorithm to ensure the stable operation of the grid and meet user needs, including: using deep neural networks to estimate strategies and Value function, reward function Ensure continuous and stable power output of the simulation model.

[0045] Furthermore, deep neural networks are used to estimate strategies, including policy networks and value networks.

[0046] Furthermore, the policy network includes the input of the network as the observed state of the environment s , the hidden layer is a fully connected neural network, the activation function is ReLU, the output layer is also a fully connected neural network, the activation function is the hyperbolic tangent function, and the final output state s Actions taken a , the function limits the output to the interval [-1,1] and obtains the actual action value through linear transformation.

[0047] Furthermore, the value network includes the observation state of the environment as the input of the network s and the actions taken a , the hidden layer is a fully connected neural network, the activation function is ReLU, the output layer is also a fully connected neural network, there is no activation function, and the final output state action value Q( s , a ).

[0048] Further, using deep neural networks to estimate value functions includes:

[0049] Experience replay: The agent interacts with the environment and collects experience tuples and stores it in the playback buffer D middle;

[0050] Policy network training: Using samples from the replay buffer, train the policy network to maximize the cumulative reward: ,in By Strategy Network Decide, is a small exploration noise:

[0051] ;

[0052] Value network training: Use samples from the playback buffer to train A value network to estimate the value of each state-action pair, The learning goal of the value network is to minimize the TD error:

[0053] ;

[0054] in, is the discount factor, is the goal Value network, and is the target policy network, is the loss function, It is experience replay;

[0055] Target network updates: Regularly update target policy networks and targets The value network is used to reduce the estimation variance through a soft update strategy:

[0056] ;

[0057] in, and are the parameters of the original network and the target network respectively, is the soft update coefficient;

[0058] Policy noise addition: Introduce target policy noise to increase exploration:

[0059] ;

[0060] in, is the variance of the noise;

[0061] At each time step, the agent chooses an action and receive rewards and the new state , this information is used to update the experience replay buffer D , is a normal distribution.

[0062] Furthermore, the reward function Defined as:

[0063] ;

[0064] in, is the real-time output power of the PV-SOFC system, is the actual power demand, is the input power of the system, which consists of the power generated by the PV panels and the power provided by the SOFC. , and is the weight parameter.

[0065] To achieve the above object, the present invention also provides the following technical solutions:

[0066] Photovoltaic-solid oxide fuel cell hybrid energy system control system, including:

[0067] A model unit, used to use a PV-SOFC hybrid energy system as a simulation model;

[0068] Research unit, used to study the control strategies of the simulation model in networked operation mode and island operation mode;

[0069] The output unit is used to simulate the model based on the TD3 algorithm to adaptively adjust the continuous and stable power output and integrate it into the power grid, ensuring the stable operation of the power grid and meeting user needs.

[0070] To achieve the above object, the present invention also provides the following technical solutions:

[0071] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0072] To achieve the above object, the present invention also provides the following technical solutions:

[0073] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of any one of the methods described above.

[0074] Compared with the prior art, the present invention has the following beneficial effects:

[0075] Based on the sequential decision model between microgrid operation optimization and deep reinforcement learning, this paper proposes a microgrid optimization method of deep reinforcement learning, and deeply defines the state, action and reward function of the reinforcement learning model. In particular, the TD3 algorithm is selected, whose robustness and performance are fully reflected in the continuous action space, providing new possibilities for intelligent control strategies.

[0076] The present invention deepens the research on the selection of state space and feature engineering in the control strategy of fuel cells, ensuring that the TD3 algorithm is highly matched with the characteristics of fuel cells. At the same time, by clarifying the definition and constraints of the action space, the practical feasibility of the algorithm output is ensured. After a series of in-depth experiments, it is confirmed that the TD3 algorithm has obvious advantages in improving the efficiency and response speed of fuel cells, and its performance is more outstanding than the traditional PID control method.

[0077] The present invention adopts the TD3 algorithm, which is precisely adjusted to adapt to the specific needs of fuel cells, such as state and action space definition and reward function design, thereby significantly improving the effect of the control strategy and verifying its potential value in power system control. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 This is a flow chart of the photovoltaic-solid oxide fuel cell hybrid energy system control method of the present invention.

[0079] Figure 2 Schematic diagram of the strategic network structure of the present invention.

[0080] Figure 3 It is a schematic diagram of the value network structure of the present invention.

[0081] Figure 4 This is a diagram of the PV-SOFC hybrid energy system of the present invention.

[0082] Figure 5 This is a power flow diagram of the hybrid energy system of the SOFC fuel cell of the present invention in island mode.

[0083] Figure 6 This is the cumulative reward curve of the DDPG algorithm of the fuel cell of the present invention in the island mode.

[0084] Figure 7 This is a hydrogen flow rate diagram under the island mode of the present invention.

[0085] Figure 8 is the sample value of the DDPG algorithm in the island mode of the present invention.

[0086] Fig. 9 This is a power flow diagram of the networked hybrid energy system of the present invention.

[0087] Fig.10 This is a cumulative reward curve diagram of the DDPG algorithm in the networking mode of the present invention.

[0088] Fig.11 This is a hydrogen flow rate diagram in the networking mode of the present invention.

[0089] Fig.12This is a sample value diagram of the DDPG algorithm in the networking mode of the present invention.

[0090] Fig.13 It is a diagram of the internal structure of the computer device of the present invention. DETAILED DESCRIPTION

[0091] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0092] See also Figures 1 to 13 , the present invention provides a technical solution:

[0093] This invention is committed to solving the core problem in power demand side management by adopting the advanced TD3 algorithm in reinforcement learning to optimize the control mechanism of the fuel cell system. Fuel cell technology, due to its high efficiency and low carbon characteristics, plays a vital role in the sustainable energy transition. However, the precise control of the fuel cell system to meet dynamic power demand is still an unresolved challenge.

[0094] In the environment of power demand side management, the traditional PID control method faces many limitations. To this end, based on the sequential decision model between microgrid operation optimization and deep reinforcement learning, this paper innovatively proposes a microgrid optimization method of deep reinforcement learning, and deeply defines the state, action and reward function of the reinforcement learning model. In particular, the TD3 algorithm is selected, and its robustness and performance are fully reflected in the continuous action space, providing new possibilities for intelligent control strategies.

[0095] In the control strategy of fuel cells, the selection of state space and feature engineering research have been deepened to ensure that the TD3 algorithm is highly matched with the characteristics of fuel cells. At the same time, by clarifying the definition and constraints of the action space, the practical feasibility of the algorithm output is ensured. After a series of in-depth experiments, it is confirmed that the TD3 algorithm has obvious advantages in improving the efficiency and response speed of fuel cells, and its performance is more outstanding than the traditional PID control method.

[0096] In summary, this paper demonstrates the great value and potential of reinforcement learning in power demand side management. It provides strong support for the continuous advancement of fuel cell technology and the popularization of sustainable energy. Therefore, this paper aims to explore the control strategy of fuel cells in depth, especially in the microgrid environment, hoping to provide theoretical and technical support for the realization of a more efficient and stable microgrid system.

[0097] Research content

[0098] The global energy transformation is making headway, and microgrid technology is particularly critical in the clean energy transformation due to its high flexibility and integration capabilities. The present invention constructs an intelligent microgrid system that integrates photovoltaics, wind energy, solid oxide fuel cells (SOFCs) and battery energy storage, which is designed to serve residential, commercial and medical facilities, and aims to ensure the stability, economic benefits and environmental friendliness of power supply.

[0099] Fuel cells, as an energy conversion technology with high efficiency and low emissions, have their core competitiveness in efficient control strategies. Although traditional PID control has its applications in the early stages, it has limitations in controlling highly nonlinear systems such as fuel cells.

[0100] The TD3 algorithm of reinforcement learning has emerged in the field of control in recent years, especially in dealing with continuous action spaces. This invention adopts the TD3 algorithm, which is precisely adjusted to adapt to the specific needs of fuel cells, such as state and action space definition and reward function design, thereby significantly improving the effect of the control strategy and verifying its potential value in power system control.

[0101] Basics: Reinforcement Learning Algorithms

[0102] Introduction to Reinforcement Learning

[0103] Reinforcement learning is a core subfield of machine learning. Its main goal is to train intelligent agents to make decisions that maximize cumulative rewards in the process of interacting with the environment. In this process, the intelligent agent constantly explores and tries and fails, and receives corresponding rewards or penalties based on the actions taken. Its ultimate goal is to master a set of optimal strategies to ensure that the intelligent agent can make the best choice in various environmental states.

[0104] Markov decision making

[0105] The Markov decision process (MDP) is a mathematical modeling framework used to describe systems involving randomness and decision making. It mainly consists of four elements: state, action, transition probability, and reward.

[0106] state( ): At any time t, the characteristic quantity of the environment observed by the agent is , S represents the space of all possible states.

[0107] action( ): At any time t, the actions that the agent can take are , A represents the space of all possible actions.

[0108] award( ): Reward is a feedback signal provided by the environment to evaluate the quality of an action. Rewards can be positive (encourage actions), negative (punish actions), or zero (neutral feedback). To evaluate the effect of an agent taking action a in state s, the mathematical form of reward is as follows: .

[0109] Learning Algorithms

[0110] The learning algorithm is a common value-based reinforcement learning algorithm, assuming that there is a quadruple of interactions between the agent and the environment , then the iterative update formula is as follows: .

[0111] TD3 algorithm: Introduction to TD3 algorithm

[0112] TD3 (Twin Delayed Deep Deterministic Policy Gradient) is a reinforcement learning algorithm designed specifically to solve control problems in continuous action spaces. The TD3 algorithm is an improved version of DDPG (Deep Deterministic Policy Gradient) that introduces a number of enhancements to improve the stability and performance of the algorithm. The core goal of the TD3 algorithm is to minimize The mean squared error of the value network is used to learn the best strategy so that the agent can obtain the maximum cumulative reward in the continuous action space. Due to its excellent robustness and performance, the TD3 algorithm is widely used in various problems requiring continuous control, including robot control, autonomous driving, and energy management.

[0113] In summary, the TD3 algorithm combines deep learning and classical reinforcement learning techniques to form a robust and efficient framework that is particularly suitable for continuous control tasks. The design details and robustness of its algorithm make it perform well in practical applications and provide new possibilities in the field of continuous control.

[0114] Policy Network

[0115] The TD3 algorithm adopts a deterministic strategy, that is, it directly outputs a certain action in a given state, rather than outputting the probability distribution of the action. This helps to handle continuous action spaces and improve the efficiency of the algorithm.

[0116] like Figure 2As shown in the figure, the input of the network is the observed state s of the environment, the hidden layer is a fully connected neural network, the activation function is ReLU, the output layer is also a fully connected neural network, the activation function is the hyperbolic tangent (tanh) function, and finally the action a taken under the output state s is limited to the interval [-1,1]. The actual action value can be obtained through linear transformation.

[0117] Value Network

[0118] like Figure 3 As shown in the figure, the input of the TD3 algorithm network is the observed state s of the environment and the action a taken. The hidden layer is a fully connected neural network with the activation function ReLU. The output layer is also a fully connected neural network, but there is no activation function. Finally, the state action value Q(s,a) is output. In order to cooperate with the TD3 algorithm for learning and improve the convergence speed of the neural network, the original state information and action information are normalized and then used as the input of the neural network.

[0119] Hydrogen fuel cell system

[0120] Introduction to fuel cell system: Hydrogen fuel cell (HFC) system represents the cutting-edge electrochemical energy conversion technology. This paper adopts the anode-supported SOFC (Anode-Supported SOFC) electrochemical model, which has been experimentally verified.

[0121]

[0122] This electrochemical reaction not only ensures high efficiency of energy conversion, but its only by-product is water, thus ensuring complete environmental friendliness.

[0123] Proton Exchange Membrane

[0124] The core structure of a hydrogen fuel cell is mainly composed of the following key components:

[0125] Proton Exchange Membrane (PEM): This is a semi-permeable membrane that allows only protons ( H + ) passes through, thus realizing the electron transfer between the anode and the cathode.

[0126] Anode and cathode: At the interface of these two electrodes, hydrogen and oxygen are oxidized and reduced, respectively, forming an electron flow and driving the external circuit.

[0127] Catalyst layer: The catalyst layer is located at the contact surface of the anode and cathode and is usually composed of highly active materials (such as platinum) to promote the oxidation of hydrogen and the reduction of oxygen.

[0128] Electrolyte membrane: The main task of the proton exchange membrane is to ensure efficient transmission of electrons without allowing gas molecules to pass directly through.

[0129] Current Collector: This is a highly conductive material that collects the electrons generated from the electrochemical reaction and passes them to an external circuit.

[0130] Hydrogen fuel cells convert basic chemical reactions into continuous and efficient electrical energy output through a series of carefully designed and highly integrated components, providing a powerful solution for future renewable energy systems.

[0131] Design of hybrid energy system based on power demand side management

[0132] Introduction to Power Demand Side Management

[0133] Demand side management plays a vital role in the current power system, especially in the context of a rapidly changing energy environment and growing electricity demand. Electricity demand usually changes over time, so intelligent control methods are needed to adapt to these changes. The goals of demand side management cover many aspects, including optimization of energy utilization, reduction of load peaks, improvement of system efficiency, and efficient integration of renewable energy. To achieve these goals, reinforcement learning methods, especially the TD3 algorithm, have been introduced into the field of demand side management to achieve more intelligent, flexible and adaptive control strategies.

[0134] Simulation model construction of PV-SOFC hybrid energy system

[0135] System composition and working principle

[0136] The PV-SOFC hybrid energy system is an important evolution of microgrid technology, combining photovoltaic (PV) technology and solid oxide fuel cell (SOFC) technology to provide an economical and reliable power solution.

[0137] System Configuration

[0138] like Figure 4 Photovoltaic (PV) modules: PV modules are mainly composed of a series of solar panels that can directly convert solar radiation into electrical energy. Its conversion efficiency, stability and output characteristics are related to the installation location, weather conditions and the type of solar panels.

[0139] Solid Oxide Fuel Cell (SOFC): SOFC is a high-temperature fuel cell that uses ceramic material as the electrolyte and can directly convert hydrogen and oxygen into electrical energy at an operating temperature of 800-1000°C.

[0140] Energy storage devices: To address the intermittent and unstable nature of photovoltaic-generated electricity, batteries or other types of energy storage devices are usually added to the system to store excess electricity and release it when needed.

[0141] Control system: The control system is the brain of the entire microgrid, responsible for managing and optimizing the operation of all components to ensure stable operation of the grid and meet user needs.

[0142] Working principle: The core idea of ​​the PV-SOFC hybrid energy system is to improve the energy utilization of the entire system through the complementary working characteristics of photovoltaic modules and SOFC. When there is sufficient sunlight, the photovoltaic modules serve as the main source of power supply; at night or when there is insufficient light, SOFC intervenes to provide a stable power output for the system. At the same time, any excess electricity will be stored for later use.

[0143] The control system plays a vital role in the whole process. It dynamically adjusts the working status of each component according to the real-time energy demand and supply to ensure the efficient and stable operation of the microgrid.

[0144] The biggest advantage of this system is its high adaptability to various energy inputs, whether it is sufficient solar radiation, continuous hydrogen supply, or any combination between the two, the PV-SOFC hybrid energy system can provide stable and economical power output.

[0145] Power flow and hydrogen flow analysis

[0146] In the PV-SOFC hybrid energy system, the power flow is very critical because it directly affects the operating efficiency and stability of the system. The power flow mainly depends on the output of the PV module, the power generation capacity of the SOFC, and the storage and release status of the battery or other energy storage device.

[0147] Electricity generated by photovoltaics: During the day, when solar radiation reaches a certain intensity, photovoltaic modules will become the main electricity producer. Excess electricity will be transferred to energy storage devices for storage and use at night or when there is low radiation.

[0148] Electricity generated by SOFC: When photovoltaic power generation is insufficient, such as at night or on cloudy days, SOFC will start working as the main source of electricity supply. In addition, SOFC can also operate during low load during the day to provide basic power needs for the system.

[0149] Batteries or energy storage devices: They act as a "buffer" and "bridge" in the entire system. When there is excess photovoltaic capacity, they are responsible for storing excess electricity; when system demand exceeds real-time capacity, they will release previously stored electricity.

[0150] Hydrogen flow analysis: For SOFC, hydrogen supply is the basis of its operation. Ensuring stable and efficient hydrogen flow is crucial for the normal operation of the entire system.

[0151] Hydrogen source: The hydrogen in the system can come from an external supply, such as a hydrogen tank or hydrogen produced by electrolysis of water.

[0152] Hydrogen storage: In order to ensure the continuous operation of SOFC, there is usually hydrogen storage equipment, such as high-pressure hydrogen tanks or metal hybrid hydrogen storage systems.

[0153] Hydrogen transportation: From storage equipment to SOFC, the transportation of hydrogen requires a set of sophisticated pipelines and regulation systems to ensure that the input hydrogen flow of SOFC matches its working requirements.

[0154] In short, ensuring the correct power flow and hydrogen flow is crucial for the efficient and stable operation of the PV-SOFC hybrid energy system. Mistakes or deficiencies in any link may lead to performance degradation or shutdown of the entire system.

[0155] Operating condition constraints within the system

[0156] To ensure efficient and stable operation of the PV-SOFC hybrid energy system, the following internal operating condition constraints must be met:

[0157] 1. Temperature constraints of photovoltaic modules: The efficiency of photovoltaic modules is related to their operating temperature, which can usually be expressed as the following formula:

[0158]

[0159] in, is the efficiency of the PV module, is the efficiency at standard temperature, is the temperature coefficient, is the actual operating temperature, is the reference temperature (e.g. 25°C).

[0160] 2. Temperature constraints of SOFC: To ensure the normal operation of SOFC and avoid overheating, its operating temperature must be maintained within a certain range, which can be expressed as:

[0161]

[0162] in, Yes, the operating temperature, and are the minimum and maximum allowable operating temperatures respectively.

[0163] 3. Power and load matching constraints: To ensure stable operation of the system, the total power generated by the PV panels and SOFC must match the load demand, which can be expressed as:

[0164]

[0165] in, is the total power generated by the system, is the power generated by the photovoltaic module, is the power generated, is the load demand of the system.

[0166] Considering these constraints, the design and operation of PV-SOFC hybrid energy systems require sophisticated strategies and management to ensure that individual components and the entire system operate in optimal working conditions.

[0167] Research on control strategy under network operation mode

[0168] Optimization goal definition: Minimize long-term operating costs

[0169] The optimization goal is to minimize the long-term operating cost. This paper considers the following benefits or costs: electricity sales revenue, electricity purchase costs, and system operating costs. The construction cost of the hybrid energy system is a one-time fixed cost, while the operating costs of other equipment other than SOFC can be approximately considered unchanged over a sufficiently long period of time. Therefore, the variable costs in the system operating costs only need to consider the operating costs of SOFC and the electricity sales and purchase costs. The optimization objective function is as follows:

[0170]

[0171] in, and Respectively represent the real-time electricity purchase price and the real-time electricity sales price. represents the unit operating cost of SOFC, , , They represent real-time main grid power, electricity sales power and SOFC power respectively.

[0172] Constraints:

[0173]

[0174]

[0175] in, is the real-time electrolyzer power, For the real-time hydrogen storage tank pressure, this article specifies a maximum pressure of 15,000 psi.

[0176] Reward function construction: system optimization objectives and constraints

[0177] In reinforcement learning, in order to enable the intelligent agent to learn how to operate the system to achieve the desired optimization goal, we use a reward function to provide feedback for each action of the agent. This paper directly constructs the reward function based on the system optimization goal and constraints:

[0178]

[0179]

[0180] This article takes , is the trade-off coefficient between the two parts.

[0181] Study on PV-SOFC hybrid energy system under island operation mode

[0182] Optimization objective definition: Minimize the probability of power outage and power abandonment

[0183] In the island operation mode, the PV-SOFC hybrid energy system is not connected to the main grid, but is powered independently. The power failure probability refers to the percentage of time that the load demand cannot be met in the total system operation time, reflecting the system's ability to meet the load demand; the power abandonment probability refers to the percentage of time that the system generates excess power in the total system operation time. The smaller the probability, the higher the system power utilization rate. The optimization objectives are as follows:

[0184]

[0185] in, Indicates the maximum excess power that the system can absorb. In this paper, .

[0186] Fuel cell control method based on TD3

[0187] State-space design and feature engineering

[0188] Before applying the TD3 algorithm to fuel cell control, the definition of the state space should be carefully considered because it is directly related to the performance of the reinforcement learning algorithm.

[0189] For a fuel cell system, a suitable state space should include the following key features:

[0190] Battery Temperature: Battery temperature is critical to performance and should include the current temperature and its rate of change.

[0191] Humidity: The distribution of water also has an important impact on battery operation, and relative humidity information such as the humidity of hydrogen and oxygen should be included in the status.

[0192] Voltage and Current: The voltage and current of a battery are core performance indicators and must be considered in the state.

[0193] Hydrogen and oxygen supply rates: Controlling the supply rates of hydrogen and oxygen is critical to the battery and should also be part of the status.

[0194] In addition to selecting appropriate state features, in order to adapt to the TD3 algorithm, feature engineering is also required to map the continuous state space into a processable range, which is achieved through techniques such as standardization and dimensionality reduction.

[0195] Design of reward function

[0196] The reward function plays a decisive role in reinforcement learning because it provides feedback to the agent when interacting with the environment. Especially in fuel cell control, an appropriate reward function can guide the agent to more efficient and sustainable energy management. The present invention attaches importance to energy efficiency as a core indicator because in energy-intensive applications, how to maximize the ratio of power output to input becomes the primary focus.

[0197] In order to reflect the dynamic adjustment of the PV-SOFC system under the change of real-time electricity demand, the reward function needs to consider the following elements:

[0198] Power Matching Rewards: Rewards are provided for better power matching based on the difference between the output power of the PV-SOFC system and the real-time electricity demand.

[0199] Energy efficiency: Encourage systems to meet electricity demand while ensuring minimal energy loss.

[0200] Based on the above considerations, the reward function can be defined as:

[0201]

[0202] in,

[0203] is the real-time output power of the PV-SOFC system.

[0204] is the actual power demand.

[0205] is the input power of the system, which consists of the power generated by photovoltaics and the power provided by SOFC.

[0206] , and is a weight parameter that needs to be adjusted through experiments to ensure the optimal performance of the system.

[0207] The goal of this reward function design is to ensure that the PV-SOFC system can not only meet the real-time power demand, but also achieve efficient energy utilization in the process. The first item encourages power matching to reduce the deviation between system output and actual demand; while the second item emphasizes energy efficiency to ensure that energy utilization is maximized while meeting demand. This design enables the PV-SOFC system to adjust more intelligently and respond to different environmental changes, thereby achieving better performance.

[0208] Reinforcement Learning Process

[0209] Once the state space, action space, and reward function are determined, the reinforcement learning process can be started. The TD3 algorithm uses a deep neural network to estimate the policy and value function and constantly interacts with the environment to update these estimates.

[0210] Experience replay: The agent interacts with the environment and collects experience tuples And store it in the playback buffer D.

[0211] Policy network training: Using samples from the replay buffer, train the policy network to maximize the cumulative reward: ,in By Strategy Network Decide, is a small exploration noise:

[0212]

[0213] Value network training: Use samples from the playback buffer to train A value network to estimate the value of each state-action pair, The learning goal of the value network is to minimize the TD error: ;

[0214] in, is the discount factor, is the goal Value network, and is the target policy network.

[0215] Target network updates: Regularly update target policy networks and targets The value network is used to reduce the estimation variance through a soft update strategy:

[0216] ;

[0217] in, and are the parameters of the original network and the target network respectively, is the soft update coefficient;

[0218] Policy noise addition: Introduce target policy noise to increase exploration:

[0219] ;

[0220] in, is the variance of the noise.

[0221] At each time step, the agent chooses an action and receives a reward and a new state, which is used to update the experience replay buffer.

[0222] Study Setting

[0223] Simulation Environment

[0224] This study uses a simulation-based fuel cell system model that includes multiple state variables: temperature, humidity, current and voltage, as well as control variables, hydrogen and oxygen supply rates. This simulation environment allows simulating the performance of the fuel cell system under different operating conditions.

[0225] Performance Indicators

[0226] Three performance indicators were used to comprehensively evaluate the effectiveness of the control method, including efficiency, reliability, and power demand optimization. The efficiency indicator is used to measure the effectiveness of the system's energy utilization, the reliability indicator reflects the system's stability and anti-interference, and the power demand optimization indicator is used to evaluate whether the system can intelligently adjust the supply rate according to actual demand to maximize the power demand.

[0227] Algorithm parameters

[0228] The hyperparameters of the TD3 algorithm were carefully adjusted to ensure optimal performance in fuel cell control. After multiple rounds of experiments, the parameter configuration suitable for the simulation environment was selected.

[0229] Example simulation and result analysis

[0230] Island Mode

[0231] In our experiments, we specifically observed the operation characteristics of SOFC fuel cells in island mode. Figure 5 It can be clearly seen that during the period from 0:00 to 24:00, when the solar power decreases, the peak of electricity consumption occurs. This phenomenon shows that in the absence of solar supply, SOFC fuel cells begin to decompose hydrogen in large quantities to generate electricity to meet the power demand. This ability emphasizes the important role of SOFC fuel cells in energy complementation and responding to peak power demand. In addition, compared with traditional power supply methods, this response characteristic of SOFC fuel cells provides a strong guarantee for grid stability, especially when the supply of renewable energy is unstable.

[0232] During the optimization of the DDGP (Deep Deterministic Policy Gradient) algorithm, we observed a significant trend: as time went on and the number of trainings increased, the algorithm's performance experienced a transition from negative rewards to positive rewards. This indicates that the algorithm gradually learned a better strategy that enabled it to obtain higher rewards for a given task. This gradual evolution from an initial bad strategy to an optimized strategy demonstrates the learning ability and adaptability of the DDGP algorithm. With more and more training, we can expect the algorithm to further optimize its strategy and reach higher reward levels.

[0233] from Figure 6 The effect of the fuel cell during peak electricity demand can be clearly observed in Figure 1. In particular, when the required electricity increases, the hydrogen flow rate increases significantly, which means that during periods of increased electricity demand, the fuel cell works more actively to generate more electricity to meet the demand. This correspondence shows that the fuel cell can effectively adjust its output to adapt to changing electricity demand.

[0234] In addition, if Figure 7 As shown in the figure, the graph of hydrogen flow rate shows multiple peaks and valleys, which may be related to the electricity consumption pattern in daily life. For example, there may be higher electricity demand during certain periods of daily life (such as morning and evening), while demand may decrease during other periods. The fuel cell adapts to this fluctuation by increasing the hydrogen flow rate, thereby ensuring a stable power supply.

[0235] In summary, fuel cells have proven their ability to cope with peak electricity demand, making them a very reliable and efficient energy solution.

[0236] In the DDPG algorithm, we introduced a noise network that conforms to the normal distribution to simulate the characteristics of a 24-hour time series. In this way, our goal is to increase the exploratory nature of the algorithm so that it can better cope with a variety of different environmental situations. This normally distributed noise can help the algorithm explore the state space more extensively in the early stages, avoid local optimality, and gradually converge to the optimal strategy as learning deepens.

[0237] From the provided Figure 8 In the figure, we can see that the sample values ​​show an obvious upward trend in the early stage, which may reflect the large-scale exploration of the algorithm at the beginning. Over time, the sample values ​​begin to fluctuate stably within a range, which indicates that the algorithm may have found a relatively optimized strategy and is fine-tuning around it.

[0238] In general, by introducing this normally distributed noisy network, we hope that the DDPG algorithm can more effectively learn and optimize its strategy to cope with various complex environmental changes.

[0239] Networking Mode

[0240] from Fig. 9 As can be seen from the graph, the solar power output is mainly concentrated in the time interval of 25 to 125, which is likely to correspond to daytime. As the position of the sun changes, the solar power output also fluctuates periodically. The obvious peak occurs during the period when the sun is at its highest altitude.

[0241] The red and green lines represent the power output of the SOFC fuel cell. When the solar power decreases, such as in the time intervals of 0 to 25 and 125 to 175, the power output of the SOFC fuel cell increases, especially during peak power consumption.

[0242] During peak hours, solar energy may not be able to meet all electricity demand, so other energy sources, such as SOFC fuel cells, are needed to provide additional power. The figure shows that SOFC fuel cells decompose hydrogen in large quantities to generate electricity during these periods to ensure a stable power supply.

[0243] this Fig. 9 The complementarity between solar energy and SOFC fuel cells in the interconnected mode is emphasized. By integrating different energy sources, changes in electricity demand can be more effectively responded to and the stability of the power system can be ensured.

[0244] In summary, this Fig. 9 We provide valuable information on how solar energy and SOFC fuel cells can be effectively used in a grid-connected mode to meet changing electricity demands. It is crucial to take this complementarity into account when designing and planning power systems.

[0245] pass Fig.10 It can be seen that the DDGP algorithm shows a clear trend in the optimization process: as time goes by and the number of training times increases, the performance of the algorithm gradually changes from negative rewards to positive rewards. This shows that the algorithm gradually learns better strategies, which enables it to obtain higher rewards on specific tasks. This evolution from an initial bad strategy to an optimized strategy shows the learning ability and adaptability of the DDGP algorithm. As the number of training times increases, it can be expected that the algorithm will further optimize its strategy and achieve higher levels of rewards.

[0246] Fig.11The output changes of the fuel cell when it is in networked mode and operating in sync with the power grid are shown. When the grid demand reaches its peak, the fuel cell quickly increases its output, indicating that it has excellent responsiveness and can quickly adapt to changes in grid demand. This responsiveness is a key feature in networked mode and ensures stable operation of the grid. The multiple output peaks and valleys in the figure may correspond to the actual demand pattern of the grid. In daily life, the demand for electricity increases during certain periods (such as the rush hour on weekdays) and may decrease during other periods. The fuel cell is able to adaptively adjust its output to keep pace with the demand of the grid.

[0247] Stable power supply: Despite fluctuations in demand, the fuel cell always maintains a stable output, ensuring seamless connection with the power grid. This stability shows that fuel cells can not only be used as an independent power source, but can also be perfectly integrated with the traditional power grid system in interconnected mode.

[0248] Conclusion: The fuel cell showed extremely high adaptability and stability in the grid-connected mode, proving its great potential to work in conjunction with the main grid as an auxiliary power source. This further emphasizes the key role of fuel cells in future energy networks.

[0249] In the DDPG algorithm, a normally distributed noise network is used to simulate the characteristics of a 24-hour time series, aiming to improve the algorithm's exploration performance and make it better adapt to various environmental changes. This normally distributed noise helps to explore the state space extensively in the early stages, avoid falling into local optimality, and gradually converge to the optimal strategy as learning deepens.

[0250] From the provided Fig.12 It can be observed that in the early stage, the sample values ​​show a clear upward trend, which may reflect the large-scale exploration of the algorithm at the beginning. Over time, the sample values ​​begin to fluctuate stably within a range, indicating that the algorithm may have found a relatively optimized strategy and is fine-tuning around it.

[0251] Overall, by introducing a normally distributed noisy network, it is expected that the DDPG algorithm can more effectively learn and optimize its strategy to adapt to various complex environmental changes.

[0252] Conclusion and Analysis

[0253] This invention shows that the TD3-based fuel cell control method has potential in power demand side management, which can significantly improve efficiency, reliability and optimization of power demand. This research provides strong support for further research in the field of power demand side management and the application of fuel cell technology. The TD3-based reinforcement learning method provides new ideas and methods for the intelligent control of fuel cell systems, which is expected to promote the development of sustainable energy.

[0254] like Fig.13 As shown, the device controlled by the photovoltaic-solid oxide fuel cell hybrid energy system can be various types of computing devices such as desktop computers, notebooks, PDAs, and cloud servers. This device is composed of multiple components such as processors and memories, but is not limited thereto and may include other components or different combinations. The device examples provided here are for illustrative purposes only and are not restrictive.

[0255] The processor can be various types of devices, such as CPU, DSP, ASIC, FPGA, etc., and can also include programmable logic devices, discrete gates, transistor logic devices and discrete hardware components, etc. The memory is used to store computer programs and modules. The processor runs or executes the computer program modules stored in the memory and calls the data stored in the memory, thereby realizing various functions of maximum power point tracking of photovoltaic power generation systems.

[0256] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0257] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the photovoltaic-solid oxide fuel cell hybrid energy system control method are implemented.

[0258] If the photovoltaic-solid oxide fuel cell hybrid energy system control method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device.

[0259] Based on this understanding, the present invention implements all or part of the processes in the above-mentioned photovoltaic-solid oxide fuel cell hybrid energy system control method, and can also be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned photovoltaic-solid oxide fuel cell hybrid energy system control method can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or preset intermediate form, etc.

[0260] Computer-readable storage media may refer to any entity or device that can carry computer program code, including recording media, USB flash drives, mobile hard drives, disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), software distribution media, etc.

[0261] It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0262] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A photovoltaic-solid oxide fuel cell hybrid energy system control method, characterized in that: include: The PV-SOFC hybrid energy system is used as a simulation model; Study the control strategies of the simulation model in networked operation mode and island operation mode; The simulation model adaptively adjusts the continuous and stable power output into the grid based on the TD3 algorithm to ensure the stable operation of the grid and meet user needs; The simulation model adaptively adjusts the continuous and stable power output based on the TD3 algorithm to be integrated into the power grid to ensure the stable operation of the power grid and meet user needs. It includes: using deep neural networks to estimate strategies and Q-value functions, and using reward function R1 to ensure the continuous and stable power output of the simulation model; Using deep neural networks to estimate the Q-value function includes: Experience replay: The agent interacts with the environment and collects experience tuples s t ,a t ,r t ,s t ,s t+1 And store it in the playback buffer D; Policy network training: Using samples from the replay buffer, train the policy network to maximize the cumulative reward: max a Q(s t ,a), where action a is determined by the policy network π and ∈ is a small exploration noise: a=π(s t )+∈; Q-value network training: Using samples from the playback buffer, train the Q-value network to estimate the value of each state-action pair. The learning goal of the Q-value network is to minimize T D error: Where γ is the discount factor, Q′ is the target Q-value network, π′ is the target policy network, L is the loss function, and E is the experience replay; Target network update: Regularly update the target policy network and target Q value network to reduce the estimation variance, through the soft update strategy: θ′=τθ+(1-τ)θ′; Among them, θ and θ′ are the parameters of the original network and the target network respectively, and τ is the soft update coefficient; Policy noise addition: Introduce target policy noise to increase exploration: a=π(S t )+N(0,σ 2 ); Among them, σ 2 is the variance of the noise; At each time step, the agent chooses an action a t and receive a reward t and the new state s t+1 , used to update the experience replay buffer D, N is a normal distribution; The reward function R1 is defined as: Among them, P output is the real-time output power of the PV-SOFC system, P demand is the actual power demand, P input is the input power of the system, which consists of the power generated by the PV panels and the power provided by the SOFC, and α, β and γ are weight parameters.

2. The method according to claim 1, characterized in that The PV-SOFC hybrid energy system specifically includes: Photovoltaic modules, which are used to convert solar radiation directly into electricity; SOFCs, which convert hydrogen and oxygen directly into electricity; Energy storage devices, which store excess electrical energy and release it when needed; and a control system for controlling.

3. The method according to claim 2, characterized in that The PV-SOFC hybrid energy system meets the following internal operating condition constraints: temperature constraints of PV modules, temperature constraints of SOFC, and power and load matching constraints.

4. The method according to claim 3, characterized in that The temperature constraint of the PV module is expressed as the following formula: ηpv=η0-β(TT ref ); Where ηpv is the efficiency of the photovoltaic module, η0 is the efficiency at standard temperature, β is the temperature coefficient, T is the actual operating temperature, and T ref is the reference temperature.

5. The method according to claim 3, characterized in that The temperature constraint of SOFC is expressed as: T min ≤T SOFC ≤T amx ; T SOFC is the operating temperature of SOFC, T min and T amx are respectively the minimum and maximum operating temperatures allowed for SOFC.

6. The method according to claim 3, characterized in that The power and load matching constraints are expressed as: P generated =P PV +P SOFC =P load ; P generated is the total power generated by the system, P PV is the power generated by the PV module, P SOFC is the power generated by SOFC, P load is the load demand of the system.

7. The method according to claim 1, characterized in that The control strategy of the simulation model in the networked operation mode includes the construction of optimization objectives and reward functions.

8. The method according to claim 7, characterized in that The optimization goal is to minimize the long-term operating cost. The optimization objective function is as follows: and Respectively represent the real-time electricity purchase price and the real-time electricity sales price. represents the unit operating cost of SOFC, Respectively represent the real-time main grid power, electricity sales power and SOFC power; Constraints: TP min <TP t <TP max ; in, is the real-time electrolyzer power, is the maximum electrolyzer power, TP t is the real-time hydrogen storage tank pressure, TP min and TP max They are the minimum hydrogen storage tank pressure and the maximum hydrogen storage tank pressure respectively.

9. The method according to claim 8, characterized in that The reward function is constructed based on the system optimization objectives and constraints, and R represents the reward function: Take k1 and k2 as and The trade-off coefficient between the two parts, It is the penalty term when the remaining hydrogen storage is lower than the lower limit of the constraint condition.

10. The method according to claim 1, characterized in that The control strategy of the simulation model in the island operation mode includes defining the optimization target by minimizing the probability of power failure and power abandonment. The optimization target is expressed as follows: Among them, P load is the load demand of the system, T1 is the time, P ε Indicates the maximum excess power that the system can absorb, P g is the main grid power, P loss is the power loss of the system.

11. The method according to claim 1, characterized in that The deep neural network used to estimate the policy includes a policy network and a value network.

12. The method according to claim 11, characterized in that The policy network includes a network whose input is the observed state s of the environment, a hidden layer which is a fully connected neural network, an activation function which is ReLU, an output layer which is also a fully connected neural network, an activation function which is a hyperbolic tangent function, and finally an action a taken under the output state s. The function limits the output to the interval [-1,1] and obtains the actual action value through linear transformation.

13. The method according to claim 11, characterized in that The value network includes the network input as the observed state s of the environment and the action a taken, the hidden layer is a fully connected neural network, the activation function is ReLU, the output layer is also a fully connected neural network, there is no activation function, and finally the state action value Q(s,a) is output.

14. Photovoltaic-solid oxide fuel cell hybrid energy system control system, characterized in that: include: A model unit, used to use a PV-SOFC hybrid energy system as a simulation model; Research unit, used to study the control strategies of the simulation model in networked operation mode and island operation mode; Output unit, used for simulating the model to adaptively adjust the continuous and stable power output based on the TD3 algorithm and integrate it into the power grid, so as to ensure the stable operation of the power grid and meet the needs of users; The simulation model adaptively adjusts the continuous and stable power output into the power grid based on the TD3 algorithm, which is used to ensure the stable operation of the power grid and meet user needs: using deep neural networks to estimate the strategy and Q-value function to ensure the continuous and stable power output of the simulation model with the reward function R1; Using deep neural networks to estimate the Q-value function includes: Experience replay: The agent interacts with the environment and collects experience tuples s t ,a t ,r t ,s t ,s t+1 And store it in the playback buffer D; Policy network training: Using samples from the replay buffer, train the policy network to maximize the cumulative reward: max a Q(s t ,a), where action a is determined by the policy network π and ∈ is a small exploration noise: a=π(s t )+∈; Q-value network training: Using samples from the playback buffer, train the Q-value network to estimate the value of each state-action pair. The learning goal of the Q-value network is to minimize T D error: Where γ is the discount factor, Q′ is the target Q-value network, π′ is the target policy network, L is the loss function, and L is the experience replay; Target network update: Regularly update the target policy network and target Q value network to reduce the estimation variance, through the soft update strategy: θ′=τθ+(1-τ)θ′; Among them, θ and θ′ are the parameters of the original network and the target network respectively, and τ is the soft update coefficient; Policy noise addition: Introduce target policy noise to increase exploration: a=π(S t )+N(0,σ 2 ); Among them, σ 2 is the variance of the noise; At each time step, the agent chooses an action a t and receive a reward t and the new state s t+1 , used to update the experience replay buffer D, N is a normal distribution; The reward function R1 is defined as: Among them, P output is the real-time output power of the PV-SOFC system, P demand is the actual power demand, P input is the input power of the system, which consists of the power generated by the PV panels and the power provided by the SOFC, and α, β and γ are weight parameters.

15. The system of claim 14, wherein: The PV-SOFC hybrid energy system specifically includes: Photovoltaic modules, which are used to convert solar radiation directly into electricity; SOFCs, which convert hydrogen and oxygen directly into electricity; Energy storage devices, which store excess electrical energy and release it when needed; and a control system for controlling.

16. The system of claim 15, wherein: The PV-SOFC hybrid energy system meets the following internal operating condition constraints: temperature constraints of PV modules, temperature constraints of SOFC, and power and load matching constraints.

17. The system of claim 16, wherein: The temperature constraint of the PV module is expressed as the following formula: ηpv=η0-β(TT ref ); Where ηpv is the efficiency of the photovoltaic module, η0 is the efficiency at standard temperature, β is the temperature coefficient, T is the actual operating temperature, and T ref is the reference temperature.

18. The system of claim 16, wherein: The temperature constraint of SOFC is expressed as: T min ≤T SOFC ≤T amx ; T SOFC is the operating temperature of SOFC, T min and T amx are respectively the minimum and maximum operating temperatures allowed for SOFC.

19. The system of claim 16, wherein: The power and load matching constraints are expressed as: P generated =P PV +P SOFC =P load ; P generated is the total power generated by the system, P PV is the power generated by the PV module, P SOFC is the power generated by SOFC, P load is the load demand of the system.

20. The system of claim 14, wherein: The control strategy of the simulation model in the networked operation mode includes the construction of optimization objectives and reward functions.

21. The system of claim 20, wherein: The optimization goal is to minimize the long-term operating cost. The optimization objective function is as follows: and Respectively represent the real-time electricity purchase price and the real-time electricity sales price. represents the unit operating cost of SOFC, Respectively represent the real-time main grid power, electricity sales power and SOFC power; Constraints: TP min <TP t <TP max ; in, is the real-time electrolyzer power, is the maximum electrolyzer power, TP t is the real-time hydrogen storage tank pressure, TP min and TP max They are the minimum hydrogen storage tank pressure and the maximum hydrogen storage tank pressure respectively.

22. The system of claim 21, wherein: The reward function is constructed based on the system optimization objectives and constraints, and R represents the reward function: Take k1 and k2 as and The trade-off coefficient between the two parts, It is the penalty term when the remaining hydrogen storage is lower than the lower limit of the constraint condition.

23. The system of claim 14, wherein: The control strategy of the simulation model in the island operation mode includes defining the optimization target by minimizing the probability of power failure and power abandonment. The optimization target is expressed as follows: Among them, P load is the load demand of the system, T is the actual operating temperature of the PV module, P ε Indicates the maximum excess power that the system can absorb, P g is the main network power.

24. The system of claim 14, wherein: The deep neural network used to estimate the policy includes a policy network and a value network.

25. The system of claim 24, wherein: The policy network includes a network whose input is the observed state s of the environment, a hidden layer which is a fully connected neural network, an activation function which is ReLU, an output layer which is also a fully connected neural network, an activation function which is a hyperbolic tangent function, and finally an action a taken under the output state s. The function limits the output to the interval [-1,1] and obtains the actual action value through linear transformation.

26. The system of claim 24, wherein: The value network includes the network input as the observed state s of the environment and the action a taken, the hidden layer is a fully connected neural network, the activation function is ReLU, the output layer is also a fully connected neural network, there is no activation function, and finally the state action value Q(s,a) is output.

27. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method according to any one of claims 1 to 13 when executing the computer program.

28. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 13 are implemented.