Energy storage control method and energy storage system

Abstract

The invention discloses an energy storage control method and an energy storage system, and belongs to the field of energy storage control, and the method comprises the following steps: S1, constructing a multi-physics field coupling model; s2, taking the full life cycle comprehensive benefit maximization of the energy storage system as a target, constructing a multi-target function, and generating a Pareto frontier solution; s3, fusing physical constraint and deep learning of the multi-physics field coupling model, constructing a multi-fidelity physical information neural network model, obtaining a real-time estimation value and a short-term prediction result of the state of the energy storage system, and optimizing parameters of a multi-objective function based on the real-time estimation value and the short-term prediction result; and S4, a plurality of non-dominated solutions generated after parameter optimization are substituted into the multi-physics coupling model, and the operation state of the energy storage assembly is verified. By adopting the energy storage control method and the energy storage system, the operation cost, the energy loss, the assembly service life and the environmental influence of the energy storage system can be effectively balanced, and the multi-objective optimization efficiency and the complex working condition adaptability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage control, and particularly relates to an energy storage control method and an energy storage system. Background Art

[0002] With the large-scale application of renewable energy (such as solar energy and wind energy) globally, energy storage systems (such as batteries, supercapacitors, fuel cells) as the core link for suppressing the fluctuations of intermittent power generation, their efficient control and optimization management have become the key to realizing sustainable energy development. Hybrid energy storage systems (HESS) have become a current research hotspot due to the complementary advantages of combining multiple energy storage components (such as the high energy density of lithium batteries and the high power density of supercapacitors).

[0003] However, traditional energy storage systems mostly adopt a single meta-heuristic algorithm for energy management, which is prone to falling into local optima and difficult to simultaneously balance the multi-objective optimization requirements such as operating costs, energy losses, and component lifetimes. Moreover, during the operation of the energy storage system, the output of photovoltaic / wind power, load demand, and the health state (SOH) of components change in real time, and a single algorithm is difficult to quickly respond to multi-variable coupling scenarios.

[0004] At the same time, traditional state estimation relies on pure data-driven models (such as neural networks) or single physical models, suffering from the defects of lack of physical constraints and insufficient multi-component synergy. Specifically, pure data models lack physical mechanism constraints on electrochemical reactions and temperature field distributions, resulting in significant prediction errors under complex working conditions. And a multi-physical field coupling model of battery-supercapacitor-fuel cell has not been established, unable to reflect the dynamic interaction of energy flow between components. Summary of the Invention

[0005] The purpose of the present invention is to provide an energy storage control method and an energy storage system to solve the above technical problems.

[0006] To achieve the above purpose, the present invention provides an energy storage control method, including the following steps: S1. Considering the electro-thermal-mechanical multi-physical field coupling characteristics of lithium batteries, supercapacitors, and fuel cells in the energy storage system, construct a multi-physical field coupling model; S2. Based on the multi-physical field coupling model, with the goal of maximizing the comprehensive benefit of the entire life cycle of the energy storage system, construct a multi-objective function including operating costs, energy losses, and component lifetimes, and use a cuckoo-particle swarm hybrid algorithm to generate Pareto front solutions, obtaining multiple non-dominated solutions, and each non-dominated solution corresponds to a different power allocation scheme; S3. Integrate the physical constraints of the multi-physical field coupling model and deep learning, construct a multi-fidelity physical information neural network model, obtain real-time estimates and short-term prediction results of the state of the energy storage system, and optimize the parameters of the multi-objective function described in step S2 based on the real-time estimates and short-term prediction results; S4. Substitute the multiple non-dominated solutions generated in step S2 after parameter optimization into the multi-physical field coupling model described in step S1 respectively to verify the operating status of the energy storage components.

[0007] An energy storage system for implementing an energy storage control method, comprising: A multi-physical field coupling model construction module, configured to construct a multi-physical field coupling model by considering the electro-thermal-mechanical multi-physical field coupling characteristics of lithium batteries, supercapacitors, and fuel cells in the energy storage system; A power distribution scheme generation module, configured to construct a multi-objective function including operating cost, energy loss, and component life based on the multi-physical field coupling model with the goal of maximizing the comprehensive benefit of the entire life cycle of the energy storage system, and use a cuckoo-particle swarm hybrid algorithm to generate Pareto front solutions to obtain multiple non-dominated solutions, and each non-dominated solution corresponds to a different power distribution scheme; A parameter optimization module, configured to fuse the physical constraints of the multi-physical field coupling model with deep learning, construct a multi-fidelity physics-informed neural network model, obtain real-time estimation values and short-term prediction results of the energy storage system state, and optimize the parameters of the multi-objective function based on the real-time estimation values and short-term prediction results; A feasibility verification module, configured to substitute the multiple non-dominated solutions after parameter optimization into the multi-physical field coupling model respectively to verify the operating status of the energy storage components.

[0008] Therefore, the present invention adopts the above-mentioned energy storage control method and energy storage system, and the beneficial effects are as follows: 1. Utilizing multi-physical field coupling can accurately reflect the internal state of components (such as the aging rate of lithium batteries and the change of the equivalent series resistance of supercapacitors), providing real physical constraints for optimization; realizing the collaborative modeling of energy flow and thermal dynamics between components, and improving the reliability of system-level simulation; 2. Adopting a cuckoo-particle swarm hybrid algorithm can efficiently search for the non-dominated solution set of economic indicators and environmental indicators, avoiding falling into local optima; 3. Fusing physical prior knowledge and data-driven features can suppress the problem of lack of physical constraints in pure data models, and improve the estimation accuracy of state of charge, SOH, etc. (error < 5%); 4. By updating the power distribution scheme at future moments through a rolling prediction strategy and using a multi-physical field model to correct prediction errors, it can dynamically adapt to changes in load, temperature, etc., and ensure that the power distribution scheme always meets the operating constraints of components; In summary, through the full-process innovation of "accurate modeling-intelligent optimization-reliable prediction-closed-loop control", the present invention realizes an efficient balance among the economy, environment, and reliability of the energy storage system, is applicable to multi-scenario energy storage requirements such as grid-level and household use, and has significant engineering application value and technological advancement.

[0009] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0010] Figure 1 It is a flowchart of a method for controlling energy storage according to the present invention. Detailed Embodiments

[0011] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0012] It should be noted that the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server that includes a series of steps or units does not necessarily have to be limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products or devices.

[0013] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0014] As Figure 1 shown, a method for controlling energy storage includes the following steps: S1. Considering the electro-thermal-mechanical multi-physical field coupling characteristics of lithium batteries, supercapacitors, and fuel cells in the energy storage system, a multi-physical field coupling model is constructed; Step S1 specifically includes the following steps: S11. Combining the electrochemical reaction and aging mechanism of lithium batteries, a state of charge dynamic equation and an aging model are established to quantify the influence of temperature and charge-discharge current on battery life; Among them, the expression of the state of charge dynamic equation is as follows: ; In the formula, represents the state of charge; represents the charge-discharge current of the lithium battery; represents the battery capacity of the lithium battery; represents time; represents the charge-discharge efficiency of the battery; represents the ambient temperature; represents the reference temperature, and ; represents the temperature influence coefficient; The aging model expression is as follows: ; In the formula, represents the total aging amount of the lithium battery; and respectively represent calendar aging and cycle aging; represents the calendar aging function; represents the average state of charge; represents the th cycle number; represents the total cycle number; represents the cycle aging function, represents the depth of discharge, represents the th cycle ambient temperature; represents the th cycle state of charge; Meanwhile, considering the influence of temperature on the capacitance value of the supercapacitor, an electro-thermal model of the supercapacitor during the charge and discharge process is established: ; In the formula, represents the supercapacitor terminal voltage; represents the initial voltage of the supercapacitor; represents the capacitance value of the supercapacitor; represents the charge and discharge current of the supercapacitor; represents the equivalent series resistance of the supercapacitor, and , represents the initial equivalent series resistance, represents the resistance temperature coefficient; represents the working temperature of the supercapacitor; represents the mass of the supercapacitor; represents the specific heat capacity; Considering the influence of reactant concentration and temperature, and combining the electrochemical reaction of the fuel cell and the energy conversion efficiency of the electrolyzer, a fuel cell - electrolyzer coupling model is established; ; In the formula, represents the output power of the fuel cell; represents the fuel cell efficiency; represents the input power of the electrolyzer; represents the hydrogen consumption; represents the electrolyzer efficiency; represents the low calorific value of hydrogen; and respectively represent the operating temperature and the maximum operating temperature of the fuel cell; S12. Integrate the state of charge dynamic equation with the aging model, the supercapacitor electro-thermal model, and the fuel cell - electrolyzer coupling model into a multi-physics field coupling model: ; In the formula, represents the output power of the lithium battery; represents the output power of the supercapacitor; represents the efficiency of the DC / DC converter; represents the load power; represents the total energy loss, and , represents the DC bus current, represents the total line resistance; represents the average temperature; and represent the masses of the lithium battery and the fuel cell; represents the operating temperature of the lithium battery.

[0015] S2. Based on the multi-physics field coupling model, with the goal of maximizing the comprehensive benefit of the energy storage system throughout its life cycle, construct a multi-objective function including operating cost, energy loss, and component life, and use the cuckoo-particle swarm hybrid algorithm to generate Pareto front solutions to obtain multiple non-dominated solutions, and each non-dominated solution corresponds to a different power distribution scheme; Step S2 specifically includes the following steps: S21. With the goal of maximizing the comprehensive benefit of the energy storage system throughout its life cycle, construct a multi-objective function including operating cost, energy loss, and component life: ; In the formula, represents the fitness value; represents the operating cost, and , and respectively represent the grid electricity price and the grid electricity purchase volume, represents the hydrogen cost, represents the hydrogen consumption; , , and all represent the weight coefficients, represents the component life loss, and , and respectively represent the aging amount and the maximum allowable aging amount of the lithium battery, represents the fuel cell efficiency decay amount, Indicates the initial efficiency of the fuel cell; Indicates the full-cycle cost; Indicates the environmental impact index; Indicates the total energy loss; The construction process of the economic indicators and environmental indicators described in step S21 is as follows: First step, construct the full-stage inventory data: ; ; ; In the formula, Indicates the manufacturing-stage cost; Indicates the operation-stage cost; Indicates the net cost of the recycling stage; Indicates the energy storage component Quality of; Indicates the energy storage component Production cost per unit mass of; Indicates the energy storage component Energy consumption per unit mass of; Indicates the energy storage component Environmental emissions per unit mass of; Indicates the total number of energy storage components; Indicates the Mass of maintenance materials in the Indicates the Unit price of maintenance materials in the Indicates the operation life; Indicates the Grid power purchase volume in the Indicates the Grid electricity price in the Indicates the Hydrogen consumption in the Indicates the Hydrogen cost in the Indicates the total mass of retired energy storage components; Indicates the recycling and treatment cost; Indicates the retired energy storage component Recyclable ratio of; Indicates the retired energy storage component Mass of; Indicates the recycling material reuse income; Indicates the total number of retired energy storage components; Second step, convert the full-stage inventory data into economic indicators and environmental indicators, and the expression of the economic indicators is as follows: ; In the formula, represents the discount rate; The environmental index expression is as follows: ; In the formula, and respectively represent the th type of environmental index value and its index weight, respectively represent carbon footprint, water resource consumption, soil pollution, and noise pollution.

[0016] S22. Use the cuckoo-particle swarm hybrid algorithm to solve the non-dominated solution set of the multi-objective function and generate the Pareto front solution; S221. Initialize the population: Generate particles corresponding to each group of power allocation schemes. In this embodiment, is set, and the position vector of the particle is , the velocity vector is , and the fitness is ; S222. Exploration stage: Introduce a socially learned factor with random fluctuations to simulate the long-distance random search behavior of cuckoos and search for potential solutions: ; In the formula, and respectively represent the updated position of the particle and the current particle position; represents the inertial weight in the exploration stage, and , and respectively represent the maximum and minimum inertias at the exploration tip, , , and respectively represent the maximum number of iterations and the number of iterations. In this embodiment, ; represents the socially learned factor, and ; represents the Levy flight path, and , represents a random variable that follows , represents the standard deviation, , represents the Levy distribution parameter, , is the gamma function, which is used to control the heavy-tailed characteristic of the step size; represents the current global optimal power allocation vector; S223. Development stage: Use the particle swarm optimization algorithm to refine potential solutions through cognitive learning and social learning: ; ; In the formula, and respectively represent the velocity of the updated particle and the current velocity of the particle ; represents the inertia weight in the development stage, and , and respectively represent the maximum and minimum inertia weights in the development stage, , , , and respectively represent the current fitness, population average fitness, and population maximum fitness of the particle ; and respectively represent the cognitive factor and social factor, and ; and both represent uniformly distributed random numbers subject to ; represents the historical optimal position of the particle ; represents the position of the refined particle ; and respectively represent the lower limit and upper limit of the physical constraint; represents the constraint boundary value; S224. Loop and iterate steps S222 and S223 until the fitness value is less than the set value or the maximum number of iterations is reached, and output the global optimal position , , and respectively represent the global optimal lithium battery output power, supercapacitor output power, and fuel cell output power; S23. Generate an optimal power allocation scheme based on the global optimal position output in step S224 and combined with physical constraints : ; ; ; ; In the formula, , and respectively represent the final optimal lithium battery output power, supercapacitor output power, and fuel cell output power; represents the maximum charge and discharge power of the lithium battery; represents the sign function; represents the optimal terminal voltage of the supercapacitor; and respectively represent the maximum and minimum values of the supercapacitor terminal voltage; S24. Screen the non-dominated solutions in the generated optimal power distribution scheme, and use the life cycle cost as the horizontal axis and the environmental impact index as the vertical axis to plot the Pareto front curve for decision-makers to select the balance point according to actual needs; At the same time, adopt a rolling optimization strategy to generate the optimal power distribution scheme for future moments.

[0017] S3. Integrate the physical constraints of the multi-physical field coupling model and deep learning to construct a multi-fidelity physics-informed neural network model, obtain the real-time estimation value and short-term prediction result of the energy storage system state, and optimize the parameters of the multi-objective function described in step S2 based on the real-time estimation value and short-term prediction result; Step S3 specifically includes the following steps: S31. Design the neural network architecture. The neural network architecture adopts a dual-branch fusion architecture of high-fidelity and low-fidelity data. The bottom branch is used to process the high-fidelity physical dynamic parameters output by step S1 , and the upper branch processes the low-fidelity sensor time series data , and realizes the deep fusion of physical prior knowledge and data-driven features through cross-layer connections; In step S31, the multi-fidelity physics-informed neural network model architecture includes an input layer, a hidden layer, and an output layer arranged in sequence. The hidden layer includes a physical constraint layer, a multi-scale feature extraction layer, and a fusion layer. The physical constraint layer adopts a 64-node fully connected layer, and the activation function is Swish; the multi-scale feature extraction layer is used to capture the 15-minute and 1-hour time series fluctuation patterns; the fusion layer fuses the output of the physical constraint layer and the output of the multi-scale feature extraction layer through skip connections, and the activation function is Tanh, realizing the non-linear coupling of physical constraint features and data-driven features.

[0018] S32. Configure the loss function of the multi-fidelity physics-informed neural network model architecture: ; In the formula, represents the comprehensive loss value; , and respectively represent the physical constraint loss function value, the data-driven loss value, and the boundary loss value; and both represent weight coefficients; Among them, ; In the formula, represents the number of data samples of the energy storage component; represents the gradient operator; represents the heat conduction coefficient; represents the energy storage component heat generation rate; represents the energy storage component temperature; represents the density of the lithium battery; represents the specific heat capacity of the lithium battery; represents the energy storage component charge and discharge current; represents the energy storage component internal resistance; represents the energy storage component terminal voltage; ; In the formula, and respectively represent the measured state of charge and the predicted state of charge; and respectively represent the measured state of health and the predicted state of health; and respectively represent the measured supercapacitor terminal voltage and the predicted supercapacitor terminal voltage; and respectively represent the measured fuel cell temperature and the predicted fuel cell temperature; ; In the formula, is the indicator function; S33. Input the normalized features into the trained multi-fidelity physics-informed neural network model , and output ; Among them, represents the high-fidelity physical dynamic parameter, and , represents the dynamic derivative of the lithium battery state of charge, represents the aging rate of the lithium battery, represents the dynamic derivative of the lithium battery heat, represents the dynamic derivative of the supercapacitor voltage, represents the dynamic derivative of the supercapacitor temperature, represents the power conversion rate of the fuel cell, represents the hydrogen consumption rate; represents the low-fidelity sensor timing data, and , represents the lithium battery voltage, represents the lithium battery current, represents the initial state of charge of the lithium battery, represents the ambient humidity; represents the historical timing data, and , , , and respectively represent the mean value of the historical state of charge of the lithium battery, the decay rate of the state of health, the voltage at the terminals of the supercapacitor, and the operating temperature of the fuel cell; and respectively represent the estimated values of the real-time state of charge and the state of health of the lithium battery; S34. Based on the real-time estimated values output in step S33 and the optimal power distribution scheme solved in step S2, a state prediction curve for the next 30 minutes is generated using a rolling prediction strategy, and error correction is performed through a multi-physics field coupling model: ; In the formula, represents the state prediction value at the future time; represents the LSTM prediction function; represents the state prediction value at the current time; represents the optimal power distribution scheme generated in step S2 at the future time; represents the prediction time step; represents the physical model correction term, and ; S35. Feed the real-time estimated values and the prediction results back to the multi-objective function in step S2 to dynamically adjust the optimization parameters.

[0019] In step S35, if , represents the minimum state of charge threshold, restricting ; If , then adjust the weight coefficient according to the following formula : ; In the formula, represents the base-set weight; represents the adjustment factor; Represents the lowest health state threshold; Represents The mean value of the health state of the lithium battery at a moment.

[0020] S4. Substitute the multiple non-dominated solutions generated in step S2 after parameter optimization into the multi-physical field coupling model described in step S1 to verify the operating state of the energy storage component.

[0021] Simulation experiment

[0022] In this simulation experiment, use MATLAB / Simulink software to build a multi-physical field coupling model and use Python to predict the state of the energy storage system.

[0023] Operating condition setting: Load curve: Typical daily load of residents (low load of 100 kW from 0 to 8 o'clock, medium-high load of 250 kW from 8 to 22 o'clock, low load of 80 kW from 22 to 24 o'clock). Ambient temperature: (Periodic fluctuation simulates the diurnal temperature difference). Optimization mode: Low-carbon mode: (Environmental weight is prioritized). Economic mode: (Cost weight is prioritized). Balanced mode: (Default mode).

[0024] Table 1 Energy storage component parameters ;

[0025] Based on the above conditions, use the present invention to generate solutions in the low-carbon mode, economic mode and balanced mode respectively; Among them, the solution in the low-carbon mode: ; The solution in the economic mode: ; The solution in the balanced mode: ; And substitute the solutions in the three modes into the multi-physical field coupling model, run a 24-hour simulation, and obtain the results shown in Table 2.

[0026] Table 2 Comparison of optimization indexes in three modes ;

[0027] As can be seen from Table 2, the fluctuation range of the state of charge of the lithium battery in different modes is: Low-carbon mode: Fluctuation range [55%, 70%] (always within the safe range [20%, 80%]). Economic mode: Fluctuation range [50%, 75%] (deep discharge leads to an increase in the aging amount).

[0028] Supercapacitor voltage: All three modes are stable in [25V, 38V] and do not touch the voltage boundary.

[0029] Fuel cell power: The average power in the low-carbon mode is 175 kW, which is 12.5% lower than that in the economic mode, meeting the low-carbon priority strategy.

[0030] Table 3 State prediction accuracy ;

[0031] As can be seen from Table 3, the physical model correction term significantly reduces the prediction error (the state of charge error drops by 5.3%). And the power distribution schemes of the three modes all satisfy the power balance equation ( , with an error < 2%), and the fuel cell power output is smooth, meeting the stable power supply requirements of the grid-level energy storage system, thus proving the effectiveness of the present invention.

[0032] An energy storage system for implementing an energy storage control method, comprising: A multi-physical field coupling model construction module, configured to construct a multi-physical field coupling model by considering the electro-thermal-mechanical multi-physical field coupling characteristics of lithium batteries, supercapacitors, and fuel cells in the energy storage system; A power distribution scheme generation module, configured to construct a multi-objective function including operating cost, energy loss, and component life based on the multi-physical field coupling model with the goal of maximizing the comprehensive benefit of the entire life cycle of the energy storage system, and use a cuckoo-particle swarm hybrid algorithm to generate Pareto front solutions to obtain multiple non-dominated solutions, and each non-dominated solution corresponds to a different power distribution scheme; A parameter optimization module, configured to fuse the physical constraints of the multi-physical field coupling model with deep learning to construct a multi-fidelity physical information neural network model, obtain real-time estimated values and short-term prediction results of the energy storage system state, and optimize the parameters of the multi-objective function based on the real-time estimated values and short-term prediction results; A feasibility verification module, configured to substitute the multiple non-dominated solutions after parameter optimization into the multi-physical field coupling model respectively to verify the operating state of the energy storage components.

[0033] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions of the present invention or make equivalent substitutions, and these modifications or equivalent substitutions cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A energy storage control method, characterized in that: It includes the following steps: S1. Considering the electro-thermal-mechanical multi-physics field coupling characteristics of lithium batteries, supercapacitors, and fuel cells in the energy storage system, construct a multi-physics field coupling model; S2. Based on the multi-physics field coupling model, with the goal of maximizing the comprehensive benefit of the entire life cycle of the energy storage system, construct a multi-objective function including operating cost, energy loss, and component life, and use the cuckoo-particle swarm hybrid algorithm to generate Pareto front solutions, obtaining multiple non-dominated solutions, and each non-dominated solution corresponds to a different power allocation scheme; S3. Integrate the physical constraints of the multi-physics field coupling model with deep learning to construct a multi-fidelity physics-informed neural network model, obtain the real-time estimation value and short-term prediction result of the energy storage system state, and optimize the parameters of the multi-objective function described in step S2 based on the real-time estimation value and short-term prediction result; S4. Substitute the multiple non-dominated solutions generated in step S2 after parameter optimization into the multi-physics field coupling model described in step S1 respectively to verify the operating state of the energy storage components.

2. The energy storage control method according to claim 1, wherein: Step S1 specifically includes the following steps: S11. Combining the electrochemical reaction and aging mechanism of the lithium battery, establish a state-of-charge dynamic equation and an aging model to quantify the influence of temperature and charge-discharge current on the battery life; Among them, the expression of the state-of-charge dynamic equation is as follows: ； In the formula, represents the state of charge; represents the charge and discharge current of the lithium battery; represents the battery capacity of the lithium battery; represents time; represents the charge and discharge efficiency of the battery; represents the ambient temperature; represents the reference temperature, and ; represents the temperature influence coefficient; The expression of the aging model is as follows: ； Wherein, represents the total aging amount of the lithium battery; and respectively represent calendar aging and cycle aging; represents the calendar aging function; represents the average state of charge; represents the th cycle number; represents the total cycle number; represents the cycle aging function, represents the depth of discharge, represents the th cycle ambient temperature; represents the th cycle state of charge; At the same time, introduce the influence of temperature on the capacitance value of the supercapacitor, and establish an electro-thermal model of the supercapacitor during the charge-discharge process: ； Wherein, represents the terminal voltage of the supercapacitor; represents the initial voltage of the supercapacitor; represents the capacitance value of the supercapacitor; represents the charge and discharge current of the supercapacitor; represents the equivalent series resistance of the supercapacitor, and , represents the initial equivalent series resistance, represents the temperature coefficient of resistance; represents the operating temperature of the supercapacitor; represents the mass of the supercapacitor; represents the specific heat capacity; Considering the influence of reactant concentration and temperature, combine the electrochemical reaction of the fuel cell and the energy conversion efficiency of the electrolyzer to establish a fuel cell-electrolyzer coupling model; ； In the formula, represents the output power of the fuel cell; represents the fuel cell efficiency; represents the input power of the electrolyzer; represents the hydrogen consumption; represents the electrolyzer efficiency; represents the lower heating value of hydrogen; and respectively represent the operating temperature and the maximum operating temperature of the fuel cell; S12. Integrate the state-of-charge dynamic equation, the aging model, the electro-thermal model of the supercapacitor, and the fuel cell-electrolyzer coupling model into a multi-physics field coupling model: ； Wherein, represents the output power of the lithium battery; represents the output power of the supercapacitor; represents the efficiency of the DC / DC converter; represents the load power; represents the total energy loss, and , represents the DC bus current, represents the total line resistance; represents the average temperature; and represent the masses of the lithium battery and the fuel cell; represents the operating temperature of the lithium battery.

3. The energy storage control method according to claim 2, wherein: Step S2 specifically includes the following steps: S21. With the goal of maximizing the comprehensive benefit of the entire life cycle of the energy storage system, construct a multi-objective function including operating cost, energy loss, and component life: ； In the formula, represents the fitness value; represents the operating cost, and , and respectively represent the grid electricity price and the grid electricity purchase volume, represents the hydrogen cost, represents the hydrogen consumption; , , and all represent the weight coefficients, represents the component life loss, and , and respectively represent the lithium battery aging amount and the maximum allowable aging amount, represents the fuel cell efficiency decay amount, represents the initial fuel cell efficiency; represents the full-cycle cost; represents the environmental impact index; represents the total energy loss; S22. Use the cuckoo-particle swarm hybrid algorithm to solve the non-dominated solution set of the multi-objective function and generate Pareto front solutions; S221. Initialize the population: Generate particles for each group of power allocation schemes, and set the position vector of the particles as , the velocity vector as , and the fitness as ; S222. Exploration stage: Introduce a socially learned factor with random fluctuations to simulate the long-distance random search behavior of cuckoos and search for potential solutions: ； wherein, and respectively represent the updated particle position and the current particle position; represents the inertia weight in the exploration stage, and , and respectively represent the maximum and minimum inertia of the exploration tip, , , and respectively represent the maximum number of iterations and the number of iterations; represents the social learning factor, and ; represents the Levy flight path, and , represents a random variable subject to , represents the standard deviation, , represents the Levy distribution parameter, , is the gamma function, which is used to control the heavy-tailed characteristic of the step size; represents the current global optimal power allocation vector; S223. Development stage: Use the particle swarm optimization algorithm to refine potential solutions through cognitive learning and social learning: ； ； In the formula, and respectively represent the velocity of the updated particle and the current velocity of the particle ; represents the inertia weight in the exploration stage, and , and respectively represent the maximum and minimum inertia weights in the exploration stage, , , , and respectively represent the current fitness of the particle , the average fitness of the population, and the maximum fitness of the population; and respectively represent the cognitive factor and the social factor, and ; and both represent random numbers that follow a uniform distribution ; represents the historical optimal position of the particle ; represents the position of the refined particle ; and respectively represent the lower limit and upper limit of the physical constraint; represents the constraint boundary value; Iteratively repeat steps S222 and S223 until the fitness value is less than the set value or the maximum number of iterations is reached , and output the global optimal position , 、 and represent the global optimal lithium battery output power, supercapacitor output power, and fuel cell output power, respectively; S23. Generate an optimal power allocation scheme based on the globally optimal position output in step S224 and combined with physical constraints : ； ； ； ； In the formula, , and respectively represent the final optimal output power of the lithium battery, the output power of the supercapacitor, and the output power of the fuel cell; represents the maximum charge-discharge power of the lithium battery; represents the sign function; represents the optimal terminal voltage of the supercapacitor; and respectively represent the maximum and minimum values of the supercapacitor terminal voltage; S24. Screen the generated optimal power allocation scheme Select the non-dominated solutions among them, and draw the Pareto front curve with the life cycle cost as the horizontal axis and the environmental impact index as the vertical axis, so that decision-makers can select the balance point according to actual needs; At the same time, adopt a rolling optimization strategy to generate the optimal power allocation scheme for future moments.

4. The energy storage control method according to claim 3, wherein: The construction process of the economic index and environmental index described in step S21 is as follows: The first step is to construct the inventory data for the entire stage: ； ； ； Wherein, represents the cost in the manufacturing stage; represents the cost in the operation stage; represents the net cost in the recycling stage; Represents the mass of the energy storage component ; Represents the production cost per unit mass of the energy storage component ; Represents the energy consumption per unit mass of the energy storage component ; Represents the environmental emissions per unit mass of the energy storage component ; Represents the total number of energy storage components Indicates the quality of the maintenance materials for the year; Indicates the unit price of the maintenance materials for the year; Indicates the operating years; Indicates the annual electricity purchase volume of the power grid; Indicates the annual electricity price of the power grid; Indicates the annual hydrogen consumption; Indicates the annual hydrogen cost; Represents the total mass of retired energy storage components; Represents the recycling and treatment cost; Represents the retired energy storage components Recyclable ratio; Represents the retired energy storage components Mass; Represents the revenue from the reuse of recycled materials; Represents the total number of retired energy storage components; The second step is to convert the inventory data for the entire stage into economic indicators and environmental indicators. Among them, the expression of the economic indicator is as follows: ； In the formula, represents the discount rate; The expression of the environmental indicator is as follows: ； In the formula, and respectively represent the category of environmental index values and their index weights, respectively represent carbon footprint, water resource consumption, soil pollution, and noise pollution.

5. A method for energy storage control according to claim 3, characterized in that: Step S3 specifically includes the following steps: S31. Design a neural network architecture. The neural network architecture adopts a high-fidelity and low-fidelity data dual-branch fusion architecture. The underlying branch is used to process the high-fidelity physical dynamic parameters output in step S1 , and the upper branch processes the low-fidelity sensor time-series data , and realizes the deep fusion of physical prior knowledge and data-driven features through cross-layer connections; S32. Configure the loss function of the multi-fidelity physics-informed neural network model architecture: ； In the formula, represents the comprehensive loss value; , and respectively represent the physical constraint loss function value, the data-driven loss value, and the boundary loss value; and both represent weight coefficients; Among them, ； In the formula, represents the number of data samples of the energy storage component; represents the gradient operator; represents the thermal conductivity; represents the energy storage component 's heat generation rate; represents the energy storage component 's temperature; represents the density of the lithium battery; represents the specific heat capacity of the lithium battery; represents the energy storage component 's charge and discharge current; represents the energy storage component 's internal resistance; represents the energy storage component 's terminal voltage; ； Wherein, and respectively represent the measured state of charge and the predicted state of charge; and respectively represent the measured state of health and the predicted state of health; and respectively represent the measured supercapacitor terminal voltage and the predicted supercapacitor terminal voltage; and respectively represent the measured fuel cell temperature and the predicted fuel cell temperature; ； In the formula, is an indicator function; S33. Input the normalized features into the trained multi-fidelity physics-informed neural network model , and output ; Among them, represents the physical dynamic parameters of high fidelity, and , represents the dynamic derivative of the state of charge of the lithium battery, represents the aging rate of the lithium battery, represents the thermal dynamic derivative of the lithium battery, represents the dynamic derivative of the voltage of the supercapacitor, represents the dynamic derivative of the temperature of the supercapacitor, represents the power conversion rate of the fuel cell, represents the hydrogen consumption rate; represents the low-fidelity sensor timing data, and , represents the voltage of the lithium battery, represents the current of the lithium battery, represents the initial state of charge of the lithium battery, represents the environmental humidity; represents the historical timing data, and , , , and respectively represent the mean value of the historical state of charge of the lithium battery, the attenuation rate of the health state, the terminal voltage of the supercapacitor, and the operating temperature of the fuel cell; and respectively represent the estimated values of the real-time state of charge and the health state of the lithium battery; S34. Based on the real-time estimation value output in step S33 and the optimal power allocation scheme solved in step S2, adopt a rolling prediction strategy to generate a state prediction curve for the next 30 minutes and correct the error through the multi-physics field coupling model: ； In the formula, represents the predicted value of the state at a future time; represents the LSTM prediction function; represents the predicted value of the state at the current time; represents the optimal power allocation scheme generated at the future time step S2; represents the prediction time step; represents the physical model correction term, and ; S35. Feed the real-time estimated value and the prediction result back to the multi-objective function in step S2, and dynamically adjust the optimization parameters.

6. A method for energy storage control according to claim 5, characterized in that: In step S31, the multi-fidelity physics-informed neural network model architecture includes an input layer, a hidden layer, and an output layer arranged in sequence. The hidden layer includes a physical constraint layer, a multi-scale feature extraction layer, and a fusion layer. The physical constraint layer adopts a fully connected layer with 64 nodes, and the activation function is Swish. The multi-scale feature extraction layer is used to capture the temporal fluctuation patterns at the 15-minute level and the 1-hour level; the fusion layer fuses the output of the physical constraint layer and the output of the multi-scale feature extraction layer through skip connections, and the activation function is Tanh to achieve the non-linear coupling of physical constraint features and data-driven features.

7. A energy storage control method according to claim 5, characterized in that: In step S35, if , represents the lowest state of charge threshold, restricting ; If , the weight coefficient is adjusted according to the following formula : ； In the formula, represents the basic setting weight; represents the adjustment factor; represents the lowest health state threshold; represents the mean value of the health state of the lithium battery at time 8. An energy storage system for implementing an energy storage control method according to any one of claims 1-7 above, characterized in that: It includes: A multi-physical field coupling model construction module, which is used to consider the electro-thermal-mechanical multi-physical field coupling characteristics of lithium batteries, supercapacitors, and fuel cells in the energy storage system, and construct a multi-physical field coupling model. A power distribution scheme generation module, which is used to construct a multi-objective function including operating cost, energy loss, and component life based on the multi-physical field coupling model with the goal of maximizing the comprehensive benefit of the whole life cycle of the energy storage system, and use the cuckoo-particle swarm hybrid algorithm to generate the Pareto front solutions to obtain multiple non-dominated solutions, and each non-dominated solution corresponds to a different power distribution scheme. A parameter optimization module, which is used to fuse the physical constraints of the multi-physical field coupling model with deep learning, construct a multi-fidelity physics-informed neural network model, obtain the real-time estimated value and short-term prediction result of the energy storage system state, and optimize the parameters of the multi-objective function based on the real-time estimated value and short-term prediction result. A feasibility verification module, which is used to substitute the multiple non-dominated solutions after parameter optimization into the multi-physical field coupling model respectively to verify the operating state of the energy storage components.

Images