Low-carbon optimal dynamic regulation control method for wind energy and hydrogen energy storage system

Abstract

The invention discloses a low-carbon optimal dynamic regulation control method for a wind energy-storage-hydrogen energy system, and aims to solve the problems of low prediction precision, single optimization target, incomplete constraint consideration, lack of dynamic feedback and the like in the control of an existing wind energy-storage-hydrogen energy system. The method comprises the following steps: collecting system multi-dimensional operation data in real time; using an LSTM and XGBoost fusion model to predict the wind power, the power grid load and the carbon emission coefficient in the next 24 hours; constructing a carbon emission and cost minimization dual-objective function, and setting component and power grid constraints; solving a Pareto optimal solution set through an NSGA-II algorithm, and screening optimal parameters to generate a control instruction; after the instruction is executed, actual data are collected and compared with optimal parameters, optimization and adjustment are carried out again when deviation exceeds the limit, and a closed loop is formed. The method improves prediction precision, realizes low-carbon and economic cooperation, guarantees system stability and assists energy transformation.

Description

Technical Field

[0001] This invention belongs to the field of optimization and control technology for wind energy storage hydrogen energy systems, and particularly relates to a low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems. Background Technology

[0002] Against the backdrop of global efforts to advance dual-carbon goals and transform the energy structure, the large-scale application of renewable energy has become a core path to achieving low-carbon energy. Wind energy, as a technologically mature and resource-rich renewable energy source, has seen its development and utilization scale continuously expand. However, wind energy exhibits significant intermittency, volatility, and randomness. For example, wind speed fluctuates dramatically due to factors such as season, weather, and time of day, leading to unstable wind power output. Direct grid connection can impact grid frequency and voltage stability, and even cause difficulties in grid dispatch and frequent wind curtailment, severely restricting the efficient utilization of wind energy resources.

[0003] To mitigate the impact of wind power fluctuations on the power grid, energy storage and hydrogen energy technologies are widely being incorporated into wind energy utilization systems to construct integrated wind-energy-storage-hydrogen energy systems. Energy storage devices can suppress wind power fluctuations through charge-discharge cycles, storing electrical energy when wind power is abundant and releasing it when wind power is scarce. The hydrogen energy subsystem can convert surplus wind power into hydrogen through an electrolyzer for storage, enabling long-term energy storage and cross-scenario applications, further enhancing the system's energy utilization flexibility. However, this integrated system has a complex structure, involving multiple components such as wind turbine generators, energy storage devices, electrolyzers, and hydrogen storage tanks. These components have significantly different operating characteristics and require interaction with the external power grid. Simultaneously, it must balance carbon emission control and economic operation, making traditional control methods insufficient to meet the system's requirements for efficient, low-carbon, and economical operation.

[0004] Traditional control methods often employ single-objective optimization, such as pursuing only the lowest operating cost or the highest wind power absorption rate, without incorporating total carbon emissions into the core optimization objective, thus becoming disconnected from current low-carbon development needs. In the forecasting stage, they frequently rely on single models to predict wind power generation and grid load demand. Due to the limitations of these models, the prediction accuracy is insufficient, especially under extreme weather conditions or sudden load changes, resulting in significant prediction deviations and a lack of reliable basis for subsequent control command formulation. In the optimization solution stage, traditional algorithms are prone to getting trapped in local optima when dealing with multi-objective and multi-constraint problems, making it difficult to generate globally optimal system operating parameters. Furthermore, they lack dynamic feedback adjustment mechanisms. When the actual system operating state deviates from the optimal parameters beyond the allowable range, the control strategy cannot be adjusted in a timely manner, leading to decreased system operating efficiency, excessive carbon emissions, or increased operating costs.

[0005] Furthermore, existing systems lack a systematic approach to parameter integration and intermediate variable definition, failing to fully consider the quantitative calculation of carbon emissions and various costs throughout the system's lifecycle, making it difficult to achieve precise control over system operation. For example, some control methods do not dynamically adjust the electrolyzer's carbon emission coefficient according to the energy structure, leading to deviations in carbon emission calculations; and they neglect energy storage charging and discharging losses or hydrogen storage and transportation costs in cost accounting, resulting in a disconnect between economic optimization objectives and actual operating costs. These problems collectively make it difficult for wind-energy storage-hydrogen energy systems to simultaneously achieve synergistic optimization of decarbonization, economic efficiency, and stability in actual operation, thus hindering the large-scale promotion and application of such systems.

[0006] Therefore, this invention aims to provide a low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems to solve the above-mentioned technical problems. Summary of the Invention

[0007] The purpose of this invention is to provide a low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems, so as to solve the technical problems existing in the background art.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A low-carbon optimal dynamic regulation and control method for wind energy storage hydrogen energy systems includes the following steps: S1: Real-time acquisition of system operation data, including wind turbine generator data, energy storage device data, hydrogen energy subsystem data, real-time electricity price of the power grid, carbon emission coefficient, and load demand data; S2: Construct a multi-objective prediction model, using historical wind speed data, historical power grid load data, and historical carbon emission coefficient data as input features, and output the wind power generation prediction results, power grid load demand prediction results, and carbon emission coefficient prediction results for a specified future time period. S3: Based on the wind power generation forecast, grid load demand forecast, and carbon emission coefficient forecast, construct the low-carbon optimal objective function and set the system operation constraints. S4: Based on the prediction results and combined with the constraints, the optimal solution set that satisfies the dual objectives of low carbon and economy is obtained through operations such as population initialization, crossover, mutation, non-dominated sorting and crowding calculation. The optimal operating parameters are then selected from the set to generate the corresponding dynamic adjustment and control commands. S5: Transmits dynamic adjustment and control commands to the actuators of each component of the wind energy storage hydrogen energy system, and each actuator performs corresponding operations according to the control commands; S6: Real-time acquisition of actual operating status data of each component of the system after execution of operations; transmission of actual operating status data to the comparison and analysis module of the central control unit, comparison with the optimal operating parameters, and calculation of deviation value; if the deviation value is within the allowable error range, the current control command is maintained; if the deviation value exceeds the allowable error range, the solution parameters of the multi-objective optimization algorithm are readjusted, a new dynamic adjustment control command is generated again, and steps S5 and S6 are repeated to realize closed-loop feedback adjustment of the system operating status.

[0009] Preferably, the multi-objective prediction model in step S2 includes an LSTM model and an XGBoost model; The LSTM model uses a 3-layer network architecture, including an input layer, 2 hidden layers and an output layer. Each hidden layer has 64 neurons, the activation function is ReLU, and a Dropout layer is added. The XGBoost model has 100 decision trees, a maximum tree depth of 6, a learning rate of 0.1, and uses root mean square error as the loss function.

[0010] Preferably, the specific process for predicting wind power generation in step S2 is as follows: S21: Obtain at least one year of historical wind speed data, and match it with the historical output power data of wind turbine generators for the corresponding time period. Use the sliding window method to extract the time series features of wind speed, including mean, maximum, fluctuation range and trend slope within each window, to form a historical wind speed time series feature matrix. Filter out non-time series features including date type and seasonal factors, and use one-hot encoding to convert them into numerical features. S22: Input the historical wind speed time series feature matrix into the LSTM model for training, and input the real-time collected wind turbine generator data into the LSTM model to output the preliminary prediction results of wind power generation in the next 24 hours. S23: Using the preliminary prediction results of wind power generation as the basic feature, input the non-time series features into the XGBoost model to output the corrected prediction results of wind power generation. S24: The preliminary prediction results and the revised prediction results of wind power generation are weighted and fused to obtain the final prediction value of wind power generation.

[0011] Preferably, the specific process of constructing the optimal low-carbon objective function based on the wind power generation forecast results, the power grid load demand forecast results, and the carbon emission coefficient forecast results in step S3 is as follows: S31: Integrate the final forecast results of wind power generation, grid load demand, and carbon emission coefficients by time dimension: S32: Obtain the inherent parameters of the wind energy storage hydrogen energy system as the basis for objective function calculation and constraint setting, including cost parameters, technical parameters, and efficiency parameters; S33: Based on the prediction results and system parameters, define the intermediate variables required in the objective function and constraints, and establish the calculation relationship; S34: With the dual objectives of minimizing total carbon emissions and minimizing operating costs throughout the system's entire life cycle, a low-carbon optimal dual objective function is constructed based on the preprocessed prediction results and intermediate variables.

[0012] Preferably, the specific process of step S34 is as follows: S341: Construct the first objective function: Minimize the total carbon emissions of the system. minf 1: The calculation formula is based on the predicted carbon emission factor and actual power consumption: ; in: P grid-buy (t) represents the active power purchased from the grid in time period t, that is, the power purchased from the grid when the system's own power generation is insufficient. P grid-buy (t)= max [ P grid [(t),0], P grid (t) represents the net interaction power between the system and the external power grid at time t, which is a core variable reflecting the direction and magnitude of energy flow between the system and the power grid. P grid When (t)>0, it indicates that the system's own power generation (wind power, energy storage discharge, etc.) is less than the total demand of the load and electrical equipment (such as electrolytic cells), and it needs to purchase electricity from the grid. P grid-buy (t)= P grid (t).

[0013] when P grid When (t)≤0, it indicates that the system's own power generation exceeds the total demand, and the excess electricity will be sold to the grid (or abandoned). In this case, there is no need to purchase electricity from the grid. P grid-buy (t)=0; α (t) is the carbon emission coefficient corresponding to the power grid purchase in time period t, which reflects the carbon emission per unit of electricity generated by the power grid in that time period, and varies with the power grid energy structure and time period; P elec (t) represents the power generation of the system's self-contained power source (such as a diesel generator or fuel cell) during time period t. α (t) represents the carbon emission coefficient corresponding to the power grid's electricity purchase in time period t, reflecting the carbon emission per unit of electricity generated by the power grid in that time period, which varies with the power grid's energy structure and time period.β This is the carbon emission coefficient per unit power of the electrolyzer, in kgCO2 / kWh. If the electricity for the electrolyzer comes from wind power or energy storage, it can be adjusted according to the actual energy structure. It is assumed to be included in the total carbon emissions of the system here. The summation range is the next 24 hours, that is, it covers the carbon emissions of the entire forecast period. S342: Constructing the second objective function: minimizing the system operating cost. minf 2: Operating costs encompass four categories: energy storage charging and discharging losses, electrolyzer energy consumption, hydrogen storage and transportation, and electricity purchase from the grid. All costs are quantified based on forecast results and intermediate variables. ; Energy storage charging and discharging loss costs: P bat−loss (t)× δ Based on real-time calculations of power loss and unit loss cost, P bat−loss (t) is the th t The power loss generated during the charging and discharging process of the time-limited energy storage device, and the energy loss caused by charging and discharging efficiency. δ This is the cost coefficient corresponding to the unit power loss of energy storage, reflecting the economic cost of energy storage losses. Energy consumption cost of electrolytic cell operation: P elec (t)× ϵ The cost is directly linked to the operating power of the electrolytic cell; the higher the power, the higher the cost. P elec (t) represents the operating power of the electrolytic cell in time period t. ϵ This is the energy consumption cost coefficient corresponding to the unit operating power of the electrolytic cell; Hydrogen storage and transportation costs: Q H2 (t)× ζ It is produced only when hydrogen production exceeds consumption, and is directly proportional to the amount stored and transported. Q H2 (t) represents the net increase in hydrogen that needs to be stored or transported in time period t, which is the production minus the consumption. ζ This is the storage and transportation cost coefficient per unit mass of hydrogen. Electricity purchase cost from the grid: P grid-buy (t)×γ(t), calculated based on the predicted power purchase and the corresponding electricity price for the time period. P grid-buy γ(t) represents the active power purchased from the grid in time period t, and γ(t) represents the unit price of electricity purchased from the grid in time period t, which fluctuates with time. S343: Integration of Dual Objective Functions: Integrating two single objective functions into a unified low-carbon optimal objective function, represented in vector form, clearly defining the direction of dual-objective optimization: minF =[ f 1, f 2).

[0014] Preferably, the constraints for system operation in step S3 include: Wind turbine generator output power constraints: P wind−min ≤ P wind−act (t)≤ P wind−rated , t=1,2,...,24; in, P wind−act (t) represents the actual output power of the wind turbine generator during time period t. P wind−min This is the minimum operating power of the wind turbine generator set. When the power output is below this value, the wind turbine will enter a shutdown or standby state to avoid mechanical wear caused by frequent start-stop cycles. P wind−rated This is the rated output power of the wind turbine generator set, which is also the maximum safe operating power. Exceeding this value will cause the wind turbine to overload and trigger the protection mechanism. SOC constraints: SOC boundary constraints: SOC min × C bat ≤ SOC (t)× C bat ≤ SOC max × C bat ; t=1,2,...,24; in, SOC min The lower limit of the safe state of charge of an energy storage device is usually 0.2~0.3, or 20%~30%. When it is below this value, the energy storage enters a protection state and deep discharge is prohibited. SOC max The upper limit of the safe state of charge of an energy storage device is usually 0.8~0.9, that is, 80%~90%. When it exceeds this value, the energy storage enters a protection state and charging is prohibited. C bat The rated capacity of the energy storage device, i.e., the maximum amount of electricity that can be stored, is SOC ( t )for t period The state of charge of the energy storage device is 0 ≤ SOC(t) ≤ 1. SOC(t)× C bat The actual stored electricity of the energy storage device in time period t; SOC dynamic constraints: SOC(t) = SOC(t−1) + [ P bat−ch ( t -1)× η bat - P bat−dis ( t− 1) / η bat ] / C bat ; t=2,3,...,24; Where SOC(t) is the state of charge of the energy storage device at time t, and SOC(t−1) is the state of charge of the energy storage device at time t-1; P bat−ch (t−1) represents the charging power of the energy storage device at time t−1, i.e., the power of the externally input energy storage. P bat−dis ( t -1) represents the discharge power of the energy storage device during the -1st time period, i.e., the power output by the energy storage device. η bat The charge / discharge efficiency of an energy storage device is typically taken as the average of the charge and discharge efficiencies, ranging from 0.85 to 0.95. C bat This refers to the rated capacity of the energy storage device; Charging section: P bat−ch ( t -1)× η bat Indicates the input power during charging. P bat−ch ( t -1) Cannot be 100% converted into stored energy, multiplied by efficiency. η bat The actual effective power stored is obtained; Discharge section: P bat−dis ( t− 1) / η bat This indicates that during discharge, in order to output externally... P bat−dis ( t− 1) The power required for energy storage is higher, which means the energy storage system needs to consume more electricity, divided by the efficiency. η bat To obtain the actual internal power consumed; Based on the previous SOC and charging / discharging power, calculate the current SOC. The initial SOC (1) is set as the current actual SOC of the system. Electrolytic cell operating power constraints: ; in, P elec (t) represents the actual operating power of the electrolytic cell in time period t; P elec (t-1) represents the actual operating power of the electrolytic cell during the (t-1)th time period; P elec−min This is the minimum starting power of the electrolytic cell, i.e., the minimum operating power. Below this value, the electrolysis efficiency will drop sharply, and the cell will usually be shut down. P elec−rated This is the rated operating power of the electrolytic cell, i.e., the maximum safe operating power. Exceeding this value will lead to equipment overload and shortened lifespan. P elec (t)- P elec (t-1)| represents the absolute change in the electrolytic cell's operating power between adjacent time periods; Δ P elec−max This is the maximum allowable power fluctuation range for the electrolyzer, to prevent sudden power changes from causing fluctuations in electrolyte concentration and damage to electrodes; Hydrogen storage tank capacity constraints: Based on hydrogen storage tank capacity parameters, the real-time storage range is limited, while simultaneously considering the dynamic relationship between hydrogen production and storage. Reserve boundary constraints: Q H2−min ≤ Q H2−act (t)≤ Q H2−max , t=1,2,...,24; in, Q H2−min This is the minimum emergency reserve capacity of a hydrogen storage tank (the minimum amount of hydrogen that must be maintained) to cope with sudden hydrogen demand or system failures, and is typically 5% to 10% of the tank's maximum capacity. Q H2−act (t) represents the actual storage capacity of the hydrogen storage tank in time period t, reflecting the current quality of hydrogen stored in the tank; Q H2−max This is the maximum storage capacity limit for the hydrogen storage tank. Exceeding this value will cause the tank to overpressure and trigger the safety protection mechanism. Dynamic constraints on reserves: Q H2−act (t)= Q H2−act (t−1)+ Q H2−prod (t−1)−Q H2−use (t−1), t=2,3,...,24; in, Q H2−act (t) represents the actual storage capacity of the hydrogen storage tank during time period t. Q H2−act (t−1) represents the actual storage capacity of the hydrogen storage tank during the (t−1)th time period. Q H2−prod (t−1) represents the hydrogen production of the electrolyzer during the (t−1)th time period, which is determined by the electrolysis power and electrolysis efficiency. Q H2−use (t−1) The hydrogen consumption of the system during the t−1 time period, including internal use and external supply; Power grid interaction power constraints: Based on the maximum permissible power interaction parameters of the power grid, the range of power purchased and sold is limited, while also considering the power balance of the associated system. - P grid-sell-max ≤ P grid (t)≤ P grid-buy-max , t=1,2,...,24; in P grid (t) represents the power exchange between the power grids during time period t. P grid-buy-max This is the maximum power that the power grid is allowed to purchase from the system, preventing the system from purchasing too much power from the grid and causing grid overload. P grid-sell-max This is the maximum power output allowed by the power grid system, preventing excessive power injection into the grid that could cause voltage and frequency fluctuations. P grid-sell-max This represents the lower limit of the power that can be sold to the grid, i.e., the negative maximum power that can be sold. It indicates that the power sold by the system to the grid cannot exceed [a certain limit]. P grid-sell-max .

[0015] Preferably, the specific process of solving the constructed low-carbon optimal objective function in step S4 is as follows: S41: Define and encode decision variables: Set the operating parameters that the system needs to optimize as decision variables, use real number encoding, and bind the variable dimension with the time granularity to form a single-entity encoding vector: X (t)=[ P wind-ac t(t), P bat-ch (t), P bat-dis (t), P elec (t),P grid [(t)];t=1,2,...,24; P wind−act (t): Actual output power of the fan during time period t; P bat−ch (t): Energy storage charging power during time period t; P bat−dis (t): Energy storage discharge power during time period t; P elec (t): Operating power of the electrolytic cell during time period t; P grid (t): Power grid interaction power during time period t; S42: Set core parameters: Population size: N = 100~200; Maximum number of iterations: G = 50~100; Crossover probability: P c =0.8~0.9; Mutation probability: P m =0.01~0.05; Crowding threshold: C=0.1; S43: Generating the initial population based on decision variable constraints and prediction data: For each decision variable, an initial value is randomly generated within its constraint interval; S44: Perform multi-objective optimization iterative solution.

[0016] Preferably, the specific process of S44 for multi-objective optimization iterative solution is as follows: S441: Randomly select 2 parent individuals from the population. X 1(t), X 2(t), perform crossover operation on each decision variable to generate offspring. X 1'(t), X 2'(t): X 1'(t)=0.5[(1+ β 1) X 1(t)+(1- β 1) X 2(t)]; X 2'(t)=0.5[(1- β 1) X 1(t)+(1+ β1) X 2(t)]; in β 1 represents the cross-distribution index; For each offspring individual, the decision variable X'(t) variates according to the formula: X''(t)=X'(t)+(X max (t)-X min (t))× δ 1; Where X''(t) represents the offspring decision variable value after mutation, X max (t) represents the maximum allowable value of the decision variable in time period t, determined by system constraints such as the rated power of the wind turbine and the maximum charging and discharging power of the energy storage. min (t) represents the minimum allowable value of this decision variable in time period t, determined by system constraints, such as the minimum technical output of the wind turbine, the power corresponding to the lower limit of SOC, etc. δ 1 represents the variation distribution index; S442: For the population after merging parents and offspring, calculate the two objective function values ​​for each individual. f 1. f 2: ; ; Stratification by dominance: If individual A's... f 1. f If 2 is less than or equal to individual B, and at least one objective is strictly smaller, then A dominates B; individuals without domination are classified as level 1, individuals dominated by level 1 are classified as level 2, and so on. For individuals within each layer, according to f 1. f 2. Sort them separately and calculate the objective function distance between each individual and its neighboring individuals: ; in, CD ( i (for individuals) i The level of congestion, f m max 、f m min The first m The maximum and minimum values ​​of each target are set, and the crowding degree of the boundary individuals is set to infinity (preferential retention). f m ( i+ 1) is the first i +1 individual in the first m Function values ​​on each objective function; fm ( i- 1) is the first i- One individual in the first m Function values ​​on each objective function; S443: Employing a hierarchical selection + crowding ranking strategy, N individuals are selected from the merged population to form the next generation population. Select individuals based on the non-dominant sorting hierarchy, first selecting the first level, then the second level, until the number of selected individuals approaches N; If the number of individuals in a certain layer exceeds the remaining quota, they will be selected in descending order of crowding. Repeat the crossover, mutation, sorting, and selection operations until the number of iterations reaches G, and output the final Pareto optimal solution set; S44: Pareto optimal solution set selection: Construct a comprehensive evaluation function, calculate the comprehensive score for each solution, and select the solution with the lowest score as the optimal solution: The formula for the comprehensive evaluation function is as follows: S = w × f 1 ′ + (1− w) × f 2 ′ ; in, S This represents the individual's overall evaluation score; a lower score indicates a better overall performance of the solution under the dual objectives. w The low-carbon target weighting coefficient reflects the system's priority in minimizing carbon emissions, and its value ranges from 0 to 1. w The larger the value, the more likely it is to choose a low-carbon solution. w The smaller the value, the more likely the economical solution will be chosen. To obtain the normalized carbon emission objective function value, the original f Mapping 1 to the [0,1] interval eliminates the influence of dimensions. To obtain the normalized operating cost objective function value, the original f 2 is mapped to the [0,1] interval to eliminate the influence of dimensions; S45: Optimal Operating Parameter Output: Extract the decision variables of the optimal individual to obtain the optimal parameters for each time period over 24 hours, including: Target output power of the wind turbine: P wind−obj (t), t=1,2,...,24; Target charge / discharge power for energy storage: P bat−ch−obj (t), P bat−dis−obj (t); Target operating power of the electrolyzer: P elec−obj (t); Target power exchange in the power grid: P grid−obj (t).

[0017] Preferably, the specific process of generating the corresponding dynamic adjustment control command based on the selected optimal operating parameters in step S4 is as follows: Wind turbine generator control commands: Pitch angle adjustment command: based on target power P wind-obj (t) and current actual power P wind-act The deviation of (t) is used to calculate the pitch angle adjustment Δ. θ ; Energy storage devices: mode switching and power correction based on state of charge / discharge and SOC; SOC protection and power correction: Real-time acquisition of the current SOC value of energy storage, SOC(t), and correction of the target power if it exceeds the safe range; Electrolytic cells: Start-up and shutdown control and stepped power regulation based on power thresholds; Power adjustment command generation: Calculate the deviation between the current actual power and the target power to generate a power adjustment command.

[0018] The beneficial effects of this invention include: 1. Improve prediction accuracy and provide a reliable basis for optimal control: In the prediction stage, a multi-objective prediction model integrating LSTM and XGBoost is adopted to overcome the limitations of insufficient prediction accuracy of traditional single models. For wind power prediction, the time-series features of historical wind speed are first extracted to output preliminary prediction results; then, non-time-series features are combined and corrected by the XGBoost model, and the two prediction results are weighted and fused, which effectively reduces the impact of wind speed fluctuations, seasonal changes and other factors on prediction accuracy, and significantly reduces the prediction error of wind power in the next 24 hours.

[0019] 2. Achieving Dual-Objective Synergistic Optimization, Balancing Low-Carbon and Economic Efficiency: An optimization function is constructed with the dual objectives of minimizing total carbon emissions and operating costs throughout the system's lifecycle. This comprehensively covers the system's carbon emissions and cost composition, overcoming the limitations of single-objective optimization in traditional control methods. In carbon emission calculations, multiple carbon emission sources are considered, including grid power purchases and electrolyzer operating power. The carbon emission coefficient per unit power of the electrolyzer can be dynamically adjusted according to the actual energy structure, ensuring accurate calculation of total carbon emissions. Cost accounting covers four core cost categories: energy storage charging and discharging losses, electrolyzer energy consumption, hydrogen storage and transportation, and grid power purchases, achieving comprehensive cost quantification. The dual-objective function is solved using NSGA-II to generate a Pareto optimal solution set, and the optimal operating parameters are selected through a comprehensive evaluation function, enabling the system to achieve an optimal balance between low-carbon and economic objectives.

[0020] 3. Comprehensive consideration of system constraints to ensure operational stability: This method systematically analyzes the operational constraints of each component of the wind-powered energy storage hydrogen energy system and the external power grid, including wind turbine output power constraints, energy storage SOC constraints, electrolyzer operating power constraints, and hydrogen storage tank capacity constraints. These constraints are designed to closely align with the operating characteristics of each component and the safety requirements of the power grid. During the optimization process, the optimal operating parameters generated are ensured not to exceed the system's safe operating boundaries, effectively preventing system failures caused by component overload, overcharging / over-discharging of energy storage, and hydrogen storage tank overpressure, thus guaranteeing long-term stable system operation.

[0021] 4. Dynamic Closed-Loop Feedback Adjustment for Enhanced Control Adaptability: A closed-loop feedback adjustment mechanism is introduced, overcoming the limitations of traditional one-time control methods. After the system executes control commands, real-time data on the actual operating status of each component is collected and compared with the optimal operating parameters to calculate the deviation. If the deviation exceeds the allowable error range, the solution parameters of the multi-objective optimization algorithm are immediately readjusted, generating new dynamic adjustment control commands, which are then executed and fed back, forming a dynamic cycle of "prediction-optimization-control-feedback-re-optimization." This ensures the system always operates in its optimal state, avoiding carbon emission exceedances, increased costs, or operational instability caused by initial prediction deviations or external disturbances, significantly improving the system's adaptability to complex operating conditions and changes in the external environment. Attached Figure Description

[0022] Fig. 1 This is a schematic flowchart of the low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy system of the present invention.

[0023] Fig. 2 This is a schematic diagram of the process for constructing the optimal low-carbon objective function according to the present invention.

[0024] Fig. 3 This is a schematic diagram of the architecture of the multi-objective prediction model of the present invention. Detailed Implementation

[0025] The following is in conjunction with the appendix Figs. 1 to 3 The present invention will be further described in detail below: Example 1 See appendix Fig. 1 As shown, the low-carbon optimal dynamic regulation and control method for wind energy storage hydrogen energy systems includes the following steps: S1: Real-time acquisition of system operation data, including wind turbine generator data, energy storage device data, hydrogen energy subsystem data, and real-time electricity price, carbon emission coefficient, and load demand data of the power grid. Wind turbine generator data includes real-time wind speed, output power, and pitch angle. Energy storage device data includes state of charge (SOC), charging and discharging power, and charging and discharging efficiency. Hydrogen energy subsystem data includes electrolysis efficiency of the electrolyzer, hydrogen production, and pressure and storage capacity of the hydrogen storage tank.

[0026] S2: Based on system operation data, a multi-objective prediction model is constructed. The multi-objective prediction model uses historical wind speed data, historical grid load data, and historical carbon emission coefficient data as input features. It uses a fusion algorithm of Long Short-Term Memory Network (LSTM) and Gradient Boosting Tree (XGBoost). The LSTM network is specifically designed to perform deep mining on wind speed data with strong time series characteristics to capture the pattern of wind speed changes over time and to make a preliminary prediction of wind power generation based on real-time wind speed. The preliminary wind power generation prediction result is then input into the XGBoost model along with non-time series features (date type, seasonal factors) to correct the preliminary prediction result and improve accuracy. Finally, through weighted fusion, the wind power generation prediction, grid load demand prediction, and carbon emission coefficient prediction are realized for the next 24 hours, respectively, to obtain the wind power generation prediction result, grid load demand prediction result, and carbon emission coefficient prediction result.

[0027] S3: Based on the wind power generation forecast, grid load demand forecast, and carbon emission coefficient forecast, a low-carbon optimal objective function is constructed. The low-carbon optimal objective function has two objectives: minimizing the total carbon emissions and minimizing the operating costs throughout the system's entire life cycle. The operating costs include the charging and discharging loss costs of the energy storage device, the energy consumption costs of the electrolyzer, the hydrogen storage and transportation costs, and the grid purchase costs. At the same time, constraints are set for the system operation, including the output power constraints of the wind turbine generator set, the state of charge constraints of the energy storage device, the operating power constraints of the electrolyzer, the storage capacity constraints of the hydrogen storage tank, and the power interaction constraints of the grid.

[0028] S4: The Non-Dominated Sorting Genetic Algorithm (NSGA-II) is used to solve the constructed low-carbon optimal objective function. Based on the prediction results and combined with the constraints, Pareto optimal solution set that satisfies both low-carbon and economic objectives is obtained through operations such as population initialization, crossover, mutation, non-dominated sorting, and crowding calculation. From the Pareto optimal solution set, the optimal operating parameters are selected, including the target output power of the wind turbine generator, the target charging and discharging power of the energy storage device, the target operating power of the electrolyzer, and the target interaction power with the power grid. According to the selected optimal operating parameters, corresponding dynamic adjustment and control commands are generated, including the pitch angle adjustment command of the wind turbine generator, the charging and discharging switching command and power adjustment command of the energy storage device, and the start-up and shutdown command, power adjustment command, and power control command of the electrolyzer.

[0029] S5: The generated dynamic adjustment and control commands are transmitted to the actuators of each component of the wind energy storage hydrogen energy system. The actuators include the pitch angle adjustment mechanism of the wind turbine generator, the charge and discharge controller of the energy storage device, the power controller of the electrolyzer, and the power controller of the grid interaction interface. Each actuator performs corresponding operations according to the control commands.

[0030] S6: Real-time acquisition of actual operating status data of each component of the system after execution of operations. The actual operating status data includes the actual output power of the wind turbine generator, the actual state of charge and charging / discharging power of the energy storage device, the actual operating power and hydrogen production of the electrolyzer, the actual storage capacity of the hydrogen storage tank, and the actual interaction power with the power grid. The actual operating status data is transmitted to the comparison and analysis module of the central control unit, compared with the optimal operating parameters, and the deviation value (the difference between the actual operating parameters and the optimal operating parameters) is calculated.

[0031] If the deviation is within the allowable error range (the absolute value of the deviation does not exceed 5% of the optimal operating parameters), the current control command is maintained; if the deviation exceeds the allowable error range, the solution parameters of the multi-objective optimization algorithm are readjusted, a new dynamic adjustment control command is generated again, and steps S5 and S6 are repeated to realize the closed-loop feedback adjustment of the system's operating state, ensuring that the wind energy storage hydrogen energy system always operates in a low-carbon optimal state.

[0032] In this embodiment, see Fig. 3 As shown, the multi-objective prediction model in step S2 includes an LSTM model and an XGBoost model; The LSTM model uses a 3-layer network architecture, including an input layer, 2 hidden layers, and an output layer. Each hidden layer has 64 neurons, and the activation function is ReLU. A dropout layer (dropout rate=0.2) is added to prevent overfitting. The XGBoost model has 100 decision trees, a maximum tree depth of 6, a learning rate of 0.1, and uses root mean square error (RMSE) as the loss function. The XGBoost correction model is trained using "LSTM preliminary prediction results + non-time series features" from historical periods as input and the corresponding actual power as the label.

[0033] Example 2 Based on Example 1, the specific process of wind power generation prediction in step S2 is as follows: S21: Obtain at least one year of historical wind speed data, with a time granularity of 15 minutes per data point. Simultaneously, match the historical output power data of wind turbine generators for the corresponding time period for model training labels. Remove invalid data with wind speeds <3m / s (generator not started) or >25m / s (generator shutdown protection). Use the sliding window method with a window length of 24 hours, i.e., 96 15-minute data points. Extract the mean, maximum, fluctuation range, trend slope, and other time-series features of wind speed within each window to form a historical wind speed time-series feature matrix. Filter non-time-series features, including date type (weekday / weekend / holiday) and seasonal factors (spring / summer / autumn / winter), and convert them into numerical features using one-hot encoding. Associate these features with the wind speed time-series feature matrix for later use.

[0034] S22: Input the historical wind speed time series feature matrix into the LSTM model for training, and input the real-time collected wind turbine generator data into the LSTM model to output the preliminary prediction results of wind power generation in the next 24 hours, with a time granularity of 15 minutes / data point and a total of 96 data points.

[0035] S23: Using the preliminary prediction results of wind power generation as the basic feature, input the non-time series features into the XGBoost model to output the corrected prediction results of wind power generation.

[0036] S24: The preliminary wind power generation forecast and the revised wind power generation forecast are weighted and fused to obtain the final wind power generation forecast value, as shown in the following formula: Final wind power generation forecast = (Preliminary wind power generation forecast × 0.6) + (Corrected wind power generation forecast × 0.4); The 96 merged data points are arranged in chronological order to form a wind power generation forecast for the next 24 hours, in kW.

[0037] The process of predicting grid load demand, carbon emission coefficient, and wind power generation in step S2 is obtained in the same way.

[0038] See Fig. 2As shown, the specific process of constructing the optimal low-carbon objective function based on the wind power generation forecast results, grid load demand forecast results, and carbon emission coefficient forecast results in step S3 is as follows: S31: Integrate the final forecast results of wind power generation, grid load demand, and carbon emission coefficients by time dimension: Time alignment: Based on the time granularity of the grid load demand forecast results (1 hour / data point) and carbon emission coefficient forecast results (1 hour / data point), the wind power generation forecast results (15 minutes / data point) are summarized by hour, and the average of 4 data points within each hour is taken as the wind power generation forecast value for that hour, thus forming 3 sets of data sequences for the next 24 hours with a time granularity of 1 hour.

[0039] Wind power generation forecast P wind (t), t=1,2,...,24, unit: kW; Forecast of power grid load demand P load (t), t=1,2,...,24, unit: kW; Carbon emission coefficient prediction α (t), t=1,2,...,24, unit: kgCO2 / kWh; S32: Obtain the inherent parameters of the wind energy storage hydrogen energy system as the basis for calculating the objective function and setting constraints, including: Cost parameters: Unit charge / discharge loss cost of energy storage devices δ (Unit: Yuan / kWh) Unit operating energy consumption cost of electrolyzer ε (Unit: Yuan / kWh) Unit storage and transportation costs of hydrogen ζ (Unit: Yuan / kg), Power grid unit electricity purchase cost γ(t) (varying over time).

[0040] Technical parameters: Rated power of wind turbine generator set P wind-rated (Unit: kW); Minimum technical output P wind-min (Unit: kW) Rated capacity of energy storage device C bat (Unit: kWh), State of Charge (SOC) Safety Limit SOC min (e.g., 20%), safety limit SOC max (e.g., 80%), rated operating power of the electrolytic cell P elec-rated (Unit: kW) Minimum starting power P elec-min(Unit: kW) Maximum storage capacity of hydrogen storage tank Q H2-max (Unit: kg) Emergency Reserves Q H2-min (Unit: kg) Maximum grid-connected power allowed by the power grid P grid-sell-max (Unit: kW) Maximum purchased power P grid-buy-max (Unit: kW)

[0041] Efficiency parameters: Charge and discharge efficiency of energy storage devices η bat (e.g., 90%) Electrolytic cell efficiency η elec (e.g., 75%, which is the amount of electricity consumed to produce 1 kg of hydrogen, in kWh / kg).

[0042] S33: Based on the prediction results and system parameters, define the intermediate variables needed in the objective function and constraints, and establish the calculation relationships: Energy storage device charging and discharging power: (Assume) P bat-ch (t) represents the energy storage charging power during time period t. P bat-ch (t)≥0, unit: kW; P bat-dis (t) represents the energy storage discharge power during time period t. P bat-dis (t)≥0, unit: kW, satisfying P bat-ch (t)× P bat-dis (t)=0, meaning that simultaneous charging and discharging are not allowed at the same time.

[0043] Energy storage charging and discharging power loss: P bat-loss (t)= P bat-ch (t)×(1- η bat )+ P bat-dis (t) / η bat - P bat-dis (t); that is, the charging loss is the difference between the input power and the actual stored power, and the discharging loss is the difference between the stored power and the output power.

[0044] Hydrogen production from the electrolyzer: Q H2-prod (t)= P elec(t)× η elec ;P elec (t) represents the operating power of the electrolyzer during time period t, in kW. Hydrogen production is directly proportional to the electrolysis power.

[0045] Hydrogen storage and transportation volume: Q H2 (t)= Q H2-prod (t)- Q H2-use (t); Q H2-use (t) represents the amount of hydrogen consumed during time period t. If there is no external supply or internal use of hydrogen, it can be set to 0. The unit is kg. That is, the amount of hydrogen that needs to be stored and transported is the difference between the production and consumption. Q H2 (t) represents the amount of hydrogen stored and transported during time period t. Q H2-prod (t) represents the hydrogen production of the electrolyzer during time period t.

[0046] Grid interaction power: P grid (t)= P load (t)+ P elec (t)+ P bat-ch (t)- P wind (t)- P bat-dis (t); If electricity needs to be purchased from the grid during time period t, the purchased power is: P grid (t); if it represents the electricity sold to the grid during time period t, the electricity sold is - P grid (t).

[0047] S34: With the dual objectives of minimizing total carbon emissions and minimizing operating costs throughout the system's entire life cycle, a low-carbon optimal dual objective function is constructed based on the preprocessed prediction results and intermediate variables.

[0048] In this embodiment, the specific process of step S34 is as follows: S341: Construct the first objective function: Minimize the total carbon emissions of the system. minf 1: The total carbon emissions of the system come from the carbon emissions from electricity purchases from the grid and the carbon emissions from the operation of the electrolyzers. The calculation formula is based on the predicted carbon emission coefficient and the actual power consumption. ; in:P grid-buy (t) represents the active power purchased from the grid in time period t, that is, the power purchased from the grid when the system's own power generation is insufficient. P grid-buy (t)= max [ P grid [(t),0], P grid (t) represents the net interaction power between the system and the external power grid at time t, which is a core variable reflecting the direction and magnitude of energy flow between the system and the power grid. P grid When (t)>0, it indicates that the system's own power generation (wind power, energy storage discharge, etc.) is less than the total demand of the load and electrical equipment (such as electrolytic cells), and it needs to purchase electricity from the grid. P grid-buy (t)= P grid (t).

[0049] when P grid When (t)≤0, it indicates that the system's own power generation exceeds the total demand, and the excess electricity will be sold to the grid (or abandoned). In this case, there is no need to purchase electricity from the grid. P grid-buy (t)=0; α (t) is the carbon emission coefficient corresponding to the power grid purchase in time period t, which reflects the carbon emission per unit of electricity generated by the power grid in that time period, and varies with the power grid energy structure and time period; P elec (t) represents the power generation of the system's self-contained power source (such as a diesel generator or fuel cell) during time period t. α (t) represents the carbon emission coefficient corresponding to the power grid's electricity purchase in time period t, reflecting the carbon emission per unit of electricity generated by the power grid in that time period, which varies with the power grid's energy structure and time period. β This is the carbon emission coefficient per unit power of the electrolyzer, in kgCO2 / kWh. If the electricity for the electrolyzer comes from wind power or energy storage, it can be adjusted according to the actual energy structure. It is included in the total carbon emissions of the system by default. The summation range is the next 24 hours, that is, it covers the carbon emissions of the entire forecast period.

[0050] S342: Constructing the second objective function: minimizing the system operating cost. minf 2: Operating costs encompass four categories: energy storage charging and discharging losses, electrolyzer energy consumption, hydrogen storage and transportation, and electricity purchase from the grid. All costs are quantified based on forecast results and intermediate variables. ; Energy storage charging and discharging loss costs: P bat−loss(t)× δ Based on real-time calculations of power loss and unit loss cost, P bat−loss (t) is the th t The power loss generated during the charging and discharging process of the time-limited energy storage device, and the energy loss caused by charging and discharging efficiency. δ This is the cost coefficient corresponding to the unit power loss of energy storage, reflecting the economic cost of energy storage losses. Energy consumption cost of electrolytic cell operation: P elec (t)× ϵ The cost is directly linked to the operating power of the electrolytic cell; the higher the power, the higher the cost. P elec (t) represents the operating power of the electrolytic cell in time period t. ϵ This is the energy consumption cost coefficient corresponding to the unit operating power of the electrolytic cell; Hydrogen storage and transportation costs: Q H2 (t)× ζ It is produced only when hydrogen production exceeds consumption, and is directly proportional to the amount stored and transported. Q H2 (t) represents the net increase in hydrogen that needs to be stored or transported in time period t, which is the production minus the consumption. ζ This is the storage and transportation cost coefficient per unit mass of hydrogen. Electricity purchase cost from the grid: P grid-buy (t)×γ(t), calculated based on the predicted power purchase and the corresponding electricity price for the time period. P grid-buy γ(t) represents the active power purchased from the grid during time period t, and γ(t) represents the unit price of electricity purchased from the grid during time period t, which fluctuates with time.

[0051] S343: Integration of Dual Objective Functions: Integrating two single objective functions into a unified low-carbon optimal objective function, represented in vector form, clearly defining the direction of dual-objective optimization: minF =[ f 1, f 2).

[0052] The constraints for system operation in step S3 include: Wind turbine output power constraints: Based on wind power generation prediction results P wind (t) and the unit's technical parameters limit the actual output power range: P wind−min ≤ P wind−act (t)≤ P wind−rated(t=1,2,...,24); in, P wind−act (t) represents the actual output power of the wind turbine generator during time period t. P wind−min This is the minimum operating power of the wind turbine generator set. When the power output is below this value, the wind turbine will enter a shutdown or standby state to avoid mechanical wear caused by frequent start-stop cycles. P wind−rated This is the rated output power of the wind turbine generator set, i.e., the maximum safe operating power. Exceeding this value will cause the wind turbine to overload and trigger the protection mechanism.

[0053] Energy storage device state of charge (SOC) constraints: based on rated energy storage capacity C bat This is related to the safety threshold, which limits the real-time range of SOC, and also correlates the dynamic relationship between charge / discharge power and SOC: SOC boundary constraints: SOC min × C bat ≤ SOC (t)× C bat ≤ SOC max × C bat ; t=1,2,...,24; in, SOC min The lower limit of the safe state of charge of an energy storage device is usually 0.2~0.3, or 20%~30%. When it is below this value, the energy storage enters a protection state and deep discharge is prohibited. SOC max The upper limit of the safe state of charge of an energy storage device is usually 0.8~0.9, that is, 80%~90%. When it exceeds this value, the energy storage enters a protection state and charging is prohibited. C bat The rated capacity of the energy storage device, i.e., the maximum amount of electricity that can be stored, is SOC ( t )for t The state of charge (SOC) of the time-limited energy storage device is 0 ≤ SOC(t) ≤ 1. SOC (t)× C bat t represents the actual stored electricity of the energy storage device during time period t.

[0054] SOC dynamic constraints: SOC(t) = SOC(t−1) + [ P bat−ch ( t -1)× ηbat - P bat−dis ( t− 1) / η bat ] / C bat ; t=2,3,...,24.

[0055] Where SOC(t) is the state of charge of the energy storage device at time t, and SOC(t−1) is the state of charge of the energy storage device at time t-1; P bat−ch (t−1) represents the charging power of the energy storage device at time t−1, i.e., the power of the externally input energy storage. P bat−dis ( t -1) represents the discharge power of the energy storage device during the -1st time period, i.e., the power output by the energy storage device. η bat The charge / discharge efficiency of an energy storage device is typically taken as the average of the charge and discharge efficiencies, ranging from 0.85 to 0.95. C bat This refers to the rated capacity of the energy storage device; Charging section: P bat−ch ( t -1)× η bat Indicates the input power during charging. P bat−ch ( t -1) Cannot be 100% converted into stored energy, multiplied by efficiency. η bat The actual effective power stored is obtained; Discharge section: P bat−dis ( t− 1) / η bat This indicates that during discharge, in order to output externally... P bat−dis ( t− 1) The power required for energy storage is higher, which means the energy storage system needs to consume more electricity, divided by the efficiency. η bat To obtain the actual internal power consumed.

[0056] Based on the previous SOC and charging / discharging power, the current SOC is calculated to ensure that the change in energy storage capacity conforms to the charging / discharging efficiency law. The initial SOC (1) is set as the current actual SOC of the system (obtained from real-time data).

[0057] Electrolytic cell operating power constraints: .

[0058] in, P elec (t) represents the actual operating power of the electrolytic cell in time period t; P elec (t-1) represents the actual operating power of the electrolytic cell during the (t-1)th time period; P elec−min This is the minimum starting power of the electrolytic cell, i.e., the minimum operating power. Below this value, the electrolysis efficiency will drop sharply, and the cell will usually be shut down. P elec−rated This is the rated operating power of the electrolytic cell, i.e., the maximum safe operating power. Exceeding this value will lead to equipment overload and shortened lifespan. P elec (t)- P elec (t-1)| represents the absolute change in the electrolytic cell's operating power between adjacent time periods; Δ P elec−max This is the maximum allowable power fluctuation range for the electrolyzer, preventing sudden power changes from causing fluctuations in electrolyte concentration and electrode damage.

[0059] Hydrogen storage tank capacity constraints: Based on hydrogen storage tank capacity parameters, the real-time storage range is limited, while simultaneously considering the dynamic relationship between hydrogen production and storage. Reserve boundary constraints: Q H2−min ≤ Q H2−act (t)≤ Q H2−max , t=1,2,...,24; in, Q H2−min This is the minimum emergency reserve capacity of a hydrogen storage tank (the minimum amount of hydrogen that must be maintained) to cope with sudden hydrogen demand or system failures, and is typically 5% to 10% of the tank's maximum capacity. Q H2−act (t) represents the actual storage capacity of the hydrogen storage tank in time period t, reflecting the current quality of hydrogen stored in the tank; Q H2−max This is the maximum storage capacity limit for the hydrogen storage tank. Exceeding this value will cause the tank to overpressure and trigger the safety protection mechanism.

[0060] Dynamic constraints on reserves: Q H2−act (t)= Q H2−act (t−1)+ Q H2−prod (t−1)− Q H2−use (t−1), t=2,3,...,24.

[0061] in, Q H2−act (t) represents the actual storage capacity of the hydrogen storage tank during time period t. Q H2−act (t−1) represents the actual storage capacity of the hydrogen storage tank during the (t−1)th time period. Q H2−prod (t−1) represents the hydrogen production of the electrolyzer during the (t−1)th time period, which is determined by the electrolysis power and electrolysis efficiency. Q H2−use (t−1) The system's hydrogen consumption during time period t−1, including internal consumption and external supply. Based on the previous time period's storage and hydrogen production and sales relationship, calculate the current time period's storage. (Initial time period) Q H2-act (1) Set as the current actual storage of the system, which is obtained from real-time data collection.

[0062] Power grid interaction power constraints: Based on the maximum permissible power interaction parameters of the power grid, the range of power purchased and sold is limited, while also considering the power balance of the associated system. - P grid-sell-max ≤ P grid (t)≤ P grid-buy-max , t=1,2,...,24; in P grid (t) represents the power exchange between the power grids during time period t. P grid-buy-max This is the maximum power that the power grid is allowed to purchase from the system, preventing the system from purchasing too much power from the grid and causing grid overload. P grid-sell-max This is the maximum power output allowed by the power grid system, preventing excessive power injection into the grid that could cause voltage and frequency fluctuations. P grid-sell-max This represents the lower limit of the power that can be sold to the grid, i.e., the negative maximum power that can be sold. It indicates that the power sold by the system to the grid cannot exceed [a certain limit]. P grid-sell-max This constraint relates to the system power balance equation ( P grid (t)= P load (t)+ P elec (t)+ P bat-ch (t)- P wind (t)- P bat-dis (t) linkage ensures that the interaction power is within a safe range.

[0063] Example 3 Based on Example 1 or Example 2, the specific process of solving the constructed low-carbon optimal objective function using the Non-dominated Sorting Genetic Algorithm (NSGA-II) in step S4 is as follows: S41: Define and encode decision variables: Set the operating parameters that the system needs to optimize as decision variables, use real number encoding (for easy continuous power adjustment), and bind the variable dimension with the time granularity (24 hours) to form a single-entity encoding vector: X (t)=[ P wind-ac t(t), P bat-ch (t), P bat-dis (t), P elec (t), P grid [(t)];t=1,2,...,24; P wind−act (t): Actual output power of the wind turbine (kW) during time period t, subject to P wind−min ≤ P wind−act (t)≤ P wind−rated Constraints, and with predicted values P wind (t) Deviation ≤±5%; P bat−ch (t) and P bat−dis (t): Energy storage charging power and discharging power (kW) during time period t, respectively, satisfying: P bat−ch (t)× P bat−dis (t)=0, and 0≤ P bat−ch (t)≤ P bat−ch−max ; P elec (t): Electrolytic cell operating power (kW) during time period t, satisfying... P elec (t)=0 or P elec−min ≤ P elec (t)≤ P elec−rated And power fluctuation ≤ Δ P elec−max ; Pgrid (t): Power grid interaction power (kW) during time period t, positive for power purchase and negative for power sale, influenced by - P grid−sell−max ≤ P grid (t)≤ P grid−buy−max constraint.

[0064] S42: Set NSGA-II core parameters: Population size: N = 100~200; Maximum number of iterations: G = 50~100; Crossover probability: P c =0.8~0.9; Mutation probability: P m =0.01~0.05; Crowding threshold: C=0.1.

[0065] S43: Based on the decision variable constraints and predicted data, an initial population is generated using a "random sampling + constraint verification" method. For each decision variable, an initial value is randomly generated within its constraint interval, such as Pwind-act(t) in [ max ( P wind-min 0.95 P wind (t)), min ( P wind-rated 1.05 P wind (t))] internal sampling; S44: Perform multi-objective optimization iterative solution, the specific process is as follows: S441: Randomly select 2 parent individuals from the population. X 1(t), X 2(t), for each decision variable, generate offspring according to the formula. X 1'(t), X 2'(t): X 1'(t)=0.5[(1+ β 1) X 1(t)+(1- β 1) X 2(t)]; X 2'(t)=0.5[(1- β 1) X 1(t)+(1+ β 1) X2(t)]; in β 1 is the crossover index, set to 20, which controls the similarity between offspring and parents. If the offspring exceeds the constraint boundary, it is truncated to the boundary value. For each offspring individual, the decision variable X'(t) variates according to the formula: X''(t)=X'(t)+(X max (t)-X min (t))× δ 1; Where X''(t) represents the offspring decision variable value after mutation, X max (t) represents the maximum allowable value of the decision variable in time period t, determined by system constraints such as the rated power of the wind turbine and the maximum charging and discharging power of the energy storage. min (t) represents the minimum allowable value of this decision variable in time period t, determined by system constraints, such as the minimum technical output of the wind turbine, the power corresponding to the lower limit of SOC, etc. δ 1 is the variation distribution index, with a value range of [-1, 1].

[0066] S442: For the population after merging parents and offspring (size 2N), calculate the two objective function values ​​for each individual. f 1 (Total carbon emissions) f 2 (Operating Costs): ; ; Stratification by dominance: If individual A's... f 1. f If 2 is less than or equal to individual B, and at least one objective is strictly smaller, then A dominates B; individuals without domination are classified as layer 1 (optimal layer), individuals dominated by layer 1 are classified as layer 2, and so on.

[0067] For individuals within each layer, according to f 1. f 2. Sort them separately and calculate the objective function distance between each individual and its neighboring individuals: ; in, CD ( i (for individuals) i The level of congestion, f m max 、f m min The first m The maximum and minimum values ​​of each target are set, and the crowding degree of the boundary individuals is set to infinity (preferential retention). fm ( i+ 1) is the first i +1 individual in the first m Function values ​​on each objective function; f m ( i- 1) is the first i- One individual in the first m Function values ​​on the objective function.

[0068] S443: Employing a hierarchical selection + crowding ranking strategy, N individuals are selected from the merged population to form the next generation population. Select individuals based on the non-dominant sorting hierarchy, first selecting the first level, then the second level, until the number of selected individuals approaches N; If the number of individuals in a certain layer exceeds the remaining quota, select individuals in descending order of crowding, retaining those with high crowding to ensure that the solution set is evenly distributed in the target space. Repeat the crossover, mutation, sorting, and selection operations until the number of iterations reaches G, and output the final Pareto optimal solution set (the set of individuals in the first layer).

[0069] S44: Pareto optimal solution set selection: Construct a comprehensive evaluation function, calculate the comprehensive score for each solution, and select the solution with the lowest score as the optimal solution: The formula for the comprehensive evaluation function is as follows: S = w × f 1 ′+ (1 −w ) × f 2 ′ ; in, S This represents the individual's overall evaluation score; a lower score indicates a better overall performance of the solution under the dual objectives. w The low-carbon target weighting coefficient reflects the system's priority in minimizing carbon emissions, and its value ranges from 0 to 1. w The larger the value, the more likely it is to choose a low-carbon solution. w The smaller the value, the more likely the economical solution will be chosen. To obtain the normalized carbon emission objective function value, the original f Mapping 1 to the [0,1] interval eliminates the influence of dimensions. To obtain the normalized operating cost objective function value, the original f 2 is mapped to the [0,1] interval to eliminate the influence of dimensions.

[0070] S45: Optimal Operating Parameter Output: Extract the decision variables of the optimal individual to obtain the optimal parameters for each time period over 24 hours, including: Target output power of the wind turbine: P wind−obj (t) (t=1,2,...,24); Target charging power for energy storage: P bat−ch−obj (t); Energy storage target discharge power: P bat−dis−obj (t); Target operating power of the electrolyzer: P elec−obj (t); Target power exchange in the power grid: P grid−obj (t) (including the direction and power of electricity purchase / sale).

[0071] The specific process of generating the corresponding dynamic adjustment and control command based on the selected optimal operating parameters in step S4 is as follows: Wind turbine generator control commands: Pitch angle adjustment command: based on target power P wind-obj (t) and current actual power P wind-act The deviation of (t) is used to calculate the pitch angle adjustment Δ. θ : like P wind−act (t)< P wind−obj (t): Generates a command to reduce the pitch angle, with an adjustment amount Δ θ =-0.5°~-2° (the larger the deviation, the larger the adjustment amount); The command format is: Wind turbine pitch angle adjustment: target pitch angle for time period t. θ = θ current +Δ θ Power Target P= P wind-obj (t), by reducing the pitch angle, wind capture efficiency is improved and output power is increased; like P wind−act (t)> P wind−obj (t); Generates a command to increase the pitch angle, with an adjustment amount Δ θ =+0.5°~-2°, the instruction format is the same as above, by increasing the pitch angle to reduce the wind energy captured and reduce the output power; like P wind−act (t)= P wind−obj (t): Generates a command to maintain the current pitch angle, avoiding frequent adjustments that could lead to mechanical wear.

[0072] Energy storage devices: Mode switching and power correction based on state of charge / discharge and SOC: Charge / discharge mode determination: based on P bat−ch−obj (t) and P bat−dis−obj The value of (t) determines the energy storage operating mode: like P bat−ch−obj (t)>0 (requires charging): Generate a charging mode switching command, formatted as energy storage switching to charging mode, target charging power for time period t. P = P bat-ch-obj (t), and at the same time close the charging circuit switch and disconnect the discharging circuit; like P bat−dis−obj (t)>0 (Discharge Required): Generate a discharge mode switching command, formatted as energy storage switching to discharge mode, target discharge power for time period t. P = P bat-dis-obj (t), and at the same time close the discharge circuit switch and disconnect the charging circuit; like P bat−ch−obj (t)=P bat−dis−obj (t)=0 (no need for charging and discharging): Generate a "standby mode command" with the format of energy storage entering standby mode, maintaining the current SOC, disconnecting the charging and discharging circuit, and only retaining power supply to the control circuit.

[0073] SOC protection and power correction: The current SOC value (SOC(t)) of the energy storage is collected in real time. If it exceeds the safe range, the target power is corrected. When SOC(t)≥SOC max ×0.95 (near full charge): Generate a charging power reduction command, formatted as: Energy Storage Power Adjustment: Correct Charging Power for Time Period t P = P bat-ch-obj (t)×0.8, SOC protection to prevent overcharging; When SOC(t) ≤ SOC min ×1.05 (close to power depletion): Generate a discharge power reduction command, formatted as: Energy storage power adjustment: Correct discharge power P = P during time period t. bat-dis-obj (t)×0.8, SOC protection to prevent over-discharge; When SOC(t) is in [SOC min SOC max Within the specified range: Generate a sustaining power command based on the target power to ensure stable charging and discharging.

[0074] Electrolyzer: Start-up / shutdown control and stepped power regulation based on power threshold: Start / Stop Status Determination: based on P elec-obj The value of (t) determines the operating status of the electrolytic cell: like P elec-obj (t)≥ P elec-min Currently, the system is shut down: a start-up command is generated, formatted as "electrolytic cell start-up, power applied after 5 minutes of preheating," and the execution process is "power-on preheating → electrolyte melting → gradual loading to target power" (preheating is to avoid electrode damage caused by a sudden increase in power at low temperatures). like P elec-obj (t)=0 and currently running: Generate a shutdown command in the format of "electrolyzer shutdown", execute the pressure relief procedure, and the execution flow is "reduce power to 0 → shut down power → depressurize hydrogen storage tank → inert gas protection". like P elec-obj (t) remains at [ P elec-min , P elec-rated Within the specified range and currently running: Generate instructions to maintain operation without requiring state switching.

[0075] Power regulation command generation: Calculate the current actual power P elec-act The deviation Δ(t) from the target power P elec = P elec-obj (t)- P elec-act (t): If |Δ P elec |≤Δ P elec-max (Fluctuations within allowable range): Generate direct adjustment command, formatted as electrolytic cell power adjustment: target power P = Pelec - obj(t) for time period t, one-time adjustment. If |Δ P elec |>Δ P elec-max (Fluctuation Exceeds Upper Limit): Generate a stepped adjustment command, formatted as: Electrolyzer Power Adjustment: Target Power for Time Period t P = P elec-obj (t), adjusted in 3 steps (each step is 1 minute apart, step size Δt). P =Δ P elec / 3, by adjusting in steps, we can avoid sudden changes in electrolyte concentration from affecting electrolysis efficiency.

Claims

1. A low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems, characterized in that, Includes the following steps: S1: Real-time acquisition of system operation data, including wind turbine generator data, energy storage device data, hydrogen energy subsystem data, real-time electricity price of the power grid, carbon emission coefficient, and load demand data; S2: Construct a multi-objective prediction model, using historical wind speed data, historical power grid load data, and historical carbon emission coefficient data as input features, and output the wind power generation prediction results, power grid load demand prediction results, and carbon emission coefficient prediction results for a specified future time period. S3: Based on the wind power generation forecast, grid load demand forecast, and carbon emission coefficient forecast, construct the low-carbon optimal objective function and set the system operation constraints. S4: Based on the prediction results and combined with the constraints, the optimal solution set that satisfies the dual objectives of low carbon and economy is obtained through operations such as population initialization, crossover, mutation, non-dominated sorting and crowding calculation. The optimal operating parameters are then selected from the set to generate the corresponding dynamic adjustment and control commands. S5: Transmits dynamic adjustment and control commands to the actuators of each component of the wind energy storage hydrogen energy system, and each actuator performs corresponding operations according to the control commands; S6: Real-time acquisition of actual operating status data of each component of the system after performing operations; The actual operating status data is transmitted to the comparison and analysis module of the central control unit, compared with the optimal operating parameters, and the deviation value is calculated. If the deviation value is within the allowable error range, the current control command is maintained. If the deviation value exceeds the allowable error range, the solution parameters of the multi-objective optimization algorithm are readjusted, a new dynamic adjustment control command is generated again, and steps S5 and S6 are repeated to realize the closed-loop feedback adjustment of the system operating status.

2. The low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems according to claim 1, characterized in that, The multi-objective prediction model in step S2 includes the LSTM model and the XGBoost model; The LSTM model uses a 3-layer network architecture, including an input layer, 2 hidden layers and an output layer. Each hidden layer has 64 neurons, the activation function is ReLU, and a Dropout layer is added. The XGBoost model has 100 decision trees, a maximum tree depth of 6, a learning rate of 0.1, and uses root mean square error as the loss function.

3. The low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems according to claim 2, characterized in that, The specific process of wind power generation prediction in step S2 is as follows: S21: Obtain at least one year of historical wind speed data, and match it with the historical output power data of wind turbine generators for the corresponding time period. Use the sliding window method to extract the time series features of wind speed, including mean, maximum, fluctuation range and trend slope within each window, to form a historical wind speed time series feature matrix. Filter out non-time series features including date type and seasonal factors, and use one-hot encoding to convert them into numerical features. S22: Input the historical wind speed time series feature matrix into the LSTM model for training, and input the real-time collected wind turbine generator data into the LSTM model to output the preliminary prediction results of wind power generation in the next 24 hours. S23: Using the preliminary prediction results of wind power generation as the basic feature, input the non-time series features into the XGBoost model to output the corrected prediction results of wind power generation. S24: The preliminary prediction results and the revised prediction results of wind power generation are weighted and fused to obtain the final prediction value of wind power generation.

4. The low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems according to claim 3, characterized in that, The specific process of constructing the optimal low-carbon objective function based on the wind power generation forecast, grid load demand forecast, and carbon emission coefficient forecast in step S3 is as follows: S31: Integrate the final forecast results of wind power generation, grid load demand, and carbon emission coefficients by time dimension: S32: Obtain the inherent parameters of the wind energy storage hydrogen energy system as the basis for objective function calculation and constraint setting, including cost parameters, technical parameters, and efficiency parameters; S33: Based on the prediction results and system parameters, define the intermediate variables required in the objective function and constraints, and establish the calculation relationship; S34: With the dual objectives of minimizing total carbon emissions and minimizing operating costs throughout the system's entire life cycle, a low-carbon optimal dual objective function is constructed based on the preprocessed prediction results and intermediate variables.

5. The low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems according to claim 4, characterized in that, The specific process of step S34 is as follows: S341: Construct the first objective function: Minimize the total carbon emissions of the system. minf 1: The calculation formula is based on the predicted carbon emission factor and actual power consumption: ； in: P grid-buy (t) represents the active power purchased from the grid in time period t. P grid-buy (t)= max [ P grid [(t),0], P grid (t) represents the net interaction power between the system and the external power grid at time t; P elec (t) represents the power generation of the system's self-contained power source during time period t. α (t) represents the carbon emission coefficient corresponding to the power grid's electricity purchase in time period t, reflecting the carbon emission per unit of electricity generated by the power grid in that time period, which varies with the power grid's energy structure and time period. β The carbon emission coefficient per unit power of the electrolytic cell; S342: Constructing the second objective function: minimizing the system operating cost. minf 2: Operating costs encompass four categories: energy storage charging and discharging losses, electrolyzer energy consumption, hydrogen storage and transportation, and electricity purchase from the grid. All costs are quantified based on forecast results and intermediate variables. ； Energy storage charging and discharging loss costs: P bat−loss (t)× δ Based on real-time calculations of power loss and unit loss cost, P bat−loss (t) is the th t Power loss generated during the charging and discharging process of a time-limited energy storage device. δ This is the cost coefficient corresponding to the unit power loss of energy storage. Energy consumption cost of electrolytic cell operation: P elec (t)× ϵ , P elec (t) represents the operating power of the electrolytic cell in time period t. ϵ This is the energy consumption cost coefficient corresponding to the unit operating power of the electrolytic cell; Hydrogen storage and transportation costs: Q H2 (t)× ζ , Q H2 (t) represents the net increase in hydrogen that needs to be stored or transported during time period t. ζ This is the storage and transportation cost coefficient per unit mass of hydrogen. Electricity purchase cost from the grid: P grid-buy (t)×γ(t), P grid-buy γ(t) represents the active power purchased from the grid in time period t, and γ(t) represents the unit price of electricity purchased from the grid in time period t. S343: Integration of Dual Objective Functions: Integrating two single objective functions into a unified low-carbon optimal objective function, represented in vector form, clearly defining the direction of dual-objective optimization: minF =[ f 1, f 2).

6. The low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems according to claim 4, characterized in that, The constraints for system operation in step S3 include: Wind turbine generator output power constraints: P wind−min ≤ P wind−act (t)≤ P wind−rated ,t=1,2,...,24; in, P wind−act (t) represents the actual output power of the wind turbine generator during time period t. P wind−min This is the minimum operating power of the wind turbine generator set. P wind−rated This refers to the rated output power of the wind turbine generator set. SOC constraints: SOC boundary constraints: SOC min × C bat ≤ SOC (t)× C bat ≤ SOC max × C bat ;t=1,2,...,24; in, SOC min This is the lower limit of the safe state of charge for energy storage devices; SOC max This represents the safe upper limit for the state of charge of an energy storage device. C bat SOC (State of Charge) is the rated capacity of the energy storage device. t )for t The state of charge (SOC) of the time-limited energy storage device is 0 ≤ SOC(t) ≤ 1. SOC (t)× C bat The actual stored electricity of the energy storage device in time period t; SOC dynamic constraints: SOC(t)= SOC(t−1)+[ P bat−ch ( t −1)× η bat - P bat−dis ( t− 1) / η bat ] / C bat ; t=2,3,...,24； Where SOC(t) is the state of charge of the energy storage device at time t, and SOC(t−1) is the state of charge of the energy storage device at time t-1; P bat−ch (t−1) represents the charging power of the energy storage device at time t−1; P bat−dis ( t (−1) represents the discharge power of the energy storage device during the (t−1)th time period; η bat The charging and discharging efficiency of energy storage devices; C bat This refers to the rated capacity of the energy storage device; Based on the previous SOC and charging / discharging power, calculate the current SOC. The initial SOC (1) is set as the current actual SOC of the system. Electrolytic cell operating power constraints: ； in, P elec (t) represents the actual operating power of the electrolytic cell in time period t; P elec (t-1) represents the actual operating power of the electrolytic cell during the (t-1)th time period; P elec−min This is the minimum starting power of the electrolytic cell; P elec−rated The rated operating power of the electrolytic cell; P elec (t)- P elec (t-1)| represents the absolute change in the electrolytic cell's operating power between adjacent time periods; Δ P elec−max This refers to the maximum allowable power fluctuation range of the electrolytic cell; Hydrogen storage tank capacity constraints: Based on hydrogen storage tank capacity parameters, the real-time storage range is limited, while simultaneously considering the dynamic relationship between hydrogen production and storage. Reserve boundary constraints: Q H2−min ≤ Q H2−act (t)≤ Q H2−max , t=1,2,...,24; in, Q H2−min This is the lower limit for the emergency reserve capacity of hydrogen storage tanks; Q H2−act (t) represents the actual storage capacity of the hydrogen storage tank during time period t; Q H2−max This represents the maximum storage capacity limit of the hydrogen storage tank. Dynamic constraints on reserves: Q H2−act (t)= Q H2−act (t−1)+ Q H2−prod (t−1)− Q H2−use (t−1), t=2,3,...,24; in, Q H2−act (t) represents the actual storage capacity of the hydrogen storage tank during time period t. Q H2−act (t−1) represents the actual storage capacity of the hydrogen storage tank during the (t−1)th time period. Q H2−prod (t−1) represents the hydrogen production of the electrolyzer during the (t−1)th time period. Q H2−use (t−1) The hydrogen consumption of the system during the t−1 time period, including internal use and external supply; Power grid interaction power constraints: Based on the maximum permissible power interaction parameters of the power grid, the range of power purchased and sold is limited, while also considering the power balance of the associated system. - P grid-sell-max ≤ P grid (t)≤ P grid-buy-max ,t=1,2,...,24; in, P grid (t) represents the power exchange between the power grids during time period t. P grid-buy-max The maximum power purchase capacity allowed by the power grid system. P grid-sell-max The maximum electricity sales capacity allowed by the power grid system, - P grid-sell-max This represents the lower limit of the power that can be sold to the grid, i.e., the negative maximum power that can be sold. It indicates that the power sold by the system to the grid cannot exceed [a certain limit]. P grid-sell-max .

7. The low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems according to claim 5, characterized in that, The specific process of solving the constructed low-carbon optimal objective function in step S4 is as follows: S41: Define and encode decision variables: Set the operating parameters that the system needs to optimize as decision variables, use real number encoding, and bind the variable dimension with the time granularity to form a single-entity encoding vector: X (t)=[ P wind-ac t(t), P bat-ch (t), P bat-dis (t), P elec (t), P grid (t)];t=1,2,...,24; P wind−act (t): Actual output power of the fan during time period t; P bat−ch (t): Energy storage charging power during time period t; P bat−dis (t): Energy storage discharge power during time period t; P elec (t): Operating power of the electrolytic cell during time period t; P grid (t): Power grid interaction power during time period t; S42: Set core parameters: Population size: N = 100~200; Maximum number of iterations: G = 50~100; Crossover probability: P c =0.8~0.9; Mutation probability: P m =0.01~0.05; Crowding threshold: C=0.1; S43: Generating the initial population based on decision variable constraints and prediction data: For each decision variable, an initial value is randomly generated within its constraint interval; S44: Perform multi-objective optimization iterative solution.

8. The low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems according to claim 7, characterized in that, The specific process of multi-objective optimization iterative solution for S44 is as follows: S441: Randomly select 2 parent individuals from the population. X 1(t), X 2(t), perform crossover operation on each decision variable to generate offspring. X 1'(t), X 2'(t): X 1'(t)=0.5[(1+ β 1) X 1(t)+(1- β 1) X 2(t)]; X 2'(t)=0.5[(1- β 1) X 1(t)+(1+ β 1) X 2(t)]; in β 1 represents the cross-distribution index; For each offspring individual, the decision variable X'(t) variates according to the formula: X''(t)=X'(t)+(X max (t)-X min (t))× δ 1; Where X''(t) represents the offspring decision variable value after mutation, X max (t) represents the maximum allowable value of the decision variable in time period t, X min (t) represents the minimum allowable value of the decision variable in time period t. δ 1 represents the variation distribution index; S442: For the population after merging parents and offspring, calculate the two objective function values ​​for each individual. f 1. f 2: ； ； Stratification by dominance: If individual A's... f 1. f If 2 is less than or equal to individual B, and at least one objective is strictly smaller, then A dominates B; individuals without domination are classified as level 1, individuals dominated by level 1 are classified as level 2, and so on. For individuals within each layer, according to f 1. f 2. Sort them separately and calculate the objective function distance between each individual and its neighboring individuals: ； in, CD ( i (for individuals) i The level of congestion, f m max 、f m min The first m The maximum and minimum values ​​of each objective are set, and the crowding degree of boundary individuals is set to infinity. f m ( i+ 1) is the first i +1 individual in the first m Function values ​​on each objective function; f m ( i- 1) is the first i- One individual in the first m Function values ​​on each objective function; S443: Employing a hierarchical selection + crowding ranking strategy, N individuals are selected from the merged population to form the next generation population. Select individuals based on the non-dominant sorting hierarchy, first selecting the first level, then the second level, until the number of selected individuals approaches N; If the number of individuals in a certain layer exceeds the remaining quota, they will be selected in descending order of crowding. Repeat the crossover, mutation, sorting, and selection operations until the number of iterations reaches G, and output the final Pareto optimal solution set; S44: Pareto optimal solution set selection: Construct a comprehensive evaluation function, calculate the comprehensive score for each solution, and select the solution with the lowest score as the optimal solution: The formula for the comprehensive evaluation function is as follows: S=w×f 1 ′+(1−w)× f 2 ′ ; in, S This is the individual's overall evaluation score. w The weighting coefficients for low-carbon targets; This represents the normalized carbon emission objective function value. This represents the normalized operating cost objective function value. S45: Optimal Operating Parameter Output: Extract the decision variables of the optimal individual to obtain the optimal parameters for each time period over 24 hours, including: Target output power of the wind turbine: P wind−obj (t), t=1,2,...,24; Target charging power for energy storage: P bat−ch−obj (t); Energy storage target discharge power: P bat−dis−obj (t); Target operating power of the electrolyzer: P elec−obj (t); Target power exchange in the power grid: P grid−obj (t).

9. The low-carbon optimal dynamic adjustment and control method for wind energy storage hydrogen energy systems according to claim 1, characterized in that, The specific process of generating the corresponding dynamic adjustment and control command based on the selected optimal operating parameters in step S4 is as follows: Wind turbine generator control commands: Pitch angle adjustment command: based on target power P wind-obj (t) and current actual power P wind-act The deviation of (t) is used to calculate the pitch angle adjustment Δ. θ ; Energy storage devices: mode switching and power correction based on state of charge / discharge and SOC; SOC protection and power correction: Real-time acquisition of the current SOC value of energy storage, SOC(t), and correction of the target power if it exceeds the safe range; Electrolytic cells: Start-up and shutdown control and stepped power regulation based on power thresholds; Power adjustment command generation: Calculate the deviation between the current actual power and the target power to generate a power adjustment command.

Images