Capacity configuration method and device of wind-solar-hydrogen-ammonia-alcohol, equipment and medium

Abstract

The application discloses a capacity configuration method and device of wind-solar hydrogen storage ammonia alcohol, equipment and medium, relates to the new energy technology field, and includes: obtaining multi-dimensional input data containing historical meteorological data, system operation and supply and demand state data, wind-solar power and grid-connected regulation benchmark data;The multi-dimensional input data is input into the pre-training capacity configuration model, so that the multi-dimensional input data is predicted and analyzed by the pre-training capacity configuration model, and the full-cycle capacity configuration scheme of the wind-solar hydrogen storage ammonia alcohol system is output;Wherein, the pre-training capacity configuration model is a capacity configuration and path regulation structure of second-hour-season three time domain cooperation, and through the output full-cycle capacity configuration scheme, the efficient cooperative control of electrolytic cell power smooth regulation, synthesis path intelligent switching and seasonal resource allocation is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of new energy technology, in particular to a capacity configuration method, device and equipment of wind-solar-hydrogen-ammonia-alcohol and a medium. BACKGROUND

[0002] Wind-solar-hydrogen-alcohol integrated technology is the core development direction in the field of new energy in recent years, which combines wind power, photovoltaic power generation, energy storage system, green hydrogen production and green methanol synthesis technology, aiming to realize efficient consumption of renewable energy and low-carbon chemical production.

[0003] Most current researches tend to use a single heuristic optimization algorithm, such as particle swarm optimization algorithm, grey wolf algorithm, etc., to realize the capacity configuration of wind-solar-hydrogen-alcohol system. However, such algorithms are generally highly sensitive to parameter settings, leading to complex parameter optimization process, easy to fall into local optimal solution, and lack of systematic verification of algorithm convergence and global optimality, thus it is difficult to guarantee the consistency and robustness of the results in different application scenarios.

[0004] Although some researches have tried to combine capacity configuration with system operation scheduling, the optimization model still stays in the "typical day" or monthly time scale, which is difficult to meet the demand of second or hour level dynamic scheduling in actual engineering. This makes the system still face problems such as insufficient adjustment capacity, large grid connection impact, and frequent start-stop of hydrogen storage equipment when dealing with sudden fluctuations of renewable energy output, which restricts the stability and economy of overall operation.

[0005] In addition, in terms of objective function setting, most researches mainly focus on maximizing economic benefits or minimizing annual investment / operation cost, often ignoring the comprehensive trade-off of multi-dimensional evaluation indexes such as greenhouse gas emission reduction, environmental impact and social benefits. At the same time, due to the fact that the large-scale construction of wind-solar-hydrogen-alcohol integrated projects is still in its infancy, combined with the current price fluctuations of hydrogen, methanol and carbon trading market and other uncertain factors, there is still a lack of complete life cycle cost and environmental impact data support, which is easy to fall into the dilemma of distorted research and evaluation results. SUMMARY

[0006] Therefore, the purpose of the present application is to provide a capacity configuration method, device and equipment of wind-solar-hydrogen-ammonia-alcohol, which can realize efficient collaborative control of electrolyzer power smooth regulation, intelligent switching of synthesis path and seasonal resource allocation, and further realize long-term stable operation and economic improvement of multi-energy coupling system. The specific scheme is as follows:

[0007] In a first aspect, the present application discloses a capacity configuration method of wind-solar-hydrogen-ammonia-alcohol, comprising:

[0008] obtaining multi-dimensional input data including historical meteorological data, system operation and demand and supply state data, wind-solar power and grid connection regulation benchmark data;

[0009] predicting, by a target hydrogen demand prediction sub-model of the pre-trained capacity configuration model and based on the multi-dimensional input data, a daily hydrogen use demand value and a hydrogen storage tank capacity parameter in a future target seasonal time scale;

[0010] predicting, by a target hydrogen resource synthesis path priority prediction sub-model of the pre-trained capacity configuration model and based on the daily hydrogen use demand value and the system operation and supply-demand state data, a hydrogen resource synthesis path priority and a corresponding device start-stop control signal in each hour;

[0011] estimating, by a target electrolytic cell regulation sub-model of the pre-trained capacity configuration model and based on the hydrogen resource synthesis path priority, the device start-stop control signal, and the wind-solar power and grid-connected regulation benchmark data, a wind-solar second-level fluctuation amount, and outputting an electrolytic cell second-level power adjustment instruction;

[0012] outputting, by the pre-trained capacity configuration model and based on the hydrogen storage tank capacity parameter, the hydrogen resource synthesis path priority, the device start-stop control signal, and the electrolytic cell second-level power adjustment instruction, a full-cycle capacity configuration scheme of the wind-solar-hydrogen-ammonia-alcohol system.

[0013] Optionally, the predicting, by a target hydrogen demand prediction sub-model of the pre-trained capacity configuration model and based on the multi-dimensional input data, a daily hydrogen use demand value and a hydrogen storage tank capacity parameter in a future target seasonal time scale, comprises:

[0014] performing weighted processing on the historical meteorological data by an attention mechanism and an LSTM network of the target hydrogen demand prediction sub-model of the pre-trained capacity configuration model, to predict a daily hydrogen use demand value in a future target seasonal time scale;

[0015] calculating a hydrogen storage tank capacity parameter according to a preset hydrogen storage constraint condition based on the daily hydrogen use demand value.

[0016] Optionally, the predicting, by the target hydrogen resource synthesis path priority prediction sub-model, a hydrogen resource synthesis path priority and a corresponding device start-stop control signal in each hour, comprises:

[0017] converting the daily hydrogen use demand value into an hourly demand boundary amount, constructing the hourly demand boundary amount and the system operation and supply-demand state data as state variables, and constructing a path action space of hydrogen and ammonia synthesis paths, hydrogen and alcohol synthesis paths, and single hydrogen production paths;

[0018] selecting a corresponding target action from the path action space based on a greedy exploration strategy and a maximum reward value in a reward value table.

[0019] According to the target action selected by the double-network soft update mechanism and the priority experience replay strategy, the hydrogen resource synthesis path priority and the corresponding equipment start-stop control signal under the optimized target action per hour are obtained.

[0020] Optionally, the target electrolytic cell regulation sub-model is used to estimate the wind-solar second-level fluctuation, and an electrolytic cell second-level power regulation instruction is output, including:

[0021] Based on the wind-solar power and grid-connected regulation reference data, the error between the actual wind-solar power measurement value and the wind-solar power prediction value is calculated, so that the disturbance observer estimates the wind-solar second-level fluctuation based on the error;

[0022] Based on the wind-solar second-level fluctuation, the electrolytic cell power value that minimizes the cost function of the target electrolytic cell regulation sub-model is solved, and an electrolytic cell second-level power regulation instruction is output.

[0023] Optionally, before the multi-dimensional input data is input into the pre-trained capacity configuration model, it further includes:

[0024] Each training data in the training data set is input into an initial capacity configuration model including an initial hydrogen demand prediction sub-model, an initial hydrogen resource synthesis path priority prediction sub-model, and an initial electrolytic cell regulation sub-model connected in sequence.

[0025] The current to-be-trained sub-model is trained by the training data and / or the output result of the previous target sub-model, so that the difference between the current output result of the current to-be-trained sub-model and the corresponding label result in the training data satisfies the corresponding threshold condition, to obtain a target sub-model.

[0026] Each target sub-model is jointly trained to obtain a pre-trained capacity configuration model.

[0027] Optionally, the joint training of each target sub-model to obtain a pre-trained capacity configuration model includes:

[0028] Based on the global optimization engine, the correlation parameters between the target hydrogen demand prediction sub-model, the target hydrogen resource synthesis path priority prediction sub-model, and the target electrolytic cell regulation sub-model are dynamically adjusted, the hydrogen storage tank capacity parameters, the path priority, and the power instruction output by the control model are controlled to cooperatively satisfy the full-cycle capacity configuration target, to obtain a pre-trained capacity configuration model.

[0029] Optionally, the full-cycle capacity configuration scheme of the wind-solar hydrogen storage ammonia alcohol system is output based on the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priority, the equipment start-stop control signal, and the electrolytic cell second-level power regulation instruction, including:

[0030] Based on the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priority, the equipment start-stop control signals, and the electrolyzer second-level power adjustment commands, the operating parameters of each device in the wind-solar-hydrogen-ammonia-methanol system are adjusted to output the corresponding full-cycle capacity configuration scheme under the operating parameters.

[0031] The operating parameters are used to control the overall cost of the wind-solar-hydrogen-ammonia-methanol system to be lower than a preset cost threshold and the resource utilization rate to be higher than a preset utilization rate threshold.

[0032] Secondly, this application discloses a capacity configuration device for wind-solar-hydrogen storage ammonia-ethanol, comprising:

[0033] The data acquisition module is used to acquire multi-dimensional input data, including historical meteorological data, system operation and supply and demand status data, and wind and solar power and grid connection control benchmark data.

[0034] The first prediction module is used to predict the daily hydrogen demand and hydrogen storage tank capacity parameters under the future target seasonal time scale by using a pre-trained capacity configuration model of the target hydrogen demand prediction sub-model and based on the multi-dimensional input data; wherein, the target hydrogen demand prediction sub-model is a seasonal hydrogen demand prediction model.

[0035] The second prediction module is used to predict the priority of hydrogen resource synthesis path prediction sub-model through the target hydrogen resource synthesis path priority prediction sub-model of the pre-trained capacity configuration model, and predict the priority of hydrogen resource synthesis path and the corresponding equipment start-stop control signal for each hour based on the daily hydrogen usage demand value and the system operation and supply and demand status data.

[0036] The third prediction module is used to estimate the second-level fluctuation of wind and solar power based on the target electrolyzer control sub-model of the pre-trained capacity configuration model, the priority of the hydrogen resource synthesis path, the equipment start-up and shutdown control signal, and the wind and solar power and grid connection control benchmark data, and output the second-level power adjustment command of the electrolyzer.

[0037] The scheme generation module is used to output a full-cycle capacity configuration scheme for the wind-solar-hydrogen-ammonia-methanol system based on the pre-trained capacity configuration model and the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priority, the equipment start-stop control signal, and the second-level power adjustment command of the electrolyzer.

[0038] Thirdly, this application discloses an electronic device, including:

[0039] Memory, used to store computer programs;

[0040] A processor is used to execute the computer program to implement the steps of the aforementioned disclosed method for configuring the capacity of wind-solar-hydrogen-ammonia-ethanol.

[0041] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed capacity configuration method for wind-solar-hydrogen-ammonia-ethanol.

[0042] As can be seen, this application discloses a capacity configuration method for wind, solar, hydrogen storage, ammonia, and methanol systems, comprising: acquiring multidimensional input data including historical meteorological data, system operation and supply-demand status data, and wind and solar power and grid-connected regulation benchmark data; predicting the daily hydrogen demand value and hydrogen storage tank capacity parameters under the future target seasonal time scale using a pre-trained capacity configuration model's target hydrogen demand prediction sub-model and based on the multidimensional input data; wherein, the target hydrogen demand prediction sub-model is a seasonal hydrogen demand prediction model; and predicting the daily hydrogen demand value and hydrogen storage tank capacity parameters under the future target seasonal time scale using the pre-trained capacity configuration model's target hydrogen resource synthesis path priority prediction sub-model and based on the daily hydrogen demand value and the system... The system predicts hourly hydrogen resource synthesis path priorities and corresponding equipment start-up and shutdown control signals based on system operation and supply and demand status data; it estimates second-level fluctuations in wind and solar power based on the target electrolyzer control sub-model of the pre-trained capacity configuration model, the hydrogen resource synthesis path priorities, the equipment start-up and shutdown control signals, and the wind and solar power and grid connection control benchmark data, and outputs second-level power adjustment commands for the electrolyzer; it outputs a full-cycle capacity configuration scheme for the wind-solar-hydrogen-ammonia-methanol system based on the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priorities, the equipment start-up and shutdown control signals, and the second-level power adjustment commands for the electrolyzer. This demonstrates that by co-optimizing the three sub-models of the pre-trained capacity configuration model, each sub-model is designed with a dedicated algorithm for different time scales, avoiding the high sensitivity of a single algorithm to parameters. The seasonal sub-model predicts long-term hydrogen demand and hydrogen storage tank capacity, laying the foundation for capacity configuration. The hourly sub-model optimizes hourly path priority and equipment start-up and shutdown, enabling short- to medium-term scheduling. The second-level sub-model estimates wind and solar fluctuations through a disturbance observer and outputs power adjustment commands to cope with high-frequency dynamic changes. By constructing a three-level time scale linkage mechanism of seasonal, hourly, and second-level, seamless connection of the three scales is achieved, breaking through the limitations of daily / monthly, and realizing full-cycle coverage from long-term configuration to real-time control. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0044] Figure 1 This is a flowchart of a capacity configuration method for wind-solar-hydrogen storage ammonia-ethanol disclosed in this application;

[0045] Figure 2 This application discloses a specific method for configuring the capacity of ammonia and alcohol storage hydrogen.

[0046] Figure 3 This is a schematic diagram of the capacity configuration device for a wind-solar-hydrogen storage ammonia-ethanol system disclosed in this application;

[0047] Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0048] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0049] The integrated wind, solar, energy storage, and methanol technology is a core development direction in the new energy field in recent years. It combines wind power, photovoltaic power generation, energy storage systems, green hydrogen preparation, and green methanol synthesis technology, aiming to achieve efficient consumption of renewable energy and low-carbon chemical production.

[0050] Existing research has proposed numerous mathematical models and capacity optimization methods for wind-solar hydrogen production systems. For example, some studies use the improved CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) algorithm to decompose power signals and optimize the capacity configuration of equipment such as electrolyzers and hydrogen storage tanks, taking into account the effects of hydrogen production efficiency and equipment operating temperature, with the goal of minimizing annual investment and operating costs. Other studies construct mixed-integer linear programming models for wind-solar-hydrogen storage-ethanol systems, aiming to maximize total revenue, using MILP (Mixed Integer Linear Programming) to achieve this goal. The mixed-integer linear programming (MILP) algorithm determines the optimal capacity of batteries, hydrogen storage tanks, and electrolyzers. Typical daily analysis verifies the system's energy balance and improved economic efficiency. Another study proposes an improved multi-objective particle swarm optimization algorithm, introducing learning factors and nonlinear inertia weights to optimize the total cost and environmental costs of the wind-solar-hydrogen energy storage system, increasing wind energy integration rate by 30.71% and solar energy by 25.98%. Further research proposes a two-layer optimization strategy: coordinating and optimizing equipment capacity at the outer layer and optimizing operation scheduling at the inner layer, combining medium- and long-term and short-term production simulations to maximize resource utilization value. Additionally, a multi-timescale stochastic production model is constructed, using a seasonal hydrogen storage allocation scheme to mitigate wind and solar power fluctuations, verifying improved system reliability. Furthermore, an integrated "source-grid-load-storage" model is designed and constructed, configuring an intelligent scheduling platform to achieve dynamic capacity configuration where "load follows source." The off-grid wind-solar-hydrogen-alcohol system reduces wind and solar curtailment by adjusting the electrolyzer load to match wind and solar output in real time. All related research must be based on policy benchmark requirements.

[0051] Current research largely favors single heuristic optimization algorithms, such as particle swarm optimization and gray wolf algorithms, for capacity configuration in wind-solar-hydrogen-storage-ethanol systems. However, these algorithms are generally highly sensitive to parameter settings, leading to complex parameter optimization processes, susceptibility to local optima, and a lack of systematic verification of algorithm convergence and global optimality. Consequently, it is difficult to guarantee the consistency and robustness of results across different application scenarios. Although some research has attempted to combine capacity configuration with system operation scheduling, optimization models mostly remain at the "typical daily" or monthly time scale, failing to meet the demands of second-level or hourly dynamic scheduling in practical engineering. This results in the system still facing problems such as insufficient regulation capacity, significant grid connection impact, and frequent start-ups and shutdowns of hydrogen storage equipment when dealing with sudden fluctuations in renewable energy output, thus restricting the overall stability and economic efficiency of operation. Furthermore, in terms of objective function setting, most studies mainly focus on maximizing economic benefits or minimizing annual investment / operating costs, often neglecting the comprehensive trade-off of multi-dimensional evaluation indicators such as greenhouse gas emission reduction, environmental impact, and social benefits. Meanwhile, since the large-scale construction of integrated wind, solar, hydrogen storage, and methanol projects is still in its initial stage, coupled with uncertainties such as the current price fluctuations of hydrogen and methanol and the carbon trading market, there is a lack of complete life cycle cost and environmental impact data to support them, which may easily lead to the predicament of distorted research and evaluation results.

[0052] To this end, the present invention provides a capacity configuration scheme for wind-solar-hydrogen-ammonia-ethanol synthesis, which can achieve efficient and coordinated control of stable electrolyzer power regulation, intelligent switching of synthesis pathways and seasonal resource allocation, and further realize the long-term robust operation and economic improvement of multi-energy coupling system.

[0053] Reference Figure 1 As shown in the figure, an embodiment of the present invention discloses a method for configuring the capacity of wind-solar-hydrogen-ammonia-ethanol storage, comprising:

[0054] Step S11: Obtain multi-dimensional input data including historical meteorological data, system operation and supply and demand status data, wind and solar power and grid connection control benchmark data.

[0055] In this embodiment, wind and solar resource data are acquired, including historical 10-year wind speed data for wind farms and historical 10-year irradiance data for photovoltaic power plants. Historical temperature data for the same period is also acquired to obtain historical meteorological data. In addition, other data besides seasonal data are acquired, specifically including: carbon constraint indicators; hydrogen / ammonia / methanol market demand curves obtained from energy trading platforms or government statistics, which are regional green hydrogen demand quarterly fluctuation curves; and time-series data on synthetic ammonia / methanol market prices. Equipment cost parameters are also acquired, including the unit investment cost of alkaline electrolyzers and the volumetric cost of hydrogen storage tanks. Furthermore, hourly system operation and supply-demand status data are acquired, specifically including system operation status data: hydrogen inventory at time t, energy storage system state of charge, and historical start-up and shutdown status of methanol / ammonia production equipment; supply-demand status data: hydrogen demand forecast at time t, carbon dioxide supply capacity at time t, and nitrogen supply capacity at time t; and market dynamic data: grid-connected electricity price at time t and carbon price at time t. Finally, second-level wind and solar power and grid-connected control benchmark data are obtained, specifically including: wind and solar power data: second-level wind turbine power fluctuation and second-level photovoltaic power fluctuation; grid-connected control benchmark data: grid-connected volatility threshold; and real-time equipment status: real-time monitoring values ​​of electrolytic cell temperature / pressure. It is important to note that the above multi-dimensional input data is not used directly after collection. Instead, it undergoes data cleaning, factor correction, and other data preprocessing steps to transform each input data into target multi-dimensional input data in the target data format. It should be noted that the target data format is determined according to the data input format requirements of the corresponding pre-trained capacity configuration model and is not limited thereto.

[0056] Step S12: Predict the daily hydrogen demand and hydrogen storage tank capacity parameters for the future target seasonal timescale using the target hydrogen demand prediction sub-model of the pre-trained capacity configuration model and based on the multi-dimensional input data; wherein, the target hydrogen demand prediction sub-model is a seasonal hydrogen demand prediction model.

[0057] In this embodiment, the multidimensional input data is input into a pre-trained capacity configuration model. The historical meteorological data is weighted using the attention mechanism and LSTM network of the target hydrogen demand prediction sub-model of the pre-trained capacity configuration model. The input weights corresponding to each historical meteorological data are output, and the daily hydrogen demand and hydrogen storage tank capacity parameters under the future target seasonal time scale are predicted based on the input weights, the historical meteorological data, and the time information of each historical meteorological data. The target hydrogen demand prediction sub-model is a seasonal hydrogen demand prediction model.

[0058] In this embodiment, before inputting the multidimensional input data into the pre-trained capacity configuration model, the method further includes: inputting each training data in the training dataset into the initial capacity configuration model, which includes the initial hydrogen demand prediction sub-model, the initial hydrogen resource synthesis path priority prediction sub-model, and the initial electrolyzer control sub-model connected in sequence.

[0059] If the current sub-model to be trained is the initial hydrogen demand prediction sub-model, then the step of training the current sub-model to be trained using the training data and / or the output of the previous target sub-model, so that the difference between the current output of the current sub-model to be trained and the corresponding label result in the training data satisfies the corresponding threshold condition, to obtain the target sub-model, includes:

[0060] The historical meteorological training data and corresponding hydrogen usage are input into the initial hydrogen demand prediction sub-model. The attention mechanism of the initial hydrogen demand prediction sub-model is used to calculate the training weights of the historical meteorological training data. The training weights, historical meteorological training data, and the time information of the historical meteorological training data are then input into an LSTM (Long Short-Term Memory) network. This allows the LSTM network to learn the correlation between the historical meteorological training data, the time information of the historical meteorological training data, and hydrogen usage. Iterative training is performed until the error between the output historical daily hydrogen usage demand value and the hydrogen usage meets a preset threshold condition, resulting in the trained target hydrogen demand prediction sub-model. It is understood that the purpose of constructing and training a seasonal hydrogen demand prediction model is to predict the hydrogen usage demand in future quarters, so as to reasonably plan the redundancy of hydrogen storage tank capacity and optimize the allocation ratio of hydrogen between alcohol and ammonia pathways, avoiding production scheduling problems caused by a significant decrease in prediction accuracy during seasonal transitions.

[0061] The traditional LSTM model structure is as follows:

[0062] ;

[0063] in, Temperature data; This is wind speed data; For radiation data; Let t be the amount of hydrogen used. This represents the input vector at time t, which is the set of input data received by the LSTM model at that time.

[0064] To address the issues of long-term dependency and information redundancy in assigning dynamic attention weights to different time steps and input variables, an attention mechanism is introduced to calculate the weights of relevant inputs for these variables:

[0065] ;

[0066] ;

[0067] ;

[0068] in, and These are the parameter matrix and bias matrix under the attention mechanism; The projection vector; Let be the attention coefficient of the i-th variable at time t; The symbol represents the element-wise multiplication (Hadamard product) operator, which multiplies corresponding elements of two vectors or matrices of the same dimension. To incorporate the weighted input after the attention mechanism is introduced, an external input is introduced as part of the control input into the hidden layer of the LSTM, and the predicted value is obtained through the LSTM. This value is also the predicted amount of hydrogen. :

[0069] ;

[0070] Meanwhile, to avoid the error from increasing over long-term predictions and the model becoming overly sensitive to certain outliers, a time-weighted mean squared error loss function is introduced. To improve the robustness and generalization ability of hydrogen demand forecasting under extreme weather or abrupt changes, a higher weight is given to the prediction error at future time steps, while a robustness regularization term is added to penalize the model's response to input disturbances. This is achieved by using a specific loss function with weighted mean square error and a robustness penalty term. The weighted mean square error term controls for jumps in the prediction curve, and the robustness regularization term improves robustness to anomalous inputs, as detailed below:

[0071] ;

[0072] in, The actual value at time t. The predicted value at time t; To adjust the dynamic weights based on the predicted situation, ; It is the regular intensity coefficient; This is a collection of relevant parameters. This is a robustness regularization term used to suppress the model's response to input perturbations.

[0073] At the same time, the capacity of the hydrogen storage tank also needs to be considered. Perform planned capacity configuration, and design constraints satisfy the confidence interval. The principles of ≥95%, minimum redundancy ratio, and hydrogen storage upper limit not less than the peak value for N consecutive days are expressed as follows:

[0074] ;

[0075] ;

[0076] in, For demand forecasts over N consecutive days; For the standard deviation of the prediction bias The weighted safety margin is calculated based on the 95% confidence interval. .

[0077] In this way, the data linkage logic of the three-layer model is as follows: the capacity plan output by the seasonal model serves as the boundary constraint condition of the hourly model; the production plan generated by the hourly model serves as the setpoint input for the second-level model; and the operating performance data of the second-level model (such as wind curtailment rate and electrolyzer efficiency) is fed back to the seasonal model to correct long-term forecast parameters.

[0078] The training process for the initial hydrogen demand prediction sub-model is as follows:

[0079] Parameter initialization: LSTM weights can be initialized using Xavier, and attention parameters... and Random initialization was performed to obtain an initial hydrogen demand prediction sub-model;

[0080] Forward propagation: Input each meteorological training sequence in the training dataset, calculate the attention weights (training weights), and input the attention weights and the time information of the meteorological training sequences into the LSTM hidden layer, and output the predicted value through a fully connected layer.

[0081] Loss calculation: The corresponding loss value is calculated by adding the weighted MSE loss and L2 regularization term (λ=0.01) as described above. The loss value is obtained by inputting the predicted value and the true value of the meteorological training sequence into the weighted loss function described above.

[0082] Backpropagation: Based on the loss value and using the Adam optimizer, a learning rate decay strategy (decreasing by 10% every 10 epochs) is used to update the model parameters of the initial hydrogen demand prediction sub-model using gradients. It should be noted that the current hydrogen demand prediction sub-model is used to train and learn the correlation between historical meteorological data and the time-varying (seasonal) changes in hydrogen consumption.

[0083] Early stopping mechanism: If the validation set loss does not decrease for 5 consecutive rounds, training is terminated to prevent overfitting. Once the model terminates training, the trained target hydrogen demand prediction sub-model is obtained.

[0084] Therefore, by training and using the A-LSTM seasonal hydrogen demand forecasting model based on the attention mechanism and meteorological factors, and by using the dynamic attention mechanism to weight meteorological data and combining it with the time-weighted loss function for training, the daily hydrogen demand forecast for the next 90 days can be achieved.

[0085] In this embodiment, if the current sub-model to be trained is the initial hydrogen resource synthesis path priority prediction sub-model, then the training of the current sub-model to be trained using the training data and / or the output of the previous target sub-model, so that the difference between the current output of the current sub-model to be trained and the corresponding label result in the training data meets the corresponding threshold condition, to obtain the target sub-model, can be understood as inputting the historical daily hydrogen demand value output by the target hydrogen demand prediction sub-model, the historical hourly system operation and supply and demand status data of the training data, and the corresponding path decision labels into the initial hydrogen resource synthesis path priority prediction sub-model. Based on the constructed state space and action space, iterative training is performed through reward function evaluation, dual-network soft update, and priority experience replay mechanism until the matching degree between the output path priority and equipment start-stop control signal and the path decision label meets the preset matching degree threshold, to obtain the trained target hydrogen resource synthesis path priority prediction sub-model.

[0086] In this way, on an hourly timescale, based on predicted wind and solar power, hydrogen load, carbon source supply, and market information, intelligent decisions are made regarding the priority of hydrogen use for ammonia or alcohol synthesis and the start-up and shutdown strategies of the equipment, in order to minimize overall costs, maximize resource utilization, and reduce start-up and shutdown frequency. An improved DQN (Deep Q-learning) algorithm is employed, combined with a dual-network structure + priority experience replay + - Greedy exploration combined with multi-path action space modeling of alcohols and ammonia enables intelligent optimization control of paths and start / stop operations. Specifically, the current state variable is:

[0087] ;

[0088] in, The hydrogen inventory at time t; Forecast demand at time t; Let t represent the carbon dioxide supply capacity. Let t represent the nitrogen supply capacity. The state of charge of the energy storage power station at time t; The grid-connected electricity price at time t; Let t be the carbon price. This refers to the start-up and shutdown status of the alcohol and ammonia production equipment.

[0089] A discrete action combination strategy is adopted, and the path switching decision function is used. include:

[0090] ;

[0091] reward function The specific components are as follows:

[0092] ;

[0093] ;

[0094] ;

[0095] in, These are the start / stop flags for hydrogen + ammonia, hydrogen + alcohol, and hydrogen-only production, respectively. For the profits from alcohol and ammonia production; Energy consumption costs; Penalty for device start / stop; This is a penalty for hydrogen inventory overflow or ammonia / methanol production shortfall; These are the corresponding penalty coefficients.

[0096] To assess the future impact of state and path variables, a dual-network approach is innovatively employed. The network system has a structure that allows it to learn path strategy combination decisions under complex conditions. Meanwhile, the network... Instead of using the conventional method of completely replicating the main network every N steps, The parameters are determined using the Polyak method by employing a soft update rate. Update the network:

[0097] ;

[0098] Update the Q-function in the target network DQN reinforcement learning algorithm:

[0099] ;

[0100] in, For the greedy algorithm's exploration rate; This is the algorithm discount factor; calculated... The algorithm evaluates the Q-values ​​of all possible actions in the next state and selects the largest one as the expected future value, representing the agent's anticipated reward under the optimal policy. However, if the algorithm always chooses the action currently considered optimal, it may never discover a better policy and become trapped in a local optimum. Therefore, a greedy exploration rate is introduced. To achieve linear decay:

[0101] ;

[0102] in, Minimum exploration rate; To explore the decay rate; This represents the initial exploration rate.

[0103] In the improved reinforcement learning algorithm, a quadruple empirical variable is generated after each agent decision. :

[0104] ;

[0105] Because the system accumulates a large amount of data during operation, Q-networks also suffer from data redundancy and data pollution, leading to poor learning efficiency and a gradual decline in learning efficiency. The slow convergence and instability become increasingly pronounced. Therefore, another difference between this invention and traditional DQN is the Prioritized Experience Replay (PER) mechanism, which improves policy learning efficiency and convergence stability. In multi-path start-stop control problems, the distribution of experience samples is highly uneven. A few "critical events," such as path switching, high-cost start-stop, or large reward fluctuations, have a significant impact on policy learning but are easily ignored in uniform sampling. The PER mechanism calculates the Temporal-Difference (TD) error for each experience sample. As a priority metric, the probability of sampling is dynamically adjusted, ensuring that high-value, policy-boundary-sensitive, or error-prone experiences are sampled more frequently during training. This not only accelerates the policy's ability to identify high-yield path combinations but also effectively strengthens the model's ability to avoid "incorrect policies" or "unreasonable switching."

[0106] Wherein, the priority function corresponding to the t-th sample data is: This data was used in the improved DQN with a sampling rate of :

[0107] ;

[0108] ;

[0109] Meanwhile, to avoid training bias, the PER mechanism combines importance sampling correction, achieving a balance between experience utilization efficiency and policy robustness while ensuring training stability. In high-dimensional path combination spaces, the PER policy can significantly improve the convergence speed and policy generalization ability of the DQN controller, especially showing obvious advantages in complex energy scenarios (such as wind and solar power with high uncertainty and high path switching costs).

[0110] Specifically, the training process of the initial hydrogen resource synthesis pathway priority prediction sub-model is as follows:

[0111] First, a training environment is constructed. Specifically, training involves building a wind and solar power output scenario generator through a simulation platform to generate 1000 sets of wind / solar / load / price fluctuation scenarios. The constructed hydrogen-ammonia-ethanol path physical model is used as the equipment dynamic response model. Then, an 8-dimensional initial state vector is input into the equipment dynamic response model. The content of the state vector has been described in the previous embodiments and will not be repeated here. Next, the corresponding execution action is selected based on the action space defined in advance through a discrete action set, which includes hydrogen production path, alcohol production path, and pure hydrogen path, and the maximum reward value in the current reward value table. Then, the sub-model parameter optimization process is carried out based on dual network update, PER mechanism, and exponential decay strategy. Finally, training stops when the Q-value volatility is less than the preset volatility threshold for 1000 consecutive steps and the path switching cost meets the preset cost condition. The target hydrogen resource synthesis path priority prediction sub-model is then output.

[0112] In this way, the path switching decision accuracy is improved by prioritizing the learning of high TD error samples through the PER mechanism, the Q value fluctuation is reduced through Polyak soft update, the exploration efficiency is improved through the exponential decay strategy, and the synthesis tower life is extended through the start-stop penalty term.

[0113] In this embodiment, if the current sub-model to be trained is the initial electrolyzer control sub-model, the current sub-model to be trained is trained using the training data and / or the output of the previous target sub-model. The training aims to ensure that the difference between the current output of the current sub-model and the corresponding label result in the training data meets a corresponding threshold condition, thus obtaining the target sub-model. This can be understood as inputting the historical hydrogen resource synthesis path priority and equipment start-up / stop control signals output by the target hydrogen resource synthesis path priority prediction sub-model, the historical second-level wind and solar power and grid-connected control benchmark data from the training data, and the corresponding historical second-level power adjustment commands of the electrolyzer to the initial electrolyzer control sub-model. The historical second-level fluctuations of wind and solar power are estimated using a disturbance observer to iteratively optimize the model parameters until the tracking error between the output electrolyzer power adjustment command and the historical second-level power adjustment command meets a preset threshold condition, thus obtaining the target electrolyzer control sub-model.

[0114] It is understood that this embodiment improves upon the traditional MPC (Model Predictive Control) cost function to apply the improved MPC cost function to the initial electrolyzer control sub-model of this invention. First, the traditional MPC cost function is as follows:

[0115] ;

[0116] in, This refers to the power of the electrolytic cell; This represents the actual grid-connected power of the new energy base; Target grid-connected power; For precision weights; For smoothing weights; To predict the control cycle. By analyzing... The system increases the observations of the disturbance observer and dynamically adjusts the weighting coefficients according to the volatility, thereby improving the robustness and response speed of grid-connected scheduling and avoiding the problems of over-response or delayed response that are prone to occur in traditional second-level control.

[0117] It is important to note that while traditional MPC (Multi-Level Monitoring) possesses good predictive and control capabilities, it suffers from slow response to rapid nonlinear fluctuations in wind and solar resources and struggles to simultaneously balance tracking performance and system robustness. Therefore, a Disturbance Observer (DOB) is embedded into the MPC prediction framework to achieve feedforward compensation for system disturbances. Simultaneously, an adjustable cost function weighting coefficient is introduced to enhance steady-state control during periods of severe fluctuation and improve tracking accuracy during periods of stable fluctuation. Thus, the improved MPC cost function is as follows:

[0118] ;

[0119] ;

[0120] ;

[0121] ;

[0122] in, To increase the grid-connected power of new energy bases after disturbance observation; For dynamic precision weights; For dynamic smoothing weights; For the volatility of power generation in new energy bases; This refers to the sensitivity parameter. The power generation fluctuation estimated by the disturbance observer is based on the error between the current measured wind and solar power and the output of the prediction model. The second-level disturbance component is estimated, and the control quantity is corrected in real time through a feedforward structure, which significantly improves the system's regulation stability and response rate under high-frequency wind and solar power fluctuations. Specifically, this can be manifested as follows:

[0123] ;

[0124] ;

[0125] in, Let t be the expected value of the low-frequency filter output using the moving average method at time t.

[0126] Specifically, the training process of the initial electrolyzer control sub-model is as follows:

[0127] First, real-time data (historical second-level wind and solar power and grid-connected control benchmark data) and the current operating status are input into the initial electrolyzer control sub-model. Feedforward compensation is performed through a disturbance observer, and the result after feedforward compensation is input into the IMPC controller. The IMPC controller can perform corresponding calculations based on the improved MPC algorithm to output the corresponding electrolyzer power adjustment command to the electrolyzer PLC (Programmable Logic Controller). The electrolyzer can then adjust according to the electrolyzer power adjustment command and output a new current operating status. During this training process, the tracking error is judged by comparing the electrolyzer power adjustment command with the historical second-level power adjustment command corresponding to the historical data. When the tracking error meets the error threshold condition, the model parameter optimization process is stopped, and the target electrolyzer control sub-model is obtained.

[0128] In this way, it can be seen that the designed disturbance observation-model prediction joint control strategy runs in the main controller of the electrolysis system, which acquires the wind and solar power output and system status in real time, enhances the prediction accuracy through disturbance compensation, and outputs second-level electrolyzer adjustment commands to the underlying PLC execution system, thereby achieving efficient coordination between feedforward active suppression of fluctuation sources and equipment response.

[0129] In this embodiment, the target sub-models are jointly trained to obtain a pre-trained capacity configuration model. This can be understood as jointly training the target hydrogen demand prediction sub-model, the target hydrogen resource synthesis path priority prediction sub-model, and the target electrolyzer control sub-model to obtain the pre-trained capacity configuration model. Specifically, based on a global optimization engine, the correlation parameters between the target hydrogen demand prediction sub-model, the target hydrogen resource synthesis path priority prediction sub-model, and the target electrolyzer control sub-model are dynamically adjusted to control the hydrogen storage tank capacity parameters, path priorities, and power commands output by the models to collaboratively meet the full-cycle capacity configuration target, thereby obtaining the pre-trained capacity configuration model. Understandably, to improve the generalization and convergence stability of the model, a parameter-self-adjusting PSO-GA-Hybrid hybrid optimizer is proposed. It uses Particle Swarm Optimization (PSO) for global search and Genetic Algorithm (GA) for local search, exchanging superior individuals to avoid premature convergence. Furthermore, it dynamically adjusts the particle swarm inertia weights and crossover probabilities based on the fitness change rate. Finally, the Lyapunov convergence criterion is applied to ensure the algorithm's convergence stability. This optimization engine, acting as the "top-level brain" for system-level coordination and global optimization, serves the following objectives in the three-level linkage mechanism. Table 1 shows the optimization variables and objectives for each level:

[0130] Table 1

[0131]

[0132] All control layer parameters to be optimized are uniformly encoded into multi-dimensional vectors and normalized:

[0133] ;

[0134] in, These are the specific relevant parameters in the second-level, hour-level, and seasonal-level models, respectively.

[0135] The algorithm for finding the global optimum through particle swarm optimization is as follows:

[0136] ;

[0137] ;

[0138] in, Let i be the particle at time t; The dynamic weight at time t; For individual optimal and global optimal values; Individual factors and group factors; A random number between 0 and 1.

[0139] By leveraging the advantages of GA crossover and mutation, local optima are avoided, while excellent individual solutions are preserved. The crossover probability is... ) and mutation (probability of Algorithm logic:

[0140] ;

[0141] ;

[0142] in, The cross ratio; This is a Gaussian noise disturbance.

[0143] To overcome the problem of fixed or manually set parameters in traditional PSO and GA algorithms, a dynamic parameter self-correction mechanism is designed to automatically adjust the search strategy based on population convergence, with dynamic weights. The design is as follows:

[0144] ;

[0145] in, The fitness standard deviation of generation t; This represents the initial fitness standard deviation. Individual and population factors in PSO are also corrected for the rate of change of the fitness mean.

[0146] ;

[0147] in, The mean change rate of fitness in generation t; Sensitivity factor; crossover probability in GA and mutation probability It is also corrected based on the rate of change of the fitness mean:

[0148] ;

[0149] To ensure the stability of the optimization results in practical applications, a convergence criterion is designed using Lyapunov functions to measure whether the population state in the optimization algorithm tends to stabilize / converge, thus facilitating better optimization:

[0150] ;

[0151] in, The tolerance threshold is set to 10. -4 ;like If the value is less than 0 for several consecutive generations, it indicates that the algorithm for that parameter is gradually converging. If... If the value gradually approaches 0, it indicates that the system is gradually falling into a local optimum, and mutation or parameter correction needs to be introduced.

[0152] In this embodiment, the historical meteorological data is weighted using the attention mechanism of the target hydrogen demand prediction sub-model of the pre-trained capacity configuration model and an LSTM network to predict the daily hydrogen demand value at the target seasonal timescale. Based on the daily hydrogen demand value and according to preset hydrogen storage constraints, the hydrogen storage tank capacity parameters are calculated. It can be understood that the attention mechanism of the target hydrogen demand prediction sub-model of the pre-trained capacity configuration model is used to calculate parameter matrices for temperature data, wind speed data, radiation data, and corresponding historical hydrogen usage, to obtain parameter matrices, bias matrices, and projection vectors. Based on the parameter matrix, the bias matrix, and the projection vector... The vector and the corresponding historical meteorological data are calculated to obtain the attention coefficients of each historical meteorological data at different time steps, and input weights are generated. The input weights are input into the LSTM hidden layer of the target hydrogen demand prediction sub-model so that the LSTM hidden layer outputs an initial hydrogen demand value based on the time information of each historical meteorological data, the historical meteorological data, and the input weights. The initial hydrogen demand value is optimized based on the mean square error loss function with time weighting to obtain the daily hydrogen demand value under the future target seasonal time scale. The hydrogen storage tank capacity parameters are calculated according to the daily hydrogen demand value and the preset hydrogen storage constraints.

[0153] In this way, as in the model working mode of the training process of the initial hydrogen demand prediction sub-model in the aforementioned embodiment, after the initial hydrogen demand prediction sub-model is trained and the target hydrogen demand prediction sub-model is obtained, the daily hydrogen usage demand value is directly predicted, and the hydrogen storage tank capacity parameters are further calculated.

[0154] Step S13: Using the target hydrogen resource synthesis path priority prediction sub-model of the pre-trained capacity configuration model, and based on the daily hydrogen usage demand value and the system operation and supply and demand status data, predict the hourly hydrogen resource synthesis path priority and the corresponding equipment start-up and shutdown control signals.

[0155] In this embodiment, the daily hydrogen usage demand value and the hourly system operation and supply and demand status data are input into the target hydrogen resource synthesis path priority prediction sub-model of the pre-trained capacity configuration model, so that the target hydrogen resource synthesis path priority prediction sub-model can predict the hydrogen resource synthesis path priority and the corresponding equipment start-stop control signal at the hourly level.

[0156] In this embodiment, the daily hydrogen demand is converted into hourly demand boundary values. These hourly demand boundary values, along with hourly system operation and supply-demand status data, are constructed as state variables. A path action space is also constructed for hydrogen and ammonia synthesis paths, hydrogen and alcohol synthesis paths, and a single hydrogen production path. Based on a greedy exploration strategy and the maximum reward value in the reward value table, a corresponding target action is selected from the path action space. The selected target action is optimized using a dual-network soft update mechanism and a priority experience replay strategy to obtain the hourly hydrogen resource synthesis path priority and the corresponding equipment start-stop control signal under the optimized target action. It can be understood that, as in the model operation mode of the initial hydrogen resource synthesis path priority prediction sub-model training process in the aforementioned embodiment, after the initial hydrogen resource synthesis path priority prediction sub-model is trained and the target hydrogen resource synthesis path priority prediction sub-model is obtained, the hourly hydrogen resource synthesis path priority and the corresponding equipment start-stop control signal are directly predicted.

[0157] Step S14: Estimate the second-level fluctuation of wind and solar power based on the target electrolyzer control sub-model of the pre-trained capacity configuration model, the priority of hydrogen resource synthesis path, the equipment start-up and shutdown control signal, and the wind and solar power and grid connection control benchmark data, and output the second-level power adjustment command of the electrolyzer.

[0158] In this embodiment, the priority of the hydrogen resource synthesis path, the equipment start-up and shutdown control signal, and the second-level wind and solar power and grid connection regulation benchmark data are input into the target electrolyzer regulation sub-model of the pre-trained capacity configuration model, so that the target electrolyzer regulation sub-model can estimate the second-level wind and solar power fluctuation through the disturbance observer and output the second-level power adjustment command of the electrolyzer.

[0159] Specifically, based on the wind and solar power and grid-connected regulation benchmark data, the error between the actual measured wind and solar power and the predicted wind and solar power is calculated so that the disturbance observer can estimate the second-level fluctuation of wind and solar power based on the error. Based on the second-level fluctuation of wind and solar power, the electrolyzer power value that minimizes the cost function of the target electrolyzer regulation sub-model is calculated, and the second-level power adjustment command for the electrolyzer is output. It can be understood that, as in the model operation mode of the initial electrolyzer regulation sub-model training process in the aforementioned embodiment, after the initial electrolyzer regulation sub-model is trained and the target electrolyzer regulation sub-model is obtained, the second-level power adjustment command for the electrolyzer is directly predicted and output.

[0160] Step S15: Output the full-cycle capacity configuration scheme of the wind-solar-hydrogen-ammonia-methanol system based on the pre-trained capacity configuration model and the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priority, the equipment start-stop control signal, and the second-level power adjustment command of the electrolyzer.

[0161] In this embodiment, based on the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priority, the equipment start-stop control signal, and the electrolyzer second-level power adjustment command, the operating parameters of each device in the wind-solar-hydrogen-ammonia-methanol system are adjusted to output the full-cycle capacity configuration scheme corresponding to the operating parameters; wherein, the operating parameters are used to control the overall cost of the wind-solar-hydrogen-ammonia-methanol system to be lower than a preset cost threshold, and the resource utilization rate to be higher than a preset utilization rate threshold. Understandably, based on hydrogen storage tank capacity parameters, hydrogen resource synthesis path priorities, equipment start-up and shutdown control signals, and second-level power adjustment commands from the electrolyzer, the PSO-GA-Hybrid hybrid optimization engine performs multi-objective collaborative optimization of the wind-solar-hydrogen-ammonia-methanol system. The multi-objective collaborative optimization constraints are as follows: economic constraints, specifically, the total lifecycle cost ≤ a preset total lifecycle cost threshold; efficiency constraints, specifically, the wind-solar absorption rate ≥ a preset wind-solar absorption rate threshold; and equipment protection constraints, specifically, the electrolyzer temperature ≤ a preset temperature threshold. Under these constraints, after PSO global search, update, and fitness evaluation, the optimized target result is output, which is the final capacity configuration scheme.

[0162] Reference Figure 3 As shown, this application discloses a specific capacity configuration scheme for wind-solar-hydrogen-ammonia-methanol storage. Specifically, in wind power meteorological scenarios, this… Figure 3 The wind-solar-hydrogen-ammonia-methanol system demonstrated is based on a three-tiered, interconnected optimization architecture of "second-hour-season." Each level achieves end-to-end control through meteorological data-driven and dynamic collaboration. The specific process is as follows:

[0163] I. Seasonal Hydrogen Demand Forecast:

[0164] Data input: Collect wind power meteorological data, including wind speed data, radiation data (core wind and solar resource parameters), historical hydrogen usage data, and other meteorological data (such as temperature), as the basic input for seasonal forecasts.

[0165] Attention mechanism processing: The above multi-dimensional data is input into the attention mechanism module, and the attention level is dynamically allocated according to the influence weight of wind power meteorological factors (such as wind speed and radiation intensity) on hydrogen demand (for example, photovoltaic output is high in seasons with strong radiation, and the corresponding hydrogen production and hydrogen demand may be higher), so as to solve the problem of information redundancy at different time steps and different meteorological variables.

[0166] LSTM model prediction: The input, after being weighted by the attention mechanism, enters the LSTM memory layer and the fully connected layer. It learns long-term dependencies by combining meteorological trends (such as seasonal wind speed changes and radiation fluctuations). The model is optimized by a loss function containing time-weighted mean square error and robust regularization term, and finally outputs the seasonal hydrogen prediction amount H(t). Based on this, the hourly hydrogen demand boundary amount for the current day is determined (providing capacity constraints for hourly scheduling).

[0167] II. Hourly-level multi-path collaborative scheduling:

[0168] State variable acquisition: Based on the hourly demand boundary of seasonal output, combined with real-time wind power meteorological derivative data (such as hourly predicted wind speed, expected wind and solar power output corresponding to radiation), carbon source (CO2) supply, nitrogen source (N2) supply, energy storage power station charge status, grid-connected electricity price, carbon price and equipment start-up and shutdown status, etc., the current state variable S(t) is constructed.

[0169] Improved DQN Decision Making: The state variable S(t) is processed using an improved deep reinforcement learning (DQN) model, where:

[0170] The reward function R(t) integrates the revenue from methanol / ammonia production, energy consumption costs (related to wind and solar power output; higher wind speed / radiation results in lower hydrogen production costs), equipment start-up and shutdown penalty costs (reducing frequent start-ups and shutdowns due to wind and solar fluctuations), and production gap penalties (avoiding supply and demand imbalances caused by weather forecast deviations).

[0171] Employing a dual-network structure, prioritizing experience replay (focusing on learning strategies during sudden weather changes in wind power) and an ε-greedy exploration mechanism, the optimal action value is calculated using the Q function, and the scheduling strategy A(t) for hydrogen in the ammonia synthesis and alcohol synthesis paths is output (i.e., determining the synthesis path after hydrogen production in the electrolyzer).

[0172] III. Real-time control of the electrolytic cell at the second level:

[0173] Disturbance sensing: Real-time acquisition of high-frequency fluctuation data of wind power meteorology (such as second-level wind speed changes and instantaneous changes in radiation), and estimation of the second-level disturbance component d(t) of wind and solar grid-connected power P(t) through the disturbance observer (DOB) (based on the error between the current wind and solar power measurement value and the predicted value), to achieve feedforward sensing of sudden fluctuations in wind power meteorology.

[0174] Improved MPC control: The disturbance observation d(t) is embedded into the improved model predictive control (IMPC) framework. The electrolyzer power command is calculated through an improved cost function (combining tracking accuracy and power smoothness, enhancing steady-state control when fluctuations are severe, and improving tracking accuracy when the conditions are stable). The electrolyzer power P_elec(t) is dynamically adjusted to match the second-level fluctuations in wind and solar power output, reduce wind and solar curtailment, and respond to the hour-level determined synthesis path requirements.

[0175] As can be seen, the seasonal level determines the demand boundary based on the long-term trend of wind power meteorology, the hourly level combines short- and medium-term meteorological forecasts and market information to optimize the synthesis path, and the second-level response adjusts the electrolyzer power in real time according to meteorological fluctuations. The three layers achieve parameter linkage through the parameter self-correcting PSO-GA-Hybrid hybrid optimization engine, ensuring the efficient and stable operation of the entire link in wind power meteorological scenarios.

[0176] As can be seen, this application discloses a capacity configuration method for wind, solar, hydrogen storage, ammonia, and methanol systems, comprising: acquiring multidimensional input data including historical meteorological data, system operation and supply-demand status data, and wind and solar power and grid-connected regulation benchmark data; predicting the daily hydrogen demand value and hydrogen storage tank capacity parameters under the future target seasonal time scale using a pre-trained capacity configuration model's target hydrogen demand prediction sub-model and based on the multidimensional input data; wherein, the target hydrogen demand prediction sub-model is a seasonal hydrogen demand prediction model; and predicting the daily hydrogen demand value and hydrogen storage tank capacity parameters under the future target seasonal time scale using the pre-trained capacity configuration model's target hydrogen resource synthesis path priority prediction sub-model and based on the daily hydrogen demand value and the system... The system predicts hourly hydrogen resource synthesis path priorities and corresponding equipment start-up and shutdown control signals based on system operation and supply and demand status data; it estimates second-level fluctuations in wind and solar power based on the target electrolyzer control sub-model of the pre-trained capacity configuration model, the hydrogen resource synthesis path priorities, the equipment start-up and shutdown control signals, and the wind and solar power and grid connection control benchmark data, and outputs second-level power adjustment commands for the electrolyzer; it outputs a full-cycle capacity configuration scheme for the wind-solar-hydrogen-ammonia-methanol system based on the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priorities, the equipment start-up and shutdown control signals, and the second-level power adjustment commands for the electrolyzer. This demonstrates that by co-optimizing the three sub-models of the pre-trained capacity configuration model, each sub-model is designed with a dedicated algorithm for different time scales, avoiding the high sensitivity of a single algorithm to parameters. The seasonal sub-model predicts long-term hydrogen demand and hydrogen storage tank capacity, laying the foundation for capacity configuration. The hourly sub-model optimizes hourly path priority and equipment start-up and shutdown, enabling short- to medium-term scheduling. The second-level sub-model estimates wind and solar fluctuations through a disturbance observer and outputs power adjustment commands to cope with high-frequency dynamic changes. By constructing a three-level time scale linkage mechanism of seasonal, hourly, and second-level, seamless connection of the three scales is achieved, breaking through the limitations of daily / monthly, and realizing full-cycle coverage from long-term configuration to real-time control.

[0177] Reference Figure 3 As shown, the present invention also provides a capacity configuration device for wind-solar-hydrogen storage of ammonia and alcohol, comprising:

[0178] Data acquisition module 11 is used to acquire multi-dimensional input data including historical meteorological data, system operation and supply and demand status data, wind and solar power and grid connection control benchmark data;

[0179] The first prediction module 12 is used to predict the daily hydrogen demand and hydrogen storage tank capacity parameters under the future target seasonal time scale by using the target hydrogen demand prediction sub-model of the pre-trained capacity configuration model and based on the multi-dimensional input data; wherein, the target hydrogen demand prediction sub-model is a seasonal hydrogen demand prediction model.

[0180] The second prediction module 13 is used to predict the hydrogen resource synthesis path priority prediction sub-model through the target hydrogen resource synthesis path priority prediction sub-model of the pre-trained capacity configuration model and predict the hourly hydrogen resource synthesis path priority and corresponding equipment start-stop control signals based on the daily hydrogen usage demand value and the system operation and supply and demand status data.

[0181] The third prediction module 14 is used to estimate the second-level fluctuation of wind and solar power based on the target electrolyzer control sub-model of the pre-trained capacity configuration model, the priority of the hydrogen resource synthesis path, the equipment start-up and shutdown control signal, and the wind and solar power and grid connection control benchmark data, and output the second-level power adjustment command of the electrolyzer.

[0182] The scheme generation module 15 is used to output a full-cycle capacity configuration scheme for the wind-solar-hydrogen-ammonia-methanol system based on the pre-trained capacity configuration model and the hydrogen storage tank capacity parameters, the priority of the hydrogen resource synthesis path, the equipment start-stop control signal, and the second-level power adjustment command of the electrolyzer.

[0183] This demonstrates that by co-optimizing the three sub-models of the pre-trained capacity configuration model, each sub-model is designed with a dedicated algorithm for different time scales, avoiding the high sensitivity of a single algorithm to parameters. The seasonal sub-model predicts long-term hydrogen demand and hydrogen storage tank capacity, laying the foundation for capacity configuration. The hourly sub-model optimizes hourly path priority and equipment start-up and shutdown, enabling short- to medium-term scheduling. The second-level sub-model estimates wind and solar fluctuations through a disturbance observer and outputs power adjustment commands to cope with high-frequency dynamic changes. By constructing a three-level time scale linkage mechanism of seasonal, hourly, and second-level, seamless connection of the three scales is achieved, breaking through the limitations of daily / monthly, and realizing full-cycle coverage from long-term configuration to real-time control.

[0184] Furthermore, embodiments of this application also disclose an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0185] Figure 4This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the capacity configuration method for wind-solar-hydrogen-ammonia-methanol storage disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be a computer.

[0186] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0187] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0188] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0189] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system 221 can be Windows Server, Netware, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the capacity configuration method for wind, solar, hydrogen storage, ammonia, and methanol disclosed in any of the foregoing embodiments executed by the electronic device 20, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0190] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned capacity configuration method for wind-solar-hydrogen-ammonia-methanol storage. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0191] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0192] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs (Compact Disc-Read Only Memory), or any other form of storage medium known in the art.

[0193] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0194] The solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for configuring the capacity of ammonia-ethanol storage for wind and solar power, characterized in that, include: Acquire multi-dimensional input data including historical meteorological data, system operation and supply and demand status data, and wind and solar power and grid connection control benchmark data; The target hydrogen demand prediction sub-model is pre-trained using a capacity configuration model, and the daily hydrogen demand and hydrogen storage tank capacity parameters are predicted based on the multidimensional input data at the target seasonal time scale; wherein, the target hydrogen demand prediction sub-model is a seasonal hydrogen demand prediction model. The target hydrogen resource synthesis path priority prediction sub-model of the pre-trained capacity configuration model is used to predict the hourly hydrogen resource synthesis path priority and corresponding equipment start-stop control signals based on the daily hydrogen usage demand value and the system operation and supply and demand status data. The target electrolyzer control sub-model of the pre-trained capacity configuration model is used to estimate the second-level fluctuation of wind and solar power based on the priority of hydrogen resource synthesis path, the equipment start-up and shutdown control signal, and the benchmark data of wind and solar power and grid connection regulation, and output the second-level power adjustment command of the electrolyzer. The pre-trained capacity configuration model is used to output a full-cycle capacity configuration scheme for the wind-solar-hydrogen-ammonia-methanol system based on the hydrogen storage tank capacity parameters, the priority of the hydrogen resource synthesis path, the equipment start-stop control signal, and the second-level power adjustment command of the electrolyzer.

2. The capacity configuration method for wind-solar-hydrogen-ammonia-ethanol storage according to claim 1, characterized in that, The target hydrogen demand prediction sub-model, which uses a pre-trained capacity configuration model, predicts daily hydrogen usage demand and hydrogen storage tank capacity parameters for the future target seasonal timescale based on the multi-dimensional input data, including: The historical meteorological data is weighted by the attention mechanism of the target hydrogen demand prediction sub-model of the pre-trained capacity configuration model and the LSTM network to predict the daily hydrogen demand value under the future target seasonal time scale. Based on the daily hydrogen usage demand, and according to the preset hydrogen storage constraints, the hydrogen storage tank capacity parameters are calculated.

3. The capacity configuration method for wind-solar-hydrogen-ammonia-ethanol storage according to claim 1, characterized in that, The target hydrogen resource synthesis path priority prediction sub-model predicts the hourly hydrogen resource synthesis path priority and the corresponding equipment start-up and shutdown control signals, including: The daily hydrogen demand value is converted into hourly demand boundary quantity. The hourly demand boundary quantity and the system operation and supply and demand status data are constructed as state variables. The path action space of hydrogen and ammonia synthesis path, hydrogen and alcohol synthesis path and single hydrogen production path are constructed. Based on the greedy exploration strategy and the maximum reward value in the reward value table, the corresponding target action is selected from the path action space; The target action is optimized based on the dual-network soft update mechanism and the priority experience playback strategy to obtain the priority of hydrogen resource synthesis path and the corresponding equipment start-up and shutdown control signals per hour under the optimized target action.

4. The capacity configuration method for wind-solar-hydrogen-ammonia-ethanol storage according to claim 1, characterized in that, The second-level fluctuation of wind and solar power is estimated through the target electrolyzer control sub-model, and the second-level power adjustment command of the electrolyzer is output, including: Based on the wind and solar power and grid-connected regulation reference data, the error between the actual measured wind and solar power and the predicted wind and solar power is calculated so that the disturbance observer can estimate the second-level fluctuation of wind and solar power based on the error. Based on the second-level fluctuation of wind and solar power, the electrolytic cell power value that minimizes the cost function of the target electrolytic cell control sub-model is calculated, and the second-level power adjustment command of the electrolytic cell is output.

5. The capacity configuration method for wind-solar-hydrogen-ammonia-ethanol storage according to any one of claims 1 to 4, characterized in that, Before inputting the multidimensional input data into the pre-trained capacity configuration model, the method further includes: Each training data point in the training dataset is input into the initial capacity configuration model, which includes the initial hydrogen demand prediction sub-model, the initial hydrogen resource synthesis path priority prediction sub-model, and the initial electrolyzer regulation sub-model, which are connected in sequence. The current sub-model to be trained is trained using the training data and / or the output of the previous target sub-model, so that the difference between the current output of the current sub-model to be trained and the corresponding label result in the training data satisfies the corresponding threshold condition, so as to obtain the target sub-model. The target sub-models are jointly trained to obtain a pre-trained capacity configuration model.

6. The capacity configuration method for wind-solar-hydrogen-ammonia-ethanol storage according to claim 5, characterized in that, The joint training of each of the target sub-models to obtain a pre-trained capacity configuration model includes: The global optimization engine dynamically adjusts the correlation parameters between the target hydrogen demand prediction sub-model, the target hydrogen resource synthesis path priority prediction sub-model, and the target electrolyzer control sub-model. This controls the hydrogen storage tank capacity parameters, path priorities, and power commands output by the model to collaboratively meet the full-cycle capacity configuration target, thereby obtaining a pre-trained capacity configuration model.

7. The capacity configuration method for wind-solar-hydrogen-ammonia-ethanol storage according to claim 1, characterized in that, The full-cycle capacity configuration scheme for the wind-solar-hydrogen-ammonia-methanol system, based on the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priority, the equipment start-up and shutdown control signals, and the second-level power adjustment command output of the electrolyzer, includes: Based on the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priority, the equipment start-stop control signals, and the electrolyzer second-level power adjustment commands, the operating parameters of each device in the wind-solar-hydrogen-ammonia-methanol system are adjusted to output the corresponding full-cycle capacity configuration scheme under the operating parameters. The operating parameters are used to control the overall cost of the wind-solar-hydrogen-ammonia-methanol system to be lower than a preset cost threshold and the resource utilization rate to be higher than a preset utilization rate threshold.

8. A capacity configuration device for wind-solar-hydrogen storage ammonia-ethanol, characterized in that, include: The data acquisition module is used to acquire multi-dimensional input data, including historical meteorological data, system operation and supply and demand status data, and wind and solar power and grid connection control benchmark data. The first prediction module is used to predict the daily hydrogen demand and hydrogen storage tank capacity parameters under the future target seasonal time scale by using a pre-trained capacity configuration model of the target hydrogen demand prediction sub-model and based on the multi-dimensional input data; wherein, the target hydrogen demand prediction sub-model is a seasonal hydrogen demand prediction model. The second prediction module is used to predict the priority of hydrogen resource synthesis path prediction sub-model through the target hydrogen resource synthesis path priority prediction sub-model of the pre-trained capacity configuration model, and predict the priority of hydrogen resource synthesis path and the corresponding equipment start-stop control signal for each hour based on the daily hydrogen usage demand value and the system operation and supply and demand status data. The third prediction module is used to estimate the second-level fluctuation of wind and solar power based on the target electrolyzer control sub-model of the pre-trained capacity configuration model, the priority of the hydrogen resource synthesis path, the equipment start-up and shutdown control signal, and the wind and solar power and grid connection control benchmark data, and output the second-level power adjustment command of the electrolyzer. The scheme generation module is used to output a full-cycle capacity configuration scheme for the wind-solar-hydrogen-ammonia-methanol system based on the pre-trained capacity configuration model and the hydrogen storage tank capacity parameters, the hydrogen resource synthesis path priority, the equipment start-stop control signal, and the second-level power adjustment command of the electrolyzer.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the capacity configuration method for wind-solar-hydrogen-ammonia-ethanol as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the capacity configuration method for wind-solar-hydrogen-ammonia-ethanol as described in any one of claims 1 to 7.

Images