An adaptive control and configuration optimization method for source-storage-load power to improve the safety of energy storage systems

Abstract

This invention discloses a source-storage-load power adaptive control and configuration optimization method for improving the safety of energy storage systems. The method includes the following steps: First, a wind power time-series prediction network and a solar spectral attention mechanism are used to quickly predict the power of wind power and photovoltaic power, respectively, and generate a total power control command for adjustable loads. Second, a multi-strategy adaptive controller is used to track this command, and the state of charge of the energy storage battery is introduced for adaptive correction and protection. Next, an adjustable capacity proportional allocation strategy is used to distribute the total command to each load unit. Finally, based on the above-mentioned collaborative operation data, a multi-objective quantum spatiotemporal modal programming algorithm is used to collaboratively optimize the power and energy capacity of the energy storage system, aiming to minimize the net present value cost over the entire system's lifecycle. This invention improves the safety and stability of energy storage systems, reduces energy losses, and optimizes the operating efficiency and economy of the power grid through source-storage-load power adaptive control and configuration optimization.

Description

Technical Field

[0001] This invention relates to a power system operation and control technology, and more particularly to a source-storage-load power adaptive control and configuration optimization method for improving the safety of energy storage systems. Background Technology

[0002] With the increasing penetration rate of new energy sources such as wind power and photovoltaics, the volatility and uncertainty of power output on the source side of the power system have significantly increased, posing higher requirements for system power balance, frequency and voltage regulation, and safe and stable operation. Energy storage systems, with their characteristics of fast response, flexible power regulation, and controllable charging and discharging, have become an important technical means to improve the stability and security of power systems with a high proportion of new energy sources.

[0003] However, existing energy storage technologies still have shortcomings in planning and operation. At the planning level, current energy storage capacity configuration methods mainly focus on economic efficiency, peak shaving and valley filling, or renewable energy consumption, rarely translating the dynamic support capabilities required for the safe and stable operation of the system into quantitative constraints on energy storage power capacity and energy capacity. This easily leads to insufficient capacity configuration or redundant investment. At the operation level, if the main task of power regulation is undertaken by the energy storage system, the energy storage will be in a state of high-frequency, deep charging and discharging for a long time, resulting in accelerated equipment aging, shortened lifespan, and increased operating costs. Furthermore, it may lead to insufficient regulation margin in extreme scenarios such as drastic fluctuations in renewable energy or sudden load changes, affecting the safe operation of the system.

[0004] Meanwhile, demand-side adjustable loads, as an important flexibility resource, possess certain power regulation potential. However, existing demand response technologies mostly employ price signals or discrete dispatching methods, resulting in limited response speed and control precision. They primarily serve peak shaving and load shifting, making it difficult to meet the power system's demands for rapid, continuous, and precise power regulation. Their role in sharing the pressure of energy storage regulation and improving system security has not yet been fully realized.

[0005] Currently, the optimization and allocation of energy storage and the scheduling of demand-side resources are still relatively separate. There is a lack of unified modeling and collaborative control mechanisms between new energy sources, energy storage systems and adjustable loads. It is difficult to dynamically allocate adjustment responsibilities based on the output of new energy sources, load changes and the operating status of energy storage, resulting in greater adjustment pressure on energy storage systems and the failure to fully realize the advantages of source-storage-load resource synergy.

[0006] Therefore, existing technologies urgently need a source-storage-load power adaptive control and configuration optimization method for improving energy storage safety. This method should be able to collaboratively determine energy storage capacity and adjustable load regulation capability during the planning stage, and achieve dynamic coordination and responsibility sharing of resources on the source side, energy storage side, and load side during the operation stage. This would reduce the pressure on energy storage operation, improve system safety margin, and take into account both the renewable energy absorption capacity and overall operational economy. Summary of the Invention

[0007] Purpose of the invention: The purpose of this invention is to provide a source-storage-load power adaptive control and configuration optimization method for improving the safety of energy storage systems, which coordinates and optimizes operation and planning, reduces the pressure of energy storage regulation, and improves economy and safety.

[0008] Technical solution: The source-storage-load power adaptive control and configuration optimization method for improving the safety of energy storage systems, as described in this invention, includes the following steps:

[0009] (1) A wind power time series prediction network and a solar spectrum attention mechanism are used to perform fast power prediction on wind power and photovoltaic power generation systems respectively to obtain the predicted power value of wind and solar combined power generation. Based on the predicted value and the planned reserved power of adjustable load, a total adjustable load power control command is generated. The wind power time series prediction network is a time series network based on memory units, and the solar spectrum attention mechanism is an attention model based on frequency domain random feature mapping.

[0010] (2) A multi-strategy adaptive controller is used to track the total adjustable load power control command and correct it according to the state of charge of the energy storage battery to obtain the corrected adjustable load total power adjustment reference command; wherein, the working mode of the multi-strategy adaptive controller includes multiple parameter adjustment strategies outputting candidate parameters in parallel, predicting the performance of each strategy based on forward simulation, and fusing the output of each strategy according to dynamic reputation to generate the final control parameters.

[0011] (3) Based on the adjustable load total power adjustment reference instruction, an adjustable capacity ratio allocation strategy is adopted to allocate adjustment power instructions to each adjustable load unit and generate the final power control setting value for each adjustable load.

[0012] (4) Based on the system operation process and data determined in steps (1) to (3), a multi-objective quantum spatiotemporal modal programming algorithm is adopted to optimize the power capacity and energy capacity of the energy storage system with the goal of minimizing the net present value cost of the entire life cycle of the wind-solar-storage-electrodecoupling system; wherein, the multi-objective quantum spatiotemporal modal programming algorithm includes mode decomposition, spatiotemporal feature aggregation and multi-objective optimization.

[0013] Preferably, in step (1), the time-series network based on memory units is a memory unit network that includes a forget gate, an input gate, candidate memory units, and an output gate.

[0014] Preferably, in step (1), the calculation process of the solar spectral attention mechanism includes:

[0015] (11) Extract local features from the input sequence using a one-dimensional convolutional layer;

[0016] (12) Perform a linear transformation on the features output by the convolution to obtain the original query feature matrix, the original key feature matrix and the projection weight matrix;

[0017] (13) Using a mapping function based on random Fourier features, the original query feature matrix and the original key feature matrix are mapped to the frequency domain respectively to obtain frequency domain query mapping and frequency domain key mapping, wherein the mapping function is constructed by transforming the product of the matrix and the randomly generated Gaussian matrix by sine and cosine functions;

[0018] (14) Based on the frequency domain query mapping, frequency domain key mapping and projection weight matrix, calculate the attention output as the predicted value of photovoltaic power.

[0019] Preferably, in step (2), the operation of the multi-strategy adaptive controller includes:

[0020] (21) Multiple adaptive algorithms run in parallel and output candidate values ​​for control parameters;

[0021] (22) Use an online simplified model to perform forward simulation on each group of candidate parameters to predict their future performance;

[0022] (23) Based on the dynamic reputation that integrates historical reliability and real-time prediction performance, the candidate values ​​of the output parameters of each strategy are weighted and fused to generate the final control parameters that are optimal in the whole.

[0023] Preferably, in step (2), the correction based on the state of charge of the energy storage battery specifically involves: dynamically scaling and offsetting the adjustable load power control command according to the degree to which the battery state of charge deviates from the safe operating range, so as to prevent the battery from being overcharged or discharged.

[0024] Preferably, the adjustable load power control command is dynamically scaled and offset adjusted using the following formula:

[0025]

[0026] Where: β is the SOC correction offset of the energy storage battery; K SOC Correct the SOC gain for the energy storage battery; The adjustable load power control command generated at time τ To allow for flexible adjustment of power commands after the SOC correction of the energy storage battery.

[0027] Preferably, in step (3), the adjustable capacity ratio allocation strategy is as follows:

[0028] When it is necessary to increase the total power load, the power shall be allocated according to the proportion of the current remaining adjustable capacity of each adjustable load;

[0029] When it is necessary to reduce the total electrical load, power should be allocated according to the current operating power ratio of each adjustable load.

[0030] Preferably, in step (4), the execution process of the multi-objective quantum spatiotemporal modal programming algorithm includes the following stages:

[0031] (41) Mode decomposition and feature extraction stage: The net load fluctuation sequence of the wind-solar-storage-electrolysis coupling system is decomposed to obtain the intrinsic mode components and the energy density index of each mode component is calculated to quantify the energy storage demand of fluctuations at different time scales;

[0032] (42) Spatiotemporal graph convolution feature aggregation stage: Construct a graph model containing wind power nodes, photovoltaic nodes and load nodes, aggregate spatiotemporal features through graph convolution network, and extract key fluctuation patterns and related features that affect energy storage configuration;

[0033] (43) Dual-scale dynamic collaborative optimization stage: Establish a dual-time-scale optimization model that couples short-term operating costs and long-term investment costs, with the goal of minimizing the net present value cost of the entire system life cycle, and collaboratively solve for the optimal energy storage power capacity and energy capacity;

[0034] (44) Quantum heuristic global search stage: A quantum heuristic cooperative search algorithm is used to perform a parallel global search in the continuous solution space composed of the power capacity and energy capacity to obtain an approximately optimal energy storage capacity configuration scheme that satisfies multiple constraints.

[0035] Preferably, in step (4), the net present value cost of the wind-solar-storage-electrolysis coupling system throughout its entire life cycle includes the total investment cost of the system, operation and maintenance cost, grid interaction cost, and revenue from electrolysis products.

[0036] Preferably, the total investment cost of the system includes at least the power and capacity investment costs of the energy storage system, photovoltaic system, wind turbine, and electrolyzer.

[0037] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) It realizes the fast and accurate tracking of wind and solar fluctuations of adjustable load, effectively reducing the real-time adjustment pressure of energy storage system; (2) It realizes the coordinated optimization configuration of energy storage power and energy capacity, significantly reducing the system's total life cycle cost; (3) Through the coordinated operation-planning process, it improves the overall economy while ensuring the stability of the power grid. Attached Figure Description

[0038] Figure 1 This is a structural diagram of the adjustable load power control and energy storage system capacity optimization configuration of the present invention;

[0039] Figure 2This is a schematic diagram of the wind power time-series prediction network of the present invention;

[0040] Figure 3 This is a schematic diagram of the solar spectrum attention mechanism of the present invention;

[0041] Figure 4 This is a schematic diagram of the multi-strategy adaptive controller of the present invention;

[0042] Figure 5 This is a schematic diagram of the multi-objective quantum spatiotemporal modal programming algorithm of the present invention. Detailed Implementation

[0043] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0044] As shown in the attached figure, this invention provides a source-storage-load power adaptive control and configuration optimization method for improving the safety of energy storage systems. This method is applied to a wind-solar-storage-electrodecoupled system comprising wind power, photovoltaics, energy storage, and adjustable loads, performing collaborative optimization at both the operation control and planning configuration levels. First, in the operation control loop, a fast power prediction algorithm is used to predict the wind and solar power generation in advance, generating power control commands for the adjustable loads based on the prediction results. Next, a multi-strategy adaptive controller achieves high-precision tracking control, and the state of charge (SOC) of the energy storage batteries is introduced for protective correction. Then, a proportional allocation strategy is used to issue the total control command to each adjustable load unit for execution. The system net load characteristic data generated during the above operation process is used as input to the planning loop. Through a multi-objective quantum spatiotemporal modal programming algorithm, the optimal power capacity and energy capacity configuration scheme of the energy storage system is finally solved.

[0045] Step 1: Rapid Power Forecasting and Load Control Command Generation for Wind and Solar Power Generation

[0046] The core of this step is to quickly and accurately predict the combined output of wind and solar power, and based on this, generate preliminary adjustable load adjustment commands.

[0047] (1) A wind power time series prediction network is used to predict the power of the wind farm. (See attached diagram) Figure 2 As shown, this network is a time-series network with memory function, and its mathematical model is defined by the following set of formulas:

[0048]

[0049] in: The input vector at the current time step; The hidden state of the input vector from the previous time step; is the activation function; tanh is the hyperbolic tangent function; , , , These are the forget gate weight matrix, input gate weight matrix, candidate memory weight matrix, and output gate weight matrix, respectively. , , , These are the inherent tendencies of the forget gate, the input gate, the candidate memory generation, and the output gate, respectively. , , , , , These are the forget gate vector, input gate vector, candidate memory vector, updated long-term memory vector, output gate vector, and current hidden state vector, respectively.

[0050] The algorithm utilizes the synergistic effects of unit states, forget gates, input gates, and output gates to sequentially complete historical information filtering, real-time data fusion, and memory state updates. Ultimately, it extracts key patterns from the dynamically evolving temporal features and directly outputs predicted wind power values ​​for future moments, accurately depicting the temporal evolution of wind energy.

[0051] (2) The solar spectral attention mechanism is used to predict the power of photovoltaic power plants. This mechanism directly extracts the characteristics of the irradiance period and cloud change through frequency domain transformation, maps the time series information into a spectrum representation, and captures global dependencies in parallel to achieve end-to-end high-precision and fast prediction of photovoltaic power from historical data to future power.

[0052] (21) Local features of output photovoltaic power generation data and historical weather data are extracted through a one-dimensional convolutional layer. The formula is as follows:

[0053]

[0054] in: This is the input to the initial convolutional layer. The input features are at time t; This is the result after the operation of the i-th convolutional layer; This is the output of the previous layer; These are weight parameters; It is a deviation term; To correct the activation function of the linear unit; Conv1D(⋅) is a one-dimensional convolution operation.

[0055] (22) Convert the output value of the convolution As input to the execution layer, the attention mechanism of the execution layer bypasses complex similarity calculation steps by introducing orthogonal random features, mapping the input data to a high-dimensional space. Then, it approximates the self-attention kernel using a random projection method, thus simplifying the calculation of the attention score. Finally, it obtains the original query feature matrix C, the original key feature matrix J, and the projection weight matrix Z through a linear transformation, as shown in the formula:

[0056]

[0057]

[0058] in: The output value of the convolution; For the input of the execution layer, , , This is the weight matrix; , For the original query feature matrix, It is the original key feature matrix. is the projection weight matrix; L is the dimension of the input sequence, and d is the dimension of the hidden layer.

[0059] (23) Solar spectral attention is calculated using random eigenmaps, and the formula is as follows:

[0060]

[0061] Where: diag(⋅) extracts the main diagonal elements of the matrix to form a vector; sin(⋅) and cos(⋅) are the sine and cosine functions, respectively; This is the feature matrix for deep querying; It is the deep key feature matrix; D is the dimension of the features; This is a dynamically calculated scaling factor; A randomly generated Gaussian matrix; For frequency domain lookup mapping; Frequency domain bond mapping; Attention is solar spectral attention.

[0062] (3) Combining the prediction results of wind power and photovoltaic power, the rapid power prediction value of the wind and solar power generation system is obtained. Subsequently, this prediction value and the reserved power of the adjustable load plan are used to generate the total power control command of the adjustable load. The calculation formula is as follows:

[0063]

[0064]

[0065] Where: m is the total number of wind farms; l is the wind farm number index; This represents the actual active power measurement value of the l-th wind farm at time τ-k; This represents the actual active power measurement value of the photovoltaic power station at time τ-k; and are the backtracking window length and smoothing coefficient in the wind power time series prediction network; τ represents the time index, k represents the discrete time index; τ represents the time scale of the adjustable load power control strategy; t represents the time index in the power prediction time scale of the wind and solar power generation system. This represents the rapid power prediction value of the wind and solar power generation system at time τ-1; is the rapid power prediction value of the wind and solar power generation system at time τ; n is the total number of adjustable loads; i is the index; Reserved power for the i-th adjustable load at time t; This refers to the adjustable load power control command generated at time τ.

[0066] Step 2: Multi-strategy adaptive control and energy storage battery SOC correction

[0067] This step aims to achieve high-precision, robust tracking of the aforementioned power commands and ensure that the energy storage battery operates within a safe range.

[0068] (1) For instructions Perform SOC calibration on the energy storage battery to prevent overcharging and over-discharging. The calculation formula is as follows:

[0069]

[0070] Where: β is the SOC correction offset of the energy storage battery; K SOC Correct the SOC gain for the energy storage battery; To allow for flexible adjustment of power commands after the SOC correction of the energy storage battery.

[0071] (2) As attached Figure 4 As shown, a multi-strategy adaptive controller is used to track the corrected instructions. The core of this controller lies in its parameter adaptation mechanism:

[0072] (21) Strategy Arena: In the strategy arena, multiple adaptive algorithms run in parallel, each outputting its recommended candidate parameter values. The calculation formula is as follows:

[0073]

[0074] Where: j is the strategy number; The candidate parameter vector output by the j-th parameter tuning strategy; The proportional coefficient for the j-th strategy suggestion; is the integral coefficient of the j-th strategy suggestion; M is the total number of strategies running in parallel.

[0075] (22) Virtual evaluation: To evaluate the safety of each candidate parameter, the controller uses an online simplified model for forward simulation to predict its future performance. The calculation formula is as follows:

[0076]

[0077] in: Let be the prediction performance metric for the j-th strategy; , These are the weighting coefficients for the error term and the control variation term; For using parameters The prediction error at step i; For using parameters The control increment at step i; N is the prediction step size.

[0078] (23) Reputation Fusion: Each strategy has a dynamically updated reputation score, which integrates its historical reliability and current prediction performance to achieve long-term evaluation. The calculation formula is as follows:

[0079]

[0080] in: Let be the reputation score of the j-th strategy; Let be the reputation score of the j-th strategy at the previous time step; Forgetting factor; is the scale parameter; exp(⋅) is the exponential operation function.

[0081] (24) Weighted decision-making: The meta-decision maker assigns fusion weights to each strategy based on reputation and predictive performance. The calculation formula is as follows:

[0082]

[0083] in: Let be the fusion weight of the j-th strategy; Sensitivity coefficient; Use the strategy number index.

[0084] (25) By weighting and fusing the outputs of all strategies, the final control parameters that are optimal in combination are generated. The calculation formula is as follows:

[0085]

[0086] in: , The final adaptive proportional and integral control parameters; The proportional coefficient for the j-th strategy suggestion; Let be the integral coefficient of the j-th strategy suggestion.

[0087] 3. The controller uses the parameters obtained above. and The control quantity is calculated and output based on the power deviation. The calculation formula is as follows:

[0088]

[0089] in: and These are the control variables for the current time and the previous time, respectively; and These represent the power deviations at the current and previous times, respectively. and These are the proportional and integral coefficients of the multi-strategy adaptive controller, respectively. The sampling period.

[0090] 4. Generate a reference command for adjusting the total power of the adjustable load. The formula is:

[0091]

[0092] in: To allow for flexible adjustment of power commands after SOC correction of the energy storage battery; K P and K I For the proportional and integral coefficients of the multi-strategy adaptive controller, Reserved power for the i-th adjustable load at time t; Let be the actual operating power of the i-th adjustable load at time τ within period t; This is a reference command for adjusting the total power of adjustable loads.

[0093] Step 3: Adjustable load power distribution

[0094] This step will adjust the total power of the adjustable load according to the reference instructions. The load is allocated fairly and efficiently to each adjustable load device. The allocation strategy is based on the real-time adjustable capacity of each load, using a proportional allocation method: when the total power load needs to be increased, allocation is made according to the proportion of the remaining adjustable capacity of each load; when the total power load needs to be decreased, allocation is made according to the proportion of the current operating power (i.e., the adjustable capacity) of each load. The formula for increasing or decreasing the load allocation value for adjustable loads is as follows:

[0095]

[0096] in: Let be the rated capacity of the i-th adjustable load; This is the power adjustment command assigned to the i-th adjustable load.

[0097] Ultimately, the power control setpoint received by each adjustable load is the sum of its original planned power and the allocated adjustment command, as shown in the formula:

[0098]

[0099] in, This is the final power control setting value received by the i-th adjustable load.

[0100] Step 4: Optimize the capacity configuration of the energy storage system

[0101] This step, based on the system net load data generated by the aforementioned coordinated operation control, performs coordinated optimization configuration of the power capacity and energy capacity of the energy storage system.

[0102] As attached Figure 5 As shown, a multi-objective quantum spatiotemporal modal programming algorithm is adopted. By decomposing load fluctuation modes, extracting spatiotemporal features through graph convolution, co-optimizing through dual-scale dynamic programming, and using quantum-inspired search for global optimization, a high-precision joint configuration of energy storage power and energy capacity is achieved, improving economy and robustness. The specific steps are as follows:

[0103] (41) Modal energy decomposition and feature extraction: Modal decomposition is performed on the net load fluctuation sequence obtained from long-term operation simulation to obtain intrinsic modal components. The energy density index of each modal component is calculated to quantify the energy storage demand of fluctuations at different time scales. The formula is as follows:

[0104]

[0105] in: is the energy density index of the a-th modal component; a is the modal component index; The length of the time series; Let be the amplitude of 'a' intrinsic mode components at time h; h is the discrete-time index of the optimization algorithm; b is the regularization coefficient. The time derivative of the modal component; It is the square of the 2-norm.

[0106] (42) Spatiotemporal graph convolutional feature aggregation: Construct a graph model containing wind, solar, and load nodes, aggregate spatiotemporal features through a graph convolutional network, and extract key fluctuation patterns and related features that affect energy storage configuration. The formula is as follows:

[0107]

[0108] In the formula: and , respectively, are the feature matrices of nodes in layers r and r+1; r is the network layer index; S is the number of relation types; s is the relation type index; Let be the adjacency matrix of the s-th type of relation; Let be the degree matrix of the s-th type of relation; It is the negative 1 / 2 power of the degree matrix; Let be the learnable weight matrix of the s-th class relation in the r-th layer; Let g be the bias matrix of the r-th layer; g is the nonlinear activation function.

[0109] (43) Dual-scale dynamic programming optimization: An optimization model is established with the objective of minimizing the net present value cost of the wind-solar-storage-electrolysis coupling system over its entire life cycle. The objective function is:

[0110]

[0111] in: The minimum net present value cost of a wind-solar-storage-electrolysis coupled system; r m N is the discount rate; k For the project's lifespan; Annual maintenance costs; Annual grid interaction cost; For revenue from electrolytic products; The total investment cost of the wind-solar-storage-electrolysis coupling system is calculated using the following formula:

[0112]

[0113] in: The unit power investment cost of the electrolytic cell; This refers to the rated power of the electrolytic cell; Cost per unit power of energy storage systems; This refers to the rated power of the energy storage system. The unit energy cost of the energy storage system; This refers to the rated capacity of the energy storage system. Cost per unit peak power of a photovoltaic system; This represents the peak power of the photovoltaic array; Cost per unit rated power of wind turbine units; This refers to the rated power of the wind turbine generator set; For additional power conversion system unit cost; Rated power for the additional power conversion system.

[0114] By employing dynamic programming across two time scales, the optimal energy storage power capacity and energy capacity are determined to balance short-term operating costs with long-term investment costs. The calculation formula is as follows:

[0115]

[0116] in: Let P be the objective function for power capacity P and energy capacity E; P is the power capacity of the energy storage system; E is the energy capacity of the energy storage system. The battery charging and discharging power at time q; Let be the battery energy level at time q; Let be the instantaneous cost function at time q; q represents the total number of time points; q represents the index of the time step number. This is the terminal cost weighting coefficient; This is the terminal cost function.

[0117] (44) Quantum heuristic cooperative search: Quantum heuristic cooperative search is performed to achieve efficient global search in the continuous solution space, avoiding getting trapped in local optima. The calculation formula is as follows:

[0118]

[0119] Where: u is the iterative algebra index; v is the particle number index; and Let be the position vector of particle v in the uth and u+1th generations; Let be the local attraction point of particle v in the uth generation; The coefficient of contraction and expansion; The average optimal position; It is a uniform random vector; , , These represent quantum addition, quantum subtraction, and quantum multiplication operations, respectively; ln(⋅) is the natural logarithm function.

[0120] Through iteration, the system quickly converges to an approximately optimal energy storage capacity configuration scheme that satisfies the system stability and economic constraints.

Claims

1. A source-storage-load power adaptive control and configuration optimization method for energy storage system safety improvement, characterized in that, Includes the following steps: (1) A wind power time series prediction network and a solar spectrum attention mechanism are used to perform fast power prediction on wind power and photovoltaic power generation systems respectively to obtain the predicted power value of wind and solar combined power generation. Based on the predicted value and the planned reserved power of adjustable load, a total adjustable load power control command is generated. The wind power time series prediction network is a time series network based on memory units, and the solar spectrum attention mechanism is an attention model based on frequency domain random feature mapping. (2) A multi-strategy adaptive controller is used to track the total adjustable load power control command and correct it according to the state of charge of the energy storage battery to obtain the corrected adjustable load total power adjustment reference command; wherein, the working mode of the multi-strategy adaptive controller includes multiple parameter adjustment strategies outputting candidate parameters in parallel, predicting the performance of each strategy based on forward simulation, and fusing the output of each strategy according to dynamic reputation to generate the final control parameters. (3) Based on the adjustable load total power adjustment reference instruction, an adjustable capacity ratio allocation strategy is adopted to allocate adjustment power instructions to each adjustable load unit and generate the final power control setting value for each adjustable load. (4) Based on the system operation process and data determined in steps (1) to (3), a multi-objective quantum spatiotemporal modal programming algorithm is adopted to optimize the power capacity and energy capacity of the energy storage system with the goal of minimizing the net present value cost of the entire life cycle of the wind-solar-storage-electrodecoupling system; wherein, the multi-objective quantum spatiotemporal modal programming algorithm includes mode decomposition, spatiotemporal feature aggregation and multi-objective optimization.

2. The method of claim 1, wherein, In step (1), the time-series network based on memory units is a memory unit network that includes a forget gate, an input gate, candidate memory units, and an output gate.

3. The method of claim 1, wherein, In step (1), the calculation process of the solar spectral attention mechanism includes: (11) Extract local features from the input sequence using a one-dimensional convolutional layer; (12) Perform a linear transformation on the features output by the convolution to obtain the original query feature matrix, the original key feature matrix and the projection weight matrix; (13) Using a mapping function based on random Fourier features, the original query feature matrix and the original key feature matrix are mapped to the frequency domain respectively to obtain frequency domain query mapping and frequency domain key mapping, wherein the mapping function is constructed by transforming the product of the matrix and the randomly generated Gaussian matrix by sine and cosine functions; (14) Based on the frequency domain query mapping, frequency domain key mapping and projection weight matrix, calculate the attention output as the predicted value of photovoltaic power.

4. The method according to claim 1, characterized in that, In step (2), the operation of the multi-strategy adaptive controller includes: (21) Multiple adaptive algorithms run in parallel and output candidate values ​​for control parameters; (22) Use an online simplified model to perform forward simulation on each group of candidate parameters to predict their future performance; (23) Based on the dynamic reputation that integrates historical reliability and real-time prediction performance, the candidate values ​​of the output parameters of each strategy are weighted and fused to generate the final control parameters that are optimal in the whole.

5. The method of claim 1, wherein, In step (2), the correction based on the state of charge of the energy storage battery specifically involves: dynamically scaling and offsetting the adjustable load power control command according to the degree to which the battery state of charge deviates from the safe operating range, so as to prevent the battery from being overcharged or discharged.

6. The method of claim 5, wherein, The dynamic scaling and offset adjustment of the adjustable load power control command is achieved through the following formula: Wherein: β is the energy storage battery SOC correction offset; K SOC is the energy storage battery SOC correction gain; is the adjustable load power control instruction generated at time τ, is the flexible regulation power instruction after considering the energy storage battery SOC correction.

7. The method of claim 1, wherein, In step (3), the adjustable capacity ratio allocation strategy is as follows: When it is necessary to increase the total power load, the power shall be allocated according to the proportion of the current remaining adjustable capacity of each adjustable load; When it is necessary to reduce the total electrical load, power should be allocated according to the current operating power ratio of each adjustable load.

8. The method of claim 1, wherein, In step (4), the execution process of the multi-objective quantum spatiotemporal modality planning algorithm includes the following stages: (41) Mode decomposition and feature extraction stage: Modal decomposition is performed on the net load fluctuation sequence of the wind-solar-storage-electrolysis coupling system to obtain intrinsic mode components and calculate the energy density index of each mode component in order to quantify the energy storage demand of fluctuations at different time scales. (42) Spatiotemporal graph convolution feature aggregation stage: Construct a graph model containing wind power nodes, photovoltaic nodes and load nodes, aggregate spatiotemporal features through graph convolution network, and extract key fluctuation patterns and related features that affect energy storage configuration; (43) Dual-scale dynamic collaborative optimization stage: Establish a dual-time-scale optimization model that couples short-term operating costs and long-term investment costs, with the goal of minimizing the net present value cost of the entire system life cycle, and collaboratively solve for the optimal energy storage power capacity and energy capacity; (44) Quantum heuristic global search stage: A quantum heuristic cooperative search algorithm is used to perform a parallel global search in the continuous solution space composed of the power capacity and energy capacity to obtain an approximately optimal energy storage capacity configuration scheme that satisfies multiple constraints.

9. The method of claim 1, wherein, In step (4), the net present value cost of the wind-solar-storage-electrolysis coupling system throughout its entire life cycle includes the total investment cost of the system, operation and maintenance cost, grid interaction cost, and revenue from electrolysis products.

10. The method of claim 9, wherein, The total investment cost of the system includes at least the power and capacity investment costs of the energy storage system, photovoltaic system, wind turbine, and electrolyzer.

Images