Wind power and photovoltaic integrated energy storage prediction method

Abstract

The invention provides a prediction method supporting wind power and photovoltaic comprehensive energy storage, and relates to the technical field of machine learning. The invention provides an E-STENet prediction model which comprises a spatial correlation modeling module, a time dependence module and a prediction module, the spatial correlation modeling module is used for processing spatial correlation of data, and the time dependence modeling module is used for processing time dependence of the data. The prediction module is used for converting the hidden state into a wind power and photovoltaic comprehensive energy storage prediction result of a future time step, and all the modules cooperate with one another to complete accurate prediction of wind power and photovoltaic comprehensive energy storage.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and in particular relates to a method for predicting wind power and photovoltaic integrated energy storage. Background Art

[0002] With the transformation of the global energy structure, wind power and photovoltaics have gradually become mainstream as renewable energy sources. However, wind power and photovoltaic power generation are volatile and uncertain, and are greatly affected by weather and seasonal changes, which makes their power output difficult to predict, posing challenges to grid scheduling and stability. Therefore, how to accurately predict the power generation of wind power and photovoltaics and adjust the supply and demand of electricity through energy storage systems has become one of the key technologies to achieve the widespread application of clean energy. Accurate prediction models can reduce dependence on traditional energy and improve the utilization rate of renewable energy.

[0003] Wind power and photovoltaic integrated energy storage equipment are affected by many factors, including environmental conditions. Accurately predicting the energy storage capacity of these devices can provide a basis for the formulation of energy-saving strategies and the intelligent scheduling of equipment. For example, by storing excess electricity, the energy storage system releases electricity when power generation is low to ensure the stability of the power grid. In addition, energy storage prediction is also crucial for equipment fault diagnosis. Abnormal energy storage may indicate the potential risk of equipment failure. By identifying abnormal energy storage in advance, equipment maintenance costs can be reduced.

[0004] At present, the prediction of wind power and photovoltaic integrated energy storage faces the following problems. First, there is a significant spatial dependence between different energy storage devices. Second, under extreme conditions, the energy storage of wind power and photovoltaic equipment will fluctuate to extreme values. This abnormal energy storage phenomenon is difficult to capture through traditional linear prediction models. Finally, the impact of environmental factors and equipment characteristics on energy storage is often nonlinear. Existing statistical models are difficult to fully capture these complex relationships. Therefore, inventing a method to support wind power and photovoltaic integrated energy storage prediction is of great significance to optimizing the energy management of wind power and photovoltaic integrated energy storage equipment. Summary of the invention

[0005] The present invention provides a method for predicting wind power and photovoltaic integrated energy storage. Aiming at wind power and photovoltaic integrated energy storage data with spatial dependence, extreme value fluctuation and nonlinearity, an E-STENet prediction model is proposed, which consists of a spatial correlation modeling module, a time dependence module and a prediction module.

[0006] The technical solution adopted by the present invention to achieve the above-mentioned purpose specifically includes the following steps: S1. Collect wind power and photovoltaic integrated energy storage related data, including characteristics and target variables, and pre-process the collected wind power and photovoltaic integrated energy storage data; S2, using the Max-Min method to normalize the wind power and photovoltaic integrated energy storage data and divide the data set proportionally; S3. Construct a spatial correlation modeling module to process the spatial correlation of data, which specifically includes the following steps: S31, input normalized wind power and photovoltaic comprehensive energy storage data X t ∈R N , based on the observed values ​​of T historical wind power and photovoltaic comprehensive energy storage data, predict the wind power and photovoltaic comprehensive energy storage value in the future q steps; S32. Calculate the spatial association between wind power and photovoltaic energy storage equipment based on Mahalanobis distance and convert it into adjacency matrix weight a ij ; S33. Through the graph convolution network, the neighborhood features of the adjacency graph G = (V, E) are convolved, and the graph convolution operation is simplified using the local first-order approximation to generate the final spatiotemporal embedding representation. S4. Construct a time-dependent modeling module for processing the time dependency of data, which specifically includes the following steps: S41. Input the final feature embedding set of wind power and photovoltaic integrated energy storage data, and collect and construct the lag feature embedding set S42. Use long short-term memory network to model the time dependence of wind power and photovoltaic integrated energy storage data to obtain hidden states S43. The final wind power and photovoltaic comprehensive energy storage prediction value of q time steps is obtained through the fully connected layer.

[0007] Preferably, in S1, data related to wind power and photovoltaic integrated energy storage are collected, including photovoltaic power generation history data, wind power generation history data, wind speed data, wind direction data, solar radiation data, light intensity data, cumulative energy storage data and spatial regional characteristic data of wind power and photovoltaic energy storage equipment, and the mean method is used to fill the missing values ​​of wind power and photovoltaic integrated energy storage data. The specific formula is as follows: In the formula, X ave is the imputed mean, n is the number of missing values, x i is a non-missing value.

[0008] Preferably, the Max-Min method is used in S2 to normalize the wind power and photovoltaic integrated energy storage data, and the training set and the test set are divided proportionally. The specific formula is as follows: In the formula, x is the comprehensive energy storage data of wind power and photovoltaic power, x min is the minimum value of wind power and photovoltaic comprehensive energy storage data, x max It is the maximum value among the comprehensive energy storage data of wind power and photovoltaic power.

[0009] Preferably, the normalized wind power and photovoltaic integrated energy storage data X is input in S3 and S31. t ∈R N , where N is the number of wind power and photovoltaic energy storage equipment, a prediction model is constructed to predict the wind power and photovoltaic comprehensive energy storage value in the future q steps based on the observed values ​​of T historical wind power and photovoltaic comprehensive energy storage data. The specific formula is as follows: In the formula, G = (V, E) is the adjacency relationship graph, where V is the wind power and photovoltaic energy storage equipment nodes, E is the spatial dependency relationship, W is the learnable parameter, and F is the prediction model function. is the predicted value of the next q time steps, X1, X2, ..., X T is the observation value of T historical wind power and photovoltaic comprehensive energy storage data input.

[0010] Preferably, by inputting historical energy storage data to predict the energy storage value of future time steps, the model constructs a short-term trend prediction model based on historical observation data, which captures the time correlation in historical data and provides time-dependent characteristics for subsequent spatial modeling, ensuring the contribution of historical trends to future predictions and significantly improving the model's adaptability to time series data.

[0011] Preferably, in S3 and S32, the spatial correlation between wind power and photovoltaic energy storage devices is calculated based on the Mahalanobis distance, and the specific formula is as follows: In the formula, X i , X j are wind power and photovoltaic equipment nodes respectively, S is the covariance matrix of the feature matrix, d ij For node X i With X j The distance between ij Converted to adjacency matrix weight, the specific formula is as follows: In the formula, a ij is the node X in the adjacency matrix i and node X j The edge weight between them, σ is the scale parameter of the Gaussian kernel, which controls the decay rate of the distance to the weight, and exp is the exponential function.

[0012] Preferably, the spatial correlation between wind power and photovoltaic energy storage equipment is calculated based on the Mahalanobis distance, and the distance is converted into the adjacency matrix weight. The Mahalanobis distance calculates the spatial correlation of multidimensional features, and the adjacency matrix quantifies the spatial dependency, so that the model can better capture the mutual influence between devices and enhance the ability to understand spatial heterogeneity.

[0013] Preferably, in S3 and S33, a convolution operation is performed on the neighborhood features of the adjacency relationship graph G=(V, E) through a graph convolution network to generate a convolution operation of each wind power and photovoltaic energy storage device node. The specific formula is as follows: In the formula, L = DA, where L is the standard graph Laplacian matrix, D is the degree matrix, A is the adjacency matrix, and λ max is the maximum eigenvalue of the standard graph Laplacian matrix L, I N is the identity matrix, is the normalized standard graph Laplacian operator, and then the graph convolution operation is implemented through spectral filtering. The specific formula is as follows: X' t =P w (L)X t ; Where u is the filter order, w is u is the weight coefficient, is the Chebyshev polynomial, d is the highest order of the filter, X t is the input wind power and photovoltaic integrated energy storage data, X' t This is the feature representation of wind power and photovoltaic comprehensive energy storage data after graph convolution operation. Then the local first-order approximation is used to simplify the graph convolution operation. The specific formula is as follows: In the formula, w0 and w1 are learnable weight parameters, X t is the input wind power and photovoltaic comprehensive energy storage data, λ max is the maximum eigenvalue of the standard graph Laplacian matrix L, I N is the identity matrix, X' t To simplify the feature representation of wind power and photovoltaic integrated energy storage data after graph convolution operation, the node feature embedding is updated through multi-layer convolution, and the final spatiotemporal embedding representation is generated. The specific formula is as follows: In the formula, is the characteristic vector of the i-th wind power and photovoltaic integrated energy storage data at the initial time t, is the i-th wind power and photovoltaic comprehensive energy storage data at the initial time t, is the feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t at the k-th layer, is the feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t at the k-1 layer, is the feature embedding of the i-th wind power and photovoltaic integrated energy storage data at the Kth layer at time t, f t(k) is the activation function ReLU, W t (k) is the linear transformation weight matrix of neighborhood features, is the neighbor set of the i-th wind power and photovoltaic integrated energy storage data, is the linear transformation weight matrix of its own features, Dense is the fully connected layer, It is the final feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t.

[0014] Preferably, a graph convolutional network is used to perform convolution operations on node features in the adjacency relationship graph, and local first-order approximation is used to simplify the graph convolution calculation, which can effectively extract the spatial features between energy storage devices. Local approximation simplifies the calculation complexity and significantly improves the model's ability to model the spatial dependency relationship between devices, thereby improving the accuracy and efficiency of the prediction results.

[0015] Preferably, the final feature embedding set of wind power and photovoltaic integrated energy storage data input in S4 and S41 is The dimension is N×F, where N is the number of wind power and photovoltaic equipment, and F is the dimension of feature embedding. The feature embeddings at time t and the past p time steps are collected as the lagged feature embedding set. in It is the final feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t.

[0016] Preferably, by collecting embedded features of several past time steps, a lag feature set is formed to capture dynamic changes in time. The lag feature set records the historical status of equipment operation, provides a basis for long-term series prediction, enhances the ability to capture long-term trends, and makes the model more stable in the prediction of long-term series.

[0017] Preferably, in S4 and S42, a long short-term memory network is used to model the time dependency of wind power and photovoltaic integrated energy storage data, and a forget gate is used to control the retention and forgetting operations of historical information in the memory unit. The specific formula is as follows: In the formula, is the set of lagged feature embeddings, for The i-th lagged feature in is embedded in the hidden state of the previous time step t-1, for The i-th lagged feature is embedded in the forget gate weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The forget gate weight matrix, is the bias term of the forget gate, φ1 is the activation function Sigmoid, for The i-th lagged feature in is embedded in the forget gate activation value at time t, and then the new information at the current moment is updated to the memory unit through the input gate. The specific formula is as follows: In the formula, for The i-th lagged feature is embedded in the input gate weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The input gate weight matrix, for The i-th lagged feature is embedded in the memory unit weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The memory unit weight matrix, are the bias terms of the input gate and memory unit respectively, φ2 is the activation function tanh, They are The i-th lag feature is embedded in the activation value of the input gate and memory unit at time t. The state of the memory unit is updated through the combination of the forget gate and the input gate, and the hidden state is updated and output through the output gate. The specific formula is as follows: In the formula, for The i-th lag feature is embedded in the memory cell state at time t, for The i-th lagged feature in is embedded in the output gate activation value at time t, ⊙ is the element-by-element multiplication, for The i-th lagged feature in is embedded in the output gate weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The output gate weight matrix of is the bias term of the output gate, for The i-th lagged feature in is embedded in the hidden state at time t.

[0018] Preferably, a long short-term memory network is used to model the lagged feature set, and the update and memory of time dependence are controlled by the forget gate, input gate and output gate. This can effectively capture long-term and short-term dependencies, reduce the gradient vanishing problem of traditional time series models, and enhance the model's ability to extract time features, especially in scenarios with complex time dependencies.

[0019] Preferably, in S4 and S43, the predicted wind power and photovoltaic comprehensive energy storage prediction value of the i-th wind power and photovoltaic energy storage device in q time steps is obtained through the fully connected layer. The specific formula is as follows: In the formula, is the predicted wind power and photovoltaic comprehensive energy storage forecast value of the i-th wind power and photovoltaic energy storage equipment at time t+q, Dense is the fully connected layer, for The i-th lagged feature in is embedded in the hidden state at time t.

[0020] Preferably, the hidden state of the long short-term memory network is mapped to the target value through a fully connected layer to generate the energy storage prediction value of the future time step. The fully connected layer provides the model with nonlinear mapping capabilities, efficiently converts complex time series features into prediction values, optimizes the prediction performance, and improves the final accuracy of the energy storage prediction.

[0021] In summary, due to the adoption of the technical solution, the beneficial effects of the present invention are as follows: the present invention proposes an E-STENet prediction model, which is applied to the wind power and photovoltaic integrated energy storage prediction scenario, including a spatial correlation modeling module, a time dependence module and a prediction module. The spatial correlation modeling module is used to process the spatial correlation of data, the time dependence module is used to process the time dependence of data, and the prediction module is used to convert the hidden state into the wind power and photovoltaic integrated energy storage prediction results of the future time step. The modules cooperate with each other to achieve accurate prediction of wind power and photovoltaic integrated energy storage. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The figure is a step diagram of a method for predicting comprehensive energy storage supporting wind power and photovoltaic power.

[0023] Figure 2 This is the structure diagram of the E-STENet prediction model.

[0024] Figure 3 This is the fitting effect diagram of wind power and photovoltaic comprehensive energy storage prediction for the E-STENet prediction model. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] See also Figure 1-Figure 3 The present invention provides a technical solution: a method for predicting wind power and photovoltaic integrated energy storage, by proposing an E-STENet prediction model, constructing a spatial correlation modeling module for processing the spatial correlation of data, a time dependency module for processing the time dependency of data, and a prediction module for converting hidden states into wind power and photovoltaic integrated energy storage prediction results for future time steps, and the specific steps are as follows Figure 1 shown.

[0027] Construct the E-STENet prediction model, whose structure is as follows Figure 2 As shown, the specific steps are as follows:

[0028] S1. Collect wind power and photovoltaic integrated energy storage related data, including characteristics and target variables, and pre-process the collected wind power and photovoltaic integrated energy storage data.

[0029] Furthermore, data related to wind power and photovoltaic integrated energy storage are collected, including historical photovoltaic power generation data, historical wind power generation data, wind speed data, wind direction data, solar radiation data, light intensity data, cumulative energy storage data, and spatial regional characteristic data of wind power and photovoltaic energy storage equipment. The mean method is used to fill in the missing values ​​of wind power and photovoltaic integrated energy storage data. The specific formula is as follows: In the formula, X ave is the imputed mean, n is the number of missing values, x i is a non-missing value.

[0030] S2. Use the Max-Min method to normalize the wind power and photovoltaic integrated energy storage data and divide the data set proportionally.

[0031] Furthermore, the Max-Min method is used to normalize the wind power and photovoltaic integrated energy storage data, and the training set and test set are divided proportionally. The specific formula is as follows: In the formula, x is the comprehensive energy storage data of wind power and photovoltaic power, x min is the minimum value of wind power and photovoltaic comprehensive energy storage data, x max It is the maximum value among the comprehensive energy storage data of wind power and photovoltaic power.

[0032] S31, input normalized wind power and photovoltaic comprehensive energy storage data X t ∈R N , based on the observed values ​​of T historical wind power and photovoltaic comprehensive energy storage data, predict the wind power and photovoltaic comprehensive energy storage values ​​in the future q steps.

[0033] Furthermore, the normalized wind power and photovoltaic integrated energy storage data X is input t ∈R N , where N is the number of wind power and photovoltaic energy storage equipment, a prediction model is constructed to predict the wind power and photovoltaic comprehensive energy storage value in the future q steps based on the observed values ​​of T historical wind power and photovoltaic comprehensive energy storage data. The specific formula is as follows: In the formula, G = (V, E) is the adjacency relationship graph, where V is the wind power and photovoltaic energy storage equipment nodes, E is the spatial dependency relationship, W is the learnable parameter, and F is the prediction model function. is the predicted value of the next q time steps, X1,X2,…,X T is the observation value of T historical wind power and photovoltaic comprehensive energy storage data input.

[0034] S32. Calculate the spatial association between wind power and photovoltaic energy storage equipment based on Mahalanobis distance and convert it into adjacency matrix weight a ij .

[0035] Furthermore, the spatial correlation between wind power and photovoltaic energy storage equipment is calculated based on the Mahalanobis distance. The specific formula is as follows: In the formula, X i , X j are wind power and photovoltaic equipment nodes respectively, s is the covariance matrix of the feature matrix, d ij For node X i With X j The distance between ij Converted to adjacency matrix weight, the specific formula is as follows: In the formula, a ij is the node X in the adjacency matrix i and node X j The edge weight between them, σ is the scale parameter of the Gaussian kernel, which controls the decay rate of the distance to the weight, and exp is the exponential function.

[0036] S33. Through the graph convolution network, the neighborhood features of the adjacency graph G = (V, E) are convolved, and the graph convolution operation is simplified using the local first-order approximation to generate the final spatiotemporal embedding representation.

[0037] Furthermore, through the graph convolution network, the neighborhood features of the adjacency relationship graph G = (V, E) are convolved to generate the convolution operation of each wind power and photovoltaic energy storage device node. The specific formula is as follows: In the formula, L = DA, where L is the standard graph Laplacian matrix, D is the degree matrix, A is the adjacency matrix, and λ max is the maximum eigenvalue of the standard graph Laplacian matrix L, I N is the identity matrix, is the normalized standard graph Laplacian operator, and then the graph convolution operation is implemented through spectral filtering. The specific formula is as follows: X' t =P w (L)X t ; Where u is the filter order, w is u is the weight coefficient, is the Chebyshev polynomial, d is the highest order of the filter, X t is the input wind power and photovoltaic integrated energy storage data, X' t This is the feature representation of wind power and photovoltaic comprehensive energy storage data after graph convolution operation. Then the local first-order approximation is used to simplify the graph convolution operation. The specific formula is as follows: In the formula, w0 and w1 are learnable weight parameters, X t is the input wind power and photovoltaic comprehensive energy storage data, λ max is the maximum eigenvalue of the standard graph Laplacian matrix L, I N is the identity matrix, X' t To simplify the feature representation of wind power and photovoltaic integrated energy storage data after graph convolution operation, the node feature embedding is updated through multi-layer convolution, and the final spatiotemporal embedding representation is generated. The specific formula is as follows: In the formula, is the characteristic vector of the i-th wind power and photovoltaic integrated energy storage data at the initial time t, is the i-th wind power and photovoltaic comprehensive energy storage data at the initial time t, is the feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t at the k-th layer, is the feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t at the k-1 layer, is the feature embedding of the i-th wind power and photovoltaic integrated energy storage data at the Kth layer at time t, f t(k) is the activation function ReLU, W t (k) is the linear transformation weight matrix of neighborhood features, is the neighbor set of the i-th wind power and photovoltaic integrated energy storage data, is the linear transformation weight matrix of its own features, Dense is the fully connected layer, It is the final feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t.

[0038] S41. Input the final feature embedding set of wind power and photovoltaic integrated energy storage data, and collect and construct the lag feature embedding set

[0039] Furthermore, the final feature embedding set of wind power and photovoltaic integrated energy storage data is input The dimension is N×F, where N is the number of wind power and photovoltaic equipment, and F is the dimension of feature embedding. The feature embeddings at time t and the past p time steps are collected as the lagged feature embedding set. in It is the final feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t.

[0040] S42. Use long short-term memory network to model the time dependence of wind power and photovoltaic integrated energy storage data to obtain hidden states

[0041] Furthermore, the long short-term memory network is used to model the time dependence of wind power and photovoltaic integrated energy storage data, and the historical information in the memory unit is retained and forgotten through the forget gate. The specific formula is as follows: In the formula, is the set of lagged feature embeddings, for The i-th lagged feature in is embedded in the hidden state of the previous time step t-1, for The i-th lagged feature is embedded in the forget gate weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The forget gate weight matrix, is the bias term of the forget gate, φ1 is the activation function Sigmoid, for The i-th lagged feature in is embedded in the forget gate activation value at time t, and then the new information at the current moment is updated to the memory unit through the input gate. The specific formula is as follows: In the formula, for The i-th lagged feature is embedded in the input gate weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The input gate weight matrix, for The i-th lagged feature is embedded in the memory unit weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The memory unit weight matrix, are the bias terms of the input gate and memory unit respectively, φ2 is the activation function tanh, They are The i-th lag feature is embedded in the activation value of the input gate and memory unit at time t. The state of the memory unit is updated through the combination of the forget gate and the input gate, and the hidden state is updated and output through the output gate. The specific formula is as follows: In the formula, for The i-th lag feature is embedded in the memory cell state at time t, for The i-th lagged feature in is embedded in the output gate activation value at time t, ⊙ is the element-by-element multiplication, for The i-th lagged feature in is embedded in the output gate weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The output gate weight matrix of is the bias term of the output gate, for The i-th lagged feature in is embedded in the hidden state at time t.

[0042] S43. The predicted wind power and photovoltaic comprehensive energy storage prediction values ​​of the final q time steps are obtained through the fully connected layer.

[0043] Furthermore, the predicted wind power and photovoltaic comprehensive energy storage prediction values ​​of the i-th wind power and photovoltaic energy storage equipment at q time steps are obtained through the fully connected layer. The specific formula is as follows: In the formula, is the predicted wind power and photovoltaic comprehensive energy storage forecast value of the i-th wind power and photovoltaic energy storage equipment at time t+q, Dense is the fully connected layer, for The i-th lagged feature in is embedded in the hidden state at time t.

[0044] Furthermore, the E-STENet prediction model is written in Python, the experiment runs on the Windows operating system, Pytorch is selected as the framework in the CUDA11.27 environment, and training is performed on the GeForce RTX 3090. The optimizer uses Adam, the initial learning rate is set to 0.001, the training batch is set to 64, and the training cycle is set to 100. The data set is 30 days of wind power and photovoltaic integrated energy storage related data, which is input into the E-STENet prediction model after preprocessing.

[0045] Furthermore, the E-STENet prediction model realizes the prediction fitting effect of wind power and photovoltaic integrated energy storage as shown in the figure below. Figure 3 As shown in the figure, the horizontal axis is time (days), the vertical axis is total power generation (MWh), the dotted line represents the model prediction data, and the solid line represents the actual observation data. The model prediction data can generally follow the trend of the actual data, especially in the stages from the 2nd to the 6th day and from the 11th to the 15th day, the fluctuation trend of the prediction curve and the actual curve is basically the same, and the model can better capture the rising and falling trends of the total power generation, indicating that it has strong adaptability in dealing with nonlinear changes. The figure covers a 14-day prediction time range. Overall, the model can accurately predict the changing trend of power generation over a long time span, indicating that the model shows high robustness when processing long time series data. The experimental results show that the model performs well in predicting the total power generation of wind power and photovoltaic power generation systems.

Claims

1. A method for predicting wind power and photovoltaic integrated energy storage, characterized in that: The following steps are involved: S1. Collect wind power and photovoltaic integrated energy storage related data, including characteristics and target variables, and pre-process the collected wind power and photovoltaic integrated energy storage data; S2, using the Max-Min method to normalize the wind power and photovoltaic integrated energy storage data and divide the data set proportionally; S3. Construct a spatial correlation modeling module to process the spatial correlation of data, which specifically includes the following steps: S31, input normalized wind power and photovoltaic comprehensive energy storage data X t ∈R N , based on the observed values ​​of T historical wind power and photovoltaic comprehensive energy storage data, predict the wind power and photovoltaic comprehensive energy storage value in the future q steps; S32. Calculate the spatial association between wind power and photovoltaic energy storage equipment based on Mahalanobis distance and convert it into adjacency matrix weight a ij ; S33. Through the graph convolution network, the neighborhood features of the adjacency graph G = (V, E) are convolved, and the graph convolution operation is simplified using the local first-order approximation to generate the final spatiotemporal embedding representation. S4. Construct a time-dependent modeling module for processing the time dependency of data, which specifically includes the following steps: S41. Input the final feature embedding set of wind power and photovoltaic integrated energy storage data, and collect and construct the lag feature embedding set S42. Use long short-term memory network to model the time dependence of wind power and photovoltaic integrated energy storage data to obtain hidden states S43. The final wind power and photovoltaic comprehensive energy storage prediction value of q time steps is obtained through the fully connected layer.

2. A method for predicting wind power and photovoltaic integrated energy storage according to claim 1, characterized in that: The normalized wind power and photovoltaic integrated energy storage data X is input in S31. t ∈RN ， Where N is the number of wind power and photovoltaic energy storage equipment. A prediction model is constructed to predict the wind power and photovoltaic comprehensive energy storage value in the future q steps based on the observed values ​​of T historical wind power and photovoltaic comprehensive energy storage data. The specific formula is as follows: In the formula, G = (V, E) is the adjacency relationship graph, where V is the wind power and photovoltaic energy storage equipment nodes, E is the spatial dependency relationship, W is the learnable parameter, and F is the prediction model function. is the predicted value of the next q time steps, X1, X2, ..., X T is the observation value of T historical wind power and photovoltaic comprehensive energy storage data input.

3. A method for predicting wind power and photovoltaic integrated energy storage according to claim 2, characterized in that: In S32, there is a spatial correlation between wind power and photovoltaic energy storage devices based on the Mahalanobis distance calculation, and the specific formula is as follows: Where, X i , X j are wind power and photovoltaic equipment nodes respectively, S is the covariance matrix of the feature matrix, d ij For node X i With X j The distance between ij Converted to adjacency matrix weight, the specific formula is as follows: In the formula, a ij is the node X in the adjacency matrix i and node X j The edge weight between them, σ is the scale parameter of the Gaussian kernel, which controls the decay rate of the distance to the weight, and exp is the exponential function.

4. A method for predicting wind power and photovoltaic integrated energy storage according to claim 3, characterized in that: In S33, a convolution operation is performed on the neighborhood features of the adjacency relationship graph G=(V, E) through a graph convolution network to generate a convolution operation of each wind power and photovoltaic energy storage device node. The specific formula is as follows: In the formula, L = DA, where L is the standard graph Laplacian matrix, D is the degree matrix, A is the adjacency matrix, and λ max is the maximum eigenvalue of the standard graph Laplacian matrix L, I N is the identity matrix, is the normalized standard graph Laplacian operator, and then the graph convolution operation is implemented through spectral filtering. The specific formula is as follows: X’ t =Pw(L)X t ; Where u is the filter order, w is u is the weight coefficient, is the Chebyshev polynomial, d is the highest order of the filter, X t is the input wind power and photovoltaic integrated energy storage data, X′ t This is the feature representation of wind power and photovoltaic comprehensive energy storage data after graph convolution operation. Then the local first-order approximation is used to simplify the graph convolution operation. The specific formula is as follows: In the formula, w0 and w1 are learnable weight parameters, X t is the input wind power and photovoltaic comprehensive energy storage data, λ max is the maximum eigenvalue of the standard graph Laplacian matrix L, I N is the identity matrix, X″ t To simplify the feature representation of wind power and photovoltaic integrated energy storage data after graph convolution operation, the node feature embedding is updated through multi-layer convolution, and the final spatiotemporal embedding representation is generated. The specific formula is as follows: In the formula, is the characteristic vector of the i-th wind power and photovoltaic integrated energy storage data at the initial time t, is the i-th wind power and photovoltaic comprehensive energy storage data at the initial time t, is the feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t at the k-th layer, is the feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t at the k-1 layer, is the feature embedding of the i-th wind power and photovoltaic integrated energy storage data at the Kth layer at time t, f t (k) is the activation function ReLU, W t (k) is the linear transformation weight matrix of neighborhood features, is the neighbor set of the i-th wind power and photovoltaic integrated energy storage data, is the linear transformation weight matrix of its own features, Dense is the fully connected layer, It is the final feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t.

5. A method for predicting wind power and photovoltaic integrated energy storage according to claim 4, characterized in that: The final feature embedding set of wind power and photovoltaic integrated energy storage data input in S41 The dimension is N×F, where N is the number of wind power and photovoltaic equipment, and F is the dimension of feature embedding. The feature embeddings at time t and the past p time steps are collected as the lagged feature embedding set. in It is the final feature embedding of the i-th wind power and photovoltaic integrated energy storage data at time t.

6. A method for predicting wind power and photovoltaic integrated energy storage according to claim 5, characterized in that: In S42, a long short-term memory network is used to model the time dependency of wind power and photovoltaic integrated energy storage data, and the historical information in the memory unit is retained and forgotten by controlling the forget gate. The specific formula is as follows: In the formula, is the set of lagged feature embeddings, for The i-th lagged feature in is embedded in the hidden state of the previous time step t-1, for The i-th lagged feature is embedded in the forget gate weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The forget gate weight matrix, is the bias term of the forget gate, φ1 is the activation function Sigmoid, for The i-th lagged feature in is embedded in the forget gate activation value at time t, and then the new information at the current moment is updated to the memory unit through the input gate. The specific formula is as follows: In the formula, for The i-th lagged feature is embedded in the input gate weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The input gate weight matrix, for The i-th lagged feature is embedded in the memory unit weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The memory unit weight matrix, are the bias terms of the input gate and memory unit respectively, φ2 is the activation function tanh, They are The i-th lag feature is embedded in the activation value of the input gate and memory unit at time t. The state of the memory unit is updated through the combination of the forget gate and the input gate, and the hidden state is updated and output through the output gate. The specific formula is as follows: In the formula, for The i-th lag feature is embedded in the memory cell state at time t, for The i-th lagged feature in is embedded in the output gate activation value at time t, ⊙ is the element-by-element multiplication, for The i-th lagged feature in is embedded in the output gate weight matrix of the current time step, for The i-th lagged feature is embedded in the hidden state of the previous time step The output gate weight matrix of is the bias term of the output gate, for The i-th lagged feature in is embedded in the hidden state at time t.

7. A method for predicting wind power and photovoltaic integrated energy storage according to claim 6, characterized in that: In S43, the predicted wind power and photovoltaic comprehensive energy storage prediction value of the i-th wind power and photovoltaic energy storage device at q time steps is obtained through the fully connected layer. The specific formula is as follows: In the formula, is the predicted wind power and photovoltaic comprehensive energy storage forecast value of the i-th wind power and photovoltaic energy storage equipment at time t+q, Dense is the fully connected layer, for The i-th lagged feature in is embedded in the hidden state at time t.

8. A method for predicting wind power and photovoltaic integrated energy storage according to claim 1, characterized in that: In order to predict the wind power and photovoltaic integrated energy storage, relevant data on wind power and photovoltaic integrated energy storage are collected, including historical photovoltaic power generation data, historical wind power generation data, wind speed data, wind direction data, solar radiation data, light intensity data, cumulative energy storage data and spatial area characteristic data of wind power and photovoltaic energy storage equipment. The collected relevant data are preprocessed to ensure that there are no missing values ​​in the data. Then the processed data are divided into training sets and test sets for training and evaluating the performance of wind power and photovoltaic integrated energy storage prediction models.

Images