Solar photovoltaic power generation prediction method and distributed photovoltaic power generation charging pile system

Abstract

The invention relates to a solar photovoltaic power generation prediction method and a distributed photovoltaic power generation charging pile system, and belongs to the technical field of photovoltaic power generation. Comprising the following steps: acquiring a satellite remote sensing cloud layer image and photovoltaic sequence historical data; inputting the satellite remote sensing cloud layer image into a complex cloud layer feature capturing module to obtain remote sensing cloud layer image features; photovoltaic sequence historical data are input into a photovoltaic perception attention module to obtain photovoltaic perception attention features; and performing feature fusion on the remote sensing cloud layer image features and the photovoltaic perception attention features to obtain photovoltaic multi-modal fusion features, then performing multi-dimensional information capture to obtain photovoltaic three-dimensional attention features, and obtaining a predicted photovoltaic power generation prediction quantity through a photovoltaic power generation prediction module. The distributed photovoltaic power generation charging pile system comprises a photovoltaic power generation module, an energy storage module, an electric quantity monitoring module, a power distribution control module, a charging pile module, a communication module, a power utilization management module and the like. According to the invention, the prediction accuracy of the model and the robustness of a complex scene can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power generation, and in particular relates to a solar photovoltaic power generation prediction method and a distributed photovoltaic power generation charging pile system. Background Art

[0002] Solar photovoltaic power generation forecasting technology dates back to the rise of the new energy industry in the 20th century. Early photovoltaic power generation forecasting methods primarily relied on statistical analysis of historical power generation data and simple meteorological parameters, which presented significant limitations. Early models assumed a linear relationship between power generation variables, whereas photovoltaic power generation is affected by nonlinear factors such as irradiance-temperature nonlinear coupling and sudden cloud cover changes. At the data collection level, traditional methods relied heavily on limited parameters such as irradiance and cloud cover provided by a single ground-based meteorological station. Furthermore, meteorological data were generally updated infrequently, making them unable to capture sudden irradiance changes caused by short-term cloud movement.

[0003] With the development of computer analysis technology, early researchers primarily used machine learning algorithms such as linear regression and support vector machines to predict photovoltaic power generation. However, these methods have limited ability to model the complex nonlinear relationships in photovoltaic power generation. Furthermore, traditional models lack the ability to model spatiotemporal correlations, making it impossible to analyze cloud migration trajectories using satellite remote sensing imagery and construct short-term solar photovoltaic power generation forecasting models. Consequently, existing short-term solar photovoltaic power generation forecasting methods suffer from low accuracy and poor generalization.

[0004] Therefore, the present invention proposes a solar photovoltaic power generation prediction method and a distributed photovoltaic power generation charging pile system to solve the problems of low power generation prediction accuracy and poor generalization ability of existing solar photovoltaic power generation prediction models. Summary of the Invention

[0005] In order to solve the problems of low accuracy and insufficient multimodal information fusion capability of existing photovoltaic power generation prediction methods under complex weather conditions, the present invention provides a solar photovoltaic power generation prediction method and a distributed photovoltaic power generation charging pile system.

[0006] To achieve the above-mentioned purpose, the present invention is implemented through the following technical solutions: The present invention provides a solar photovoltaic power generation prediction method, comprising the following steps: S1. Acquire satellite remote sensing cloud images and historical photovoltaic series data; S2. Build a complex cloud feature capture module to capture the satellite remote sensing cloud images of the relevant area and the corresponding time Input to the complex cloud feature capture module, extract the features of the global and local information of the image, and output the remote sensing cloud image features ; S3. Build a photovoltaic perception attention module to convert photovoltaic sequence historical data Input to the photovoltaic perception attention module, after spatial and frequency domain feature extraction operations, output photovoltaic perception attention features ; S4. Construct photovoltaic multimodal feature fusion module to integrate remote sensing cloud image features Attention features with photovoltaic perception Input to the photovoltaic multimodal feature fusion module, after feature fusion operation, output photovoltaic multimodal fusion features ; S5. Construct a photovoltaic three-dimensional attention module to integrate photovoltaic multimodal features Input to the photovoltaic 3D attention module to capture multi-dimensional information and output photovoltaic 3D attention features ; S6. Construct a photovoltaic power generation prediction module and integrate photovoltaic three-dimensional attention features Input into the photovoltaic power generation prediction module to obtain the predicted photovoltaic power generation .

[0007] Furthermore, step S2 specifically includes: The complex cloud feature capture module includes a 1D convolution layer, a reshaping layer, a layer normalization layer, and a cloud convolution attention module; the convolution kernel size of the 1D convolution layer is The cloud convolution attention module includes a first branch, a second branch, and a third branch in parallel; S21. Satellite remote sensing cloud images of relevant areas and corresponding times Input into the complex cloud feature capture module and obtain the first convolution feature through the 1D convolution layer ; The first convolution feature After the reshape layer, the features spatial dimension Merge to sequence length Get the first reshape feature ; The first remodeling feature After the layer normalization layer is processed, the first layer normalized features are obtained. ; The formula is as follows: , , , in, Indicates the height of satellite remote sensing cloud images, Indicates the width of satellite remote sensing cloud images, Indicates the number of channels of satellite remote sensing cloud images, Indicates the number of channels after the satellite remote sensing cloud image is processed. Indicates that the convolution kernel size is The 1D convolution layer operation, represents a reshape operation, represents the transpose of the tensor matrix, Representation layer normalization operation; S22. Normalize the layer features Input to the cloud convolution attention module, in the first branch, the feature After slicing, the feature tensor of the first half of the number of channels is extracted to obtain the first slice feature. ; The first slice feature The channels are expanded to times, and get the first extended feature ; First extended feature After activation function Processing to obtain global features ; The formula is as follows: , , , in, Represents the feature The first half of the channels are sliced. and denote the first and second learnable parameters in the linear layer, respectively. express Activation function operation; S23. In the second branch, the characteristics Extract features through slicing operation The feature tensor of the last half of the number of channels, outputting the second slice feature , the second slice feature The channels are expanded to times, and obtain the second extended feature The second extended feature After the convolution kernel size is The 1D convolution layer and the convolution kernel size are 1D depth separable convolution layer to obtain local features The local features After layer normalization, the second layer normalized features are obtained ; The formula is as follows: , , , , in, Represents the feature The last half of the channels are sliced. and denote the third and fourth learnable parameters in the linear layer, respectively. Indicates that the convolution kernel size is 1D convolutional layer, Indicates that the convolution kernel size is 1D depth-wise separable convolutional layer; S24. In the third branch, the characteristics After the linear layer, the channel is expanded to times, and obtain the third extended feature ; The third extended feature After the reshaping operation, the spatial dimension is restored to , and obtain the second reshape feature The second remodeling feature After the convolution kernel size is 2D depth-separable convolution layer, activation function And the convolution kernel size is The 2D convolution layer is processed to obtain the second convolution feature ; The formula is as follows: , , , in, and denote the fifth and sixth learnable parameters in the linear layer, respectively. Indicates that the feature The spatial dimension is restored to Operation; Indicates that the convolution kernel size is 2D convolutional layer, Indicates that the convolution kernel size is 2D depth-wise separable convolutional layer; S25. Global features With the second layer normalized features Perform product fusion to obtain the first fusion feature ; The first fusion feature With the second convolution feature Perform splicing and fusion to obtain cloud convolution attention features ; Said satellite remote sensing cloud image Convolutional attention features with cloud layers Perform stitching and fusion to obtain remote sensing cloud image features ; The formula is as follows: , , , in, represents the feature concatenation operation, Represents a product fusion operation.

[0008] Furthermore, step S3 specifically includes: The photovoltaic perception attention module includes a photovoltaic frequency domain attention module and a photovoltaic space attention module; S31. Photovoltaic series historical data After convolution kernel size 1D convolution layer, and after Activation function processing, output of the third convolution feature , the formula is as follows: , in, Indicates that the convolution kernel size is 1D convolution layer; the third convolution feature The photovoltaic frequency domain attention module is used to capture the global low-frequency features and obtain the low-frequency features. ; High-frequency features are calculated by residual method ; For low frequency features and high-frequency features Perform weighted fusion to obtain photovoltaic frequency domain attention features ; The formula is as follows: , , , in, Represented as a feature The sequence length, ; and Represents the first learnable weight parameter and the second learnable weight parameter obtained through back propagation during model training; Represents the multiplication operation of matrix elements; the photovoltaic frequency domain attention feature After the convolution kernel size is The 1D convolution layer obtains the fourth convolution feature ; The fourth convolution feature After the slicing operation, the feature tensor of the first half of the number of channels is extracted to obtain the third slice feature ; The fourth convolution feature After the slicing operation, the feature tensor with half the number of channels is extracted. The fourth slice feature is obtained ; The formula is as follows: , , , in, Representation characteristics The sequence length, Representation characteristics The number of channels; Specifically refers to Slice along the first 1 / 2 of the channel dimension; Specifically refers to Slice along the last 1 / 2 of the channel dimension; S32. In the photovoltaic spatial attention module, the third slice feature After the global average pooling operation, the convolution kernel size is 1D depthwise separable convolution and Activation function to obtain the fifth convolution feature ; The fifth convolution feature With the third slice feature Perform splicing and fusion to obtain the second fusion feature ; After global average pooling operation, the convolution kernel size is 1D depthwise separable convolution and Activation function to obtain the sixth convolution feature ; The sixth convolution feature With the second fusion feature Perform splicing and fusion to obtain the third fusion feature ; The third fusion feature With the third slice feature Add element by element to get the first comprehensive feature ; The formula is as follows: , , , , , in, express activation function, represents the global average pooling operation, represents the element-by-element addition operation, and Indicates that the convolution kernel size is 1D depth-wise separable convolution and convolution kernel size is 1D depth separable convolution; similarly, the fourth slice feature After the third slice feature The same operation steps are used to obtain the second comprehensive feature ; The first comprehensive feature and the second comprehensive characteristic Perform splicing and fusion to obtain photovoltaic perception attention , the formula is: .

[0009] Furthermore, step S4 specifically includes: The remote sensing cloud image features and photovoltaic perception attention features Input to the photovoltaic multimodal feature fusion module, remote sensing cloud image features After the convolution kernel size is The 2D transposed convolution layer pairs features Perform high-resolution restoration and use the ReLU activation function for nonlinear processing to output the seventh convolution feature ; Photovoltaic perception attention features With the seventh convolution feature Perform feature splicing operation in the channel dimension to obtain splicing features ; The splicing feature is processed by convolution kernel size 2D depthwise separable convolution, and then use The activation function performs nonlinear weighted processing to obtain photovoltaic multimodal fusion features ; The formula is as follows: , , , in, represents the ReLU activation function, Indicates that the convolution kernel size is 2D transposed convolutional layer, Indicates that the convolution kernel size is 2D depthwise separable convolution.

[0010] Furthermore, step S5 specifically includes: The photovoltaic multimodal fusion features Input into the photovoltaic 3D attention module, the feature Use along the channel dimension The operation is divided into 4 groups of sub-features along the channel, and the number of channels of each sub-feature is ,in Features The number of channels is expressed as follows: , in, represents the feature splitting operation along the channel, Respectively represent the first initial sub-feature, the second initial sub-feature, the third initial sub-feature, and the fourth initial sub-feature; the first initial sub-feature Retain the sub-features through skip connections and obtain the first sub-feature ; The second initial sub-feature After the convolution kernel size is 2D convolutional layer and use The activation function performs nonlinear processing and then passes through the convolution kernel with a size of The 2D depth separable convolution layer obtains the second sub-feature The second sub-feature With the third initial sub-feature Splicing and fusion to obtain the fourth fusion feature , the fourth fusion feature After the convolution kernel size is 2D convolutional layer and use The activation function performs nonlinear processing and then passes through the convolution kernel with a size of The 2D depth separable convolution layer obtains the third sub-feature The third sub-feature With the fourth initial sub-feature Perform splicing and fusion to obtain the fifth fusion feature , the fifth fusion feature After the convolution kernel size is 2D convolutional layer and use The activation function performs nonlinear processing and then passes through the convolution kernel with a size of The 2D depth separable convolution layer obtains the fourth sub-feature ; The formula is as follows: , , , , , in, Indicates that the convolution kernel size is 2D convolutional layer, and Fenbie indicates the convolution kernel size is The 2D convolution layer and the convolution kernel size are 2D depth-wise separable convolutional layer, Indicates that the convolution kernel size is 2D depth-separable convolution layer; sub-features Splicing fusion, after convolution kernel size is 2D convolution layer to obtain photovoltaic three-dimensional attention , the formula is as follows: , in, Indicates that the convolution kernel size is 2D convolutional layer.

[0011] Furthermore, step S6 specifically includes: Focusing on photovoltaics Input into photovoltaic power generation prediction module, and transform the characteristics Flattened into a one-dimensional vector, then passed through the fully connected layer and Activation function calculation predicted photovoltaic power generation , the formula is as follows: , in, represents a full connection operation, represents a vector flattening operation, and They respectively represent the third learnable weight parameter and the fourth learnable weight parameter obtained by the model through back propagation.

[0012] Furthermore, the mean square error loss is used as the loss function, and the calculation process of the loss function is: , in, Represents the calculated value of the loss function, is the number of data samples, Expressed as the actual photovoltaic power value, Indicates that the sample is The photovoltaic power value predicted by the model at time .

[0013] The present invention also provides a distributed photovoltaic power generation charging pile system, including a photovoltaic power generation module, an energy storage module, a power monitoring module, a power distribution control module, a charging pile module, a communication module and a power management module; The output end of the photovoltaic power generation module is connected to the input end of the energy storage module. The input end of the power monitoring module is connected; the output end of the power monitoring module is connected to the input end of the power distribution control module, the output end of the energy storage module is connected to the input end of the charging pile module, the input end of the communication module is respectively connected to the output end of the power monitoring module and the output end of the power distribution control module, and the output end of the communication module is connected to the cloud; the output end of the power distribution control module is connected to the user's power consumption system, the input end of the charging pile module is also connected to the output end of the municipal power grid, the input end of the power consumption management module is connected to the output end of the power distribution control module and the output end of the municipal power grid, and the output end of the power consumption management module is connected to the input end of the charging pile module and the power consumption system; the power distribution control module is used to execute the solar photovoltaic power generation prediction method.

[0014] The advantages of the present invention are: The proposed method fully utilizes multi-source heterogeneous information, including satellite remote sensing meteorological imagery and photovoltaic system perception data. Using a complex cloud feature capture module, the method performs multi-scale feature analysis on satellite remote sensing imagery, deeply exploring the dynamic relationship between cloud morphological evolution and light intensity, significantly improving the ability to extract key features under complex meteorological conditions. The designed photovoltaic perception attention module utilizes a joint modeling mechanism in the time, space, and frequency domains to effectively capture the long- and short-term temporal dependencies of photovoltaic power generation data, enhancing the dynamic perception of the photovoltaic system's operating status. The proposed photovoltaic multimodal feature fusion module complements the fusion of heterogeneous data features. The photovoltaic three-dimensional attention module in this method fully captures the deep-level information of multimodal data, significantly improving the model's robustness in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0016] Figure 1 is a flow chart of the steps of the method of the present invention; Figure 2 This is a system block diagram of the distributed photovoltaic power generation charging pile system proposed by the present invention; Figure 3 This is a structural diagram of the distributed photovoltaic power generation charging pile system proposed in the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments derived by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0018] Example 1 In this embodiment, Figure 1 As shown, the present invention provides a solar photovoltaic power generation prediction method, which specifically includes the following steps: S1. Acquire satellite remote sensing cloud images and historical photovoltaic series data.

[0019] S2. Build a complex cloud feature capture module to capture the satellite remote sensing cloud images of the relevant area and the corresponding time Input to the complex cloud feature capture module, extract the features of the global and local information of the image, and output the remote sensing cloud image features .

[0020] Specifically, the complex cloud feature capture module includes a 1D convolution layer, a reshaping layer, a layer normalization layer, and a cloud convolution attention module; the convolution kernel size of the 1D convolution layer is The cloud convolution attention module includes a first branch, a second branch, and a third branch in parallel; S21. Satellite remote sensing cloud images of relevant areas and corresponding times Input into the complex cloud feature capture module and obtain the first convolution feature through the 1D convolution layer ; The first convolution feature After the reshape layer, the features spatial dimension Merge to sequence length Get the first reshape feature ; The first remodeling feature After the layer normalization layer is processed, the first layer normalized features are obtained. ; The formula is as follows: , , , in, Indicates the height of satellite remote sensing cloud images, Indicates the width of satellite remote sensing cloud images, Indicates the number of channels of satellite remote sensing cloud images, Indicates the number of channels after the satellite remote sensing cloud image is processed. Indicates that the convolution kernel size is The 1D convolution layer operation, represents a reshape operation, represents the transpose of the tensor matrix, Representation layer normalization operation; S22. Normalize the layer features Input to the cloud convolution attention module, in the first branch, the feature After slicing, the feature tensor of the first half of the number of channels is extracted to obtain the first slice feature. ; The first slice feature The channels are expanded to times, and get the first extended feature ; First extended feature After activation function Processing to obtain global features ; The formula is as follows: , , , in, Represents the feature The first half of the channels are sliced. and denote the first and second learnable parameters in the linear layer, respectively. express Activation function operation; S23. In the second branch, the characteristics Extract features through slicing operation The feature tensor of the last half of the number of channels, outputting the second slice feature , the second slice feature The channels are expanded to times, and obtain the second extended feature The second extended feature After the convolution kernel size is The 1D convolution layer and the convolution kernel size are 1D depth separable convolution layer to obtain local features The local features After layer normalization, the second layer normalized features are obtained ; The formula is as follows: , , , , in, Represents the feature The last half of the channels are sliced. and denote the third and fourth learnable parameters in the linear layer, respectively. Indicates that the convolution kernel size is 1D convolutional layer, Indicates that the convolution kernel size is 1D depth-wise separable convolutional layer; S24. In the third branch, the characteristics After the linear layer, the channel is expanded to times, and obtain the third extended feature ; The third extended feature After the reshaping operation, the spatial dimension is restored to , and obtain the second reshape feature The second remodeling feature After the convolution kernel size is 2D depth-separable convolution layer, activation function And the convolution kernel size is The 2D convolution layer is processed to obtain the second convolution feature ; The formula is as follows: , , , in, and denote the fifth and sixth learnable parameters in the linear layer, respectively. Indicates that the feature The spatial dimension is restored to Operation; Indicates that the convolution kernel size is 2D convolutional layer, Indicates that the convolution kernel size is 2D depth-wise separable convolutional layer; S25. Global features With the second layer normalized features Perform product fusion to obtain the first fusion feature ; The first fusion feature With the second convolution feature Perform splicing and fusion to obtain cloud convolution attention features ; Said satellite remote sensing cloud image Convolutional attention features with cloud layers Perform stitching and fusion to obtain remote sensing cloud image features ; The formula is as follows: , , , in, represents the feature concatenation operation, Represents a product fusion operation.

[0021] S3. Build a photovoltaic perception attention module to convert photovoltaic sequence historical data Input to the photovoltaic perception attention module, after spatial and frequency domain feature extraction operations, output photovoltaic perception attention features .

[0022] Specifically, the photovoltaic perception attention module includes a photovoltaic frequency domain attention module and a photovoltaic space attention module; S31. Photovoltaic series historical data After convolution kernel size 1D convolution layer, and after Activation function processing, output of the third convolution feature , the formula is as follows: , in, Indicates that the convolution kernel size is 1D convolution layer; the third convolution feature The photovoltaic frequency domain attention module is used to capture the global low-frequency features and obtain the low-frequency features. ; High-frequency features are calculated by residual method ; For low frequency features and high-frequency features Perform weighted fusion to obtain photovoltaic frequency domain attention features ; The formula is as follows: , , , in, Represented as a feature The sequence length, ; and Represents the first learnable weight parameter and the second learnable weight parameter obtained through back propagation during model training; Represents the multiplication operation of matrix elements; the photovoltaic frequency domain attention feature After the convolution kernel size is The 1D convolution layer obtains the fourth convolution feature ; The fourth convolution feature After the slicing operation, the feature tensor of the first half of the number of channels is extracted to obtain the third slice feature ; The fourth convolution feature After the slicing operation, the feature tensor of half the number of channels is extracted to obtain the fourth slice feature ; The formula is as follows: , , , in, Representation characteristics The sequence length, Representation characteristics The number of channels; Specifically refers to Slice along the first 1 / 2 of the channel dimension; Specifically refers to Slice along the last 1 / 2 of the channel dimension; S32. In the photovoltaic spatial attention module, the third slice feature After the global average pooling operation, the convolution kernel size is 1D depthwise separable convolution and Activation function to obtain the fifth convolution feature ; The fifth convolution feature With the third slice feature Perform splicing and fusion to obtain the second fusion feature ; After global average pooling operation, the convolution kernel size is 1D depthwise separable convolution and Activation function to obtain the sixth convolution feature ; The sixth convolution feature With the second fusion feature Perform splicing and fusion to obtain the third fusion feature ; The third fusion feature With the third slice feature Add element by element to get the first comprehensive feature ; The formula is as follows: , , , , , in, express activation function, represents the global average pooling operation, represents the element-by-element addition operation, and Indicates that the convolution kernel size is 1D depth-wise separable convolution and convolution kernel size is 1D depth separable convolution; similarly, the fourth slice feature After the third slice feature The same operation steps are used to obtain the second comprehensive feature ; The first comprehensive feature and the second comprehensive characteristic Perform splicing and fusion to obtain photovoltaic perception attention , the formula is: .

[0023] S4. Construct photovoltaic multimodal feature fusion module to integrate remote sensing cloud image features Attention features with photovoltaic perception Input to the photovoltaic multimodal feature fusion module, after feature fusion operation, output photovoltaic multimodal fusion features .

[0024] Specifically, the remote sensing cloud image features and photovoltaic perception attention features Input to the photovoltaic multimodal feature fusion module, remote sensing cloud image features After the convolution kernel size is The 2D transposed convolution layer pairs features Perform high-resolution restoration and use the ReLU activation function for nonlinear processing to output the seventh convolution feature ; Photovoltaic perception attention features With the seventh convolution feature Perform feature splicing operation in the channel dimension to obtain splicing features ; The splicing feature is processed by convolution kernel size 2D depthwise separable convolution, and then use The activation function performs nonlinear weighted processing to obtain photovoltaic multimodal fusion features ; The formula is as follows: , , , in, represents the ReLU activation function, Indicates that the convolution kernel size is 2D transposed convolutional layer, Indicates that the convolution kernel size is 2D depthwise separable convolution.

[0025] S5. Construct photovoltaic three-dimensional attention module to integrate photovoltaic multimodal features Input to the photovoltaic 3D attention module to capture multi-dimensional information and output photovoltaic 3D attention features .

[0026] Specifically, the photovoltaic multimodal fusion feature Input into the photovoltaic 3D attention module, the feature Use along the channel dimension The operation is divided into 4 groups of sub-features along the channel, and the number of channels of each sub-feature is ,in Characterized by The number of channels is expressed as follows: , in, represents the feature splitting operation along the channel, Respectively represent the first initial sub-feature, the second initial sub-feature, the third initial sub-feature, and the fourth initial sub-feature; the first initial sub-feature Retain the sub-features through skip connections and obtain the first sub-feature ; The second initial sub-feature After the convolution kernel size is 2D convolutional layer and use The activation function performs nonlinear processing and then passes through the convolution kernel with a size of The 2D depth separable convolution layer obtains the second sub-feature The second sub-feature With the third initial sub-feature Splicing and fusion to obtain the fourth fusion feature , the fourth fusion feature After the convolution kernel size is 2D convolutional layer and use The activation function performs nonlinear processing and then passes through the convolution kernel with a size of The 2D depth separable convolution layer obtains the third sub-feature The third sub-feature With the fourth initial sub-feature Perform splicing and fusion to obtain the fifth fusion feature , the fifth fusion feature After the convolution kernel size is 2D convolutional layer and use The activation function performs nonlinear processing and then passes through the convolution kernel with a size of The 2D depth separable convolution layer obtains the fourth sub-feature ; The formula is as follows: , , , , , in, Indicates that the convolution kernel size is 2D convolutional layer, and Fenbie indicates the convolution kernel size is The 2D convolution layer and the convolution kernel size are 2D depth-wise separable convolutional layer, Indicates that the convolution kernel size is 2D depth-separable convolution layer; sub-features Splicing fusion, after convolution kernel size is 2D convolution layer to obtain photovoltaic three-dimensional attention , the formula is as follows: , in, Indicates that the convolution kernel size is 2D convolutional layer.

[0027] S6. Construct a photovoltaic power generation prediction module and integrate photovoltaic three-dimensional attention features Input into the photovoltaic power generation prediction module to obtain the predicted photovoltaic power generation .

[0028] Specifically, the photovoltaic three-dimensional attention Input into photovoltaic power generation prediction module, and transform the characteristics Flattened into a one-dimensional vector, then passed through the fully connected layer and Activation function calculation predicted photovoltaic power generation , the formula is as follows: , in, represents a full connection operation, represents a vector flattening operation, and They respectively represent the third learnable weight parameter and the fourth learnable weight parameter obtained by the model through back propagation.

[0029] Specifically, the mean square error loss is used as the loss function, and the calculation process of the loss function is: , in, Represents the value of the calculated loss function, is the number of data samples, Expressed as the actual photovoltaic power value, Indicates that the sample is The photovoltaic power value predicted by the model at time .

[0030] Example 2 In order to verify the effectiveness of the solar photovoltaic power generation prediction method proposed in the present invention in the field of photovoltaic power generation prediction, the proposed solar photovoltaic power generation prediction method and the existing photovoltaic power generation prediction method are verified under the same experimental conditions, and the photovoltaic power generation prediction results of each model in the comparative experiment are compared and analyzed, which verifies the effectiveness of the solar photovoltaic power generation prediction method proposed in the present invention in the field of photovoltaic power generation prediction.

[0031] In the comparative test of the present invention, three mainstream photovoltaic power generation timing prediction model methods were selected as comparative experimental models with the proposed method. The three comparative models used in the experiment are: 1. The LSTM model is based on the recurrent neural network architecture, which can effectively capture the long-term dependencies in time series data, but is prone to the gradient vanishing problem when processing long sequences, and has low computational efficiency; 2. The Transformer model adopts the self-attention mechanism, is good at modeling the global dependencies of sequence data and supports parallel computing, but its computational complexity increases quadratically with the sequence length, and has high requirements on the amount and quality of training data; 3. The ViT model applies the Transformer architecture to image block sequence modeling, and performs well in global feature extraction, but lacks the ability to fine-tune modeling of local spatial details.

[0032] In the comparative experiment of the proposed solar photovoltaic power generation prediction method, three experimental indicators are used to evaluate the performance of the model: RMSE (Root Mean Square Error Percentage) measures the absolute error magnitude of the model prediction by calculating the root mean square of the squared difference between the predicted value and the actual value. The smaller its value, the higher the prediction accuracy; MAE (Mean Absolute Error Percentage) intuitively reflects the actual scale of the prediction error by calculating the average of the absolute errors between the predicted value and the actual value; MAPE (Mean Absolute Percentage Error) calculates the average of the absolute values of the relative errors between the predicted value and the actual value. Among them, RMSE and MAE are used to directly evaluate the prediction errors of the model at different time steps (0 - 4 hours), reflecting the model's ability to capture short-term and medium-term photovoltaic power fluctuations; MAPE is used to measure the relative error stability of the model under different weather conditions (sunny, cloudy, overcast).

[0033] The dataset used in the experiment is the existing dataset - Multimodal Photovoltaic Power Prediction Dataset. This dataset is jointly collected through satellite remote sensing technology and ground monitoring equipment, covering the centralized photovoltaic power station in Australia (BP Station) and 9 distributed photovoltaic sites in China (S1 - S9 Stations). The data content includes time series data such as photovoltaic power, global horizontal irradiance (GHI), and solar zenith angle with high spatio-temporal resolution, as well as dual-channel remote sensing images of infrared spectrum (IR) and shortwave radiation (SWR) provided by the Japanese Himawari-8 satellite. The dataset is divided into three types of scenarios according to weather types: sunny (CSI > 0.9), cloudy (0.3 ≤ CSI ≤ 0.9), and overcast (0 < CSI < 0.3), covering multi-season continuous observation data from 2020 - 2022 (BP Station) and 2022 - 2023 (Chinese sites). In the experiment, the dataset is divided into training set, test set, and validation set according to the ratio of 8:1:1. Among them, the BP Station contains 66,365 samples, and each Chinese site contains 31,395 samples.

[0034] The experimental results of the solar photovoltaic power generation prediction method and the experimental comparison model proposed in this invention on the data of BP Station and S1 - S9 Stations are shown in Table 1 and Table 2.

[0035] Table 1 Comparison results of models in the BP Station dataset Table 2 Comparison results of models in the S1 - S9 Stations dataset In the BP station data, the RMSE, MAE and MAPE indicators of the proposed method were 19.13%, 9.82% and 8.47% respectively. All three indicators were the best among similar models. Compared with the second-best model Transformer, the performance improvements were 0.93%, 0.29% and 0.79% respectively; in the S1-S9 station data, the RMSE, MAE and MAPE indicators of the proposed method were 58.07%, 36.54% and 27.69%, which were the best indicators among the compared models.

[0036] Example 3 like Figure 2-Figure 3 As shown, this embodiment also proposes a distributed photovoltaic power generation charging pile system, including a photovoltaic power generation module, an energy storage module, a power monitoring module, a power distribution control module, a charging pile module, a communication module and a power management module.

[0037] The output end of the photovoltaic power generation module is connected to the input end of the energy storage module, and the output end of the energy storage module is connected to the input end of the power monitoring module; the output end of the power monitoring module is connected to the input end of the power distribution control module, the output end of the energy storage module is connected to the input end of the charging pile module, the input end of the communication module is respectively connected to the output end of the power monitoring module and the output end of the power distribution control module, and the output end of the communication module is connected to the cloud; the output end of the power distribution control module is connected to the user's power consumption system, the input end of the charging pile module is also connected to the output end of the municipal power grid, the input end of the power consumption management module is connected to the output end of the power distribution control module and the output end of the municipal power grid, and the output end of the power consumption management module is connected to the input end of the charging pile module and the power consumption system; the power distribution control module is used to execute the solar photovoltaic power generation prediction method, realize the prediction of power generation so that it can be better allocated later; it has the characteristics of high efficiency, accuracy and intelligence, and provides a basis and guarantee for the formulation of the entire power generation plan and operation mode.

[0038] The introduction of power generation forecasting has the following advantages: Improve the economic benefits of power plants: By accurately predicting future power generation, we can reasonably arrange power dispatch, avoid power waste or shortage, and improve the economic benefits of power plants.

[0039] Optimize power plant operation and maintenance: Based on power generation forecast results, equipment maintenance plans can be adjusted in a timely manner to improve equipment life and efficiency.

[0040] The photovoltaic power generation module consists of several photovoltaic panels installed on the roof. The photovoltaic power generation module converts solar energy into electricity and stores it in the energy storage module. The energy storage module powers the charging station module and the user's power system. The charging station module charges new energy vehicles. The power system includes lighting, air conditioning, heating, etc., which provide lighting and heating for the user's home. For rural tiled roofs with sloping roofs, the photovoltaic panels are directly installed on the sunny side of the roof, taking advantage of the roof angle and fixing them on the side. For rural bungalows, the photovoltaic panels are fixed using a pressure block counterweight method that does not penetrate the roof.

[0041] The power monitoring module is used to monitor the real-time storage energy of the energy storage module, and is provided with a minimum storage energy threshold and a maximum storage energy threshold. When it is monitored that the storage energy of the energy storage module reaches the minimum storage energy threshold, the power monitoring module transmits the information to the power management module, and the power management module sends a control command to the power distribution control module. The power distribution control module disconnects the energy storage device from supplying energy and starts the mains power supply at the same time. When the power monitoring module monitors that the storage energy of the energy storage module reaches the maximum threshold, the power monitoring module transmits the information to the power management module. The power management module sends a control command to the power distribution control module. The power distribution control module disconnects the mains power supply and starts the energy storage device from supplying energy. The purpose of setting the above-mentioned minimum threshold is that even if a mains power failure occurs and the mains power cannot supply power to the user's home power system and the charging pile module, the energy storage of the minimum threshold can be used for emergency response to ensure the normal operation of the charging pile module and the user's power system. The maximum threshold provides a node that can use the energy storage module for power supply.

[0042] The power distribution control module can predict solar power generation, and through the power distribution control module and the power management module, unified power allocation is carried out for charging piles and power systems in users' homes to achieve effective energy management.

[0043] The communication module is a Bluetooth wireless communication module or a WIFI wireless communication module. The communication module transmits the power information of the energy storage module of the photovoltaic energy storage charging pile system and the power consumption information of the charging pile and the user's power system to the cloud. The user can view it in real time through the mobile phone app, so that the user can promptly understand the status of the charging pile and the power consumption information of the power system from the cloud.

[0044] The distributed photovoltaic power generation charging pile system proposed in this embodiment is a distributed solar power generation-new energy vehicle charging system, which has the following practical significance: 1. Save building energy consumption; 2. Supplement the grid energy structure and alleviate the pressure on the grid during peak electricity consumption periods; 3. Save energy and reduce emissions, and reduce carbon emissions; 4. Enable farmers to increase their income.

[0045] Competitive advantage analysis: The rural market is extremely vast; for the first time, solar power generation is combined with charging piles. The charging piles have high power and short charging time; they are low in price, easy for farmers to accept and easy to promote.

[0046] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A solar photovoltaic power generation prediction method, characterized in that: The following steps are involved: S1. Acquire satellite remote sensing cloud images and historical photovoltaic series data; S2. Construct a complex cloud feature capture module, which includes a 1D convolution layer, a reshaping layer, a layer normalization layer, and a cloud convolution attention module; the convolution kernel size of the 1D convolution layer is The cloud convolution attention module includes a first branch, a second branch and a third branch in parallel; the satellite remote sensing cloud images of the relevant area and the corresponding time are Input to the complex cloud feature capture module, extract the features of the global and local information of the image, and output the remote sensing cloud image features ; S3. Constructing a photovoltaic perception attention module, the photovoltaic perception attention module includes a photovoltaic frequency domain attention module and a photovoltaic space attention module; Input to the photovoltaic perception attention module, after spatial and frequency domain feature extraction operations, output photovoltaic perception attention features ; S4. Construct photovoltaic multimodal feature fusion module to integrate remote sensing cloud image features Attention features with photovoltaic perception Input to the photovoltaic multimodal feature fusion module, after feature fusion operation, output photovoltaic multimodal fusion features ; S5. Construct a photovoltaic three-dimensional attention module to integrate photovoltaic multimodal features Input to the photovoltaic 3D attention module to capture multi-dimensional information and output photovoltaic 3D attention features ; S6. Construct a photovoltaic power generation prediction module and integrate photovoltaic three-dimensional attention features Input into the photovoltaic power generation prediction module to obtain the predicted photovoltaic power generation .

2. The solar photovoltaic power generation prediction method according to claim 1, characterized in that: The 1D convolution layer, reshape layer, and layer normalization layer in step S2 specifically include: Satellite remote sensing cloud images of relevant areas and corresponding times Input into the complex cloud feature capture module and obtain the first convolution feature through the 1D convolution layer ; The first convolution feature After the reshape layer, the features spatial dimension Merge to sequence length Get the first reshape feature ; The first remodeling feature After the layer normalization layer is processed, the first layer normalized features are obtained. ; The formula is as follows: ， ， ， in, Indicates the height of satellite remote sensing cloud images, Indicates the width of satellite remote sensing cloud images, Indicates the number of channels of satellite remote sensing cloud images, Indicates the number of channels after the satellite remote sensing cloud image is processed. Indicates that the convolution kernel size is The 1D convolution layer operation, represents a reshape operation, represents the transpose of the tensor matrix, Representation layer normalization operation.

3. The solar photovoltaic power generation prediction method according to claim 2, characterized in that: The cloud convolution attention module in step S2 specifically includes: Normalize the features of the layer Input to the cloud convolution attention module, in the first branch, the feature After slicing, the feature tensor of the first half of the number of channels is extracted to obtain the first slice feature. ; The first slice feature The channels are expanded to times, and get the first extended feature ; First extended feature After activation function Processing to obtain global features ; The formula is as follows: ， ， ， in, Represents the feature The first half of the channels are sliced. and denote the first and second learnable parameters in the linear layer, respectively. express Activation function operation; In the second branch, the features Extract features through slicing operation The feature tensor of the last half of the number of channels, outputting the second slice feature , the second slice feature The channels are expanded to times, and obtain the second extended feature The second extended feature After the convolution kernel size is The 1D convolution layer and the convolution kernel size are 1D depth separable convolution layer to obtain local features The local features After layer normalization, the second layer normalized features are obtained ; The formula is as follows: ， ， ， ， in, Represents the feature The last half of the channels are sliced. and denote the third and fourth learnable parameters in the linear layer, respectively. Indicates that the convolution kernel size is 1D convolutional layer, Indicates that the convolution kernel size is 1D depth-wise separable convolutional layer; In the third branch, the characteristics After the linear layer, the channel is expanded to times, and obtain the third extended feature ; The third extended feature After the reshaping operation, the spatial dimension is restored to , and obtain the second reshape feature The second remodeling feature After the convolution kernel size is 2D depth-separable convolution layer, activation function And the convolution kernel size is The 2D convolution layer is processed to obtain the second convolution feature ; The formula is as follows: ， ， ， in, and denote the fifth and sixth learnable parameters in the linear layer, respectively. Indicates that the feature The spatial dimension is restored to Operation; Indicates that the convolution kernel size is 2D convolutional layer, Indicates that the convolution kernel size is 2D depth-wise separable convolutional layer; Global Features With the second layer normalized features Perform product fusion to obtain the first fusion feature ; The first fusion feature With the second convolution feature Perform splicing and fusion to obtain cloud convolution attention features ; Said satellite remote sensing cloud image Convolutional attention features with cloud layers Perform stitching and fusion to obtain remote sensing cloud image features ; The formula is as follows: ， ， ， in, represents the feature concatenation operation, Represents a product fusion operation.

4. The solar photovoltaic power generation prediction method according to claim 3, characterized in that: The photovoltaic frequency domain attention module in step S3 specifically includes: Photovoltaic series historical data After convolution kernel size 1D convolution layer, and after Activation function processing, output of the third convolution feature , the formula is as follows: ， in, Indicates that the convolution kernel size is 1D convolution layer; the third convolution feature The photovoltaic frequency domain attention module is used to capture the global low-frequency features and obtain the low-frequency features. ; High-frequency features are calculated by residual method ; For low frequency features and high-frequency features Perform weighted fusion to obtain photovoltaic frequency domain attention features ; The formula is as follows: ， ， ， in, Represented as a feature The sequence length, ; and Represents the first learnable weight parameter and the second learnable weight parameter obtained through back propagation during model training; Represents the multiplication operation of matrix elements; the photovoltaic frequency domain attention feature After the convolution kernel size is The 1D convolution layer obtains the fourth convolution feature ; The fourth convolution feature After the slicing operation, the feature tensor of the first half of the number of channels is extracted to obtain the third slice feature ; The fourth convolution feature After the slicing operation, the feature tensor of half the number of channels is extracted to obtain the fourth slice feature ; The formula is as follows: ， ， ， in, Representation characteristics The sequence length, Representation characteristics The number of channels; Specifically refers to Slice along the first 1 / 2 of the channel dimension; Specifically refers to Slice along the last 1 / 2 of the channel dimension.

5. The solar photovoltaic power generation prediction method according to claim 4, characterized in that: The photovoltaic space attention module in step S3 specifically includes: The third slice feature After the global average pooling operation, the convolution kernel size is 1D depthwise separable convolution and Activation function to obtain the fifth convolution feature ; The fifth convolution feature With the third slice feature Perform splicing and fusion to obtain the second fusion feature ; After global average pooling operation, the convolution kernel size is 1D depthwise separable convolution and Activation function to obtain the sixth convolution feature ; The sixth convolution feature With the second fusion feature Perform splicing and fusion to obtain the third fusion feature ; The third fusion feature With the third slice feature Add element by element to get the first comprehensive feature ; The formula is as follows: ， ， ， ， ， in, express activation function, represents the global average pooling operation, represents the element-by-element addition operation, and Indicates that the convolution kernel size is 1D depth-wise separable convolution and convolution kernel size is 1D depth separable convolution; similarly, the fourth slice feature After the third slice feature The same operation steps are used to obtain the second comprehensive feature ; The first comprehensive feature and the second comprehensive characteristic Perform splicing and fusion to obtain photovoltaic perception attention , the formula is: .

6. The solar photovoltaic power generation prediction method according to claim 5, characterized in that: Step S4 specifically includes: The remote sensing cloud image features and photovoltaic perception attention features Input to the photovoltaic multimodal feature fusion module, remote sensing cloud image features After the convolution kernel size is The 2D transposed convolution layer pairs features Perform high-resolution restoration and use the ReLU activation function for nonlinear processing to output the seventh convolution feature ; Photovoltaic perception attention features With the seventh convolution feature Perform feature splicing operation in the channel dimension to obtain splicing features ; The splicing feature is processed by convolution kernel size 2D depthwise separable convolution, and then use The activation function performs nonlinear weighted processing to obtain photovoltaic multimodal fusion features ; The formula is as follows: ， ， ， in, represents the ReLU activation function, Indicates that the convolution kernel size is 2D transposed convolutional layer, Indicates that the convolution kernel size is 2D depthwise separable convolution.

7. The solar photovoltaic power generation prediction method according to claim 6, characterized in that: Step S5 specifically includes: The photovoltaic multimodal fusion features Input into the photovoltaic 3D attention module, the feature Use along the channel dimension The operation is divided into 4 groups of sub-features along the channel, and the number of channels of each sub-feature is ,in Characterized by The number of channels is expressed as follows: ， in, represents the feature splitting operation along the channel, Respectively represent the first initial sub-feature, the second initial sub-feature, the third initial sub-feature, and the fourth initial sub-feature; the first initial sub-feature Retain the sub-features through skip connections and obtain the first sub-feature ; The second initial sub-feature After the convolution kernel size is 2D convolutional layer and use The activation function performs nonlinear processing and then passes through the convolution kernel with a size of The 2D depth separable convolution layer obtains the second sub-feature The second sub-feature With the third initial sub-feature Splicing and fusion to obtain the fourth fusion feature , the fourth fusion feature After the convolution kernel size is 2D convolutional layer and use The activation function performs nonlinear processing and then passes through the convolution kernel with a size of The 2D depth separable convolution layer obtains the third sub-feature The third sub-feature With the fourth initial sub-feature Perform splicing and fusion to obtain the fifth fusion feature , the fifth fusion feature After the convolution kernel size is 2D convolutional layer and use The activation function performs nonlinear processing and then passes through the convolution kernel with a size of The 2D depth separable convolution layer obtains the fourth sub-feature ; The formula is as follows: ， ， ， ， ， in, Indicates that the convolution kernel size is 2D convolutional layer, and Fenbie indicates the convolution kernel size is The 2D convolution layer and the convolution kernel size are 2D depth-wise separable convolutional layer, Indicates that the convolution kernel size is 2D depth-separable convolution layer; sub-features Splicing fusion, after convolution kernel size is 2D convolution layer to obtain photovoltaic three-dimensional attention , the formula is as follows: ， in, Indicates that the convolution kernel size is 2D convolutional layer.

8. The solar photovoltaic power generation prediction method according to claim 7, characterized in that: Step S6 specifically includes: Focusing on photovoltaics Input into photovoltaic power generation prediction module, and transform the characteristics Flattened into a one-dimensional vector, then passed through the fully connected layer and Activation function calculation predicted photovoltaic power generation , the formula is as follows: ， in, represents a full connection operation, represents a vector flattening operation, and They respectively represent the third learnable weight parameter and the fourth learnable weight parameter obtained by the model through back propagation.

9. The solar photovoltaic power generation prediction method according to claim 8, characterized in that: The mean square error loss is used as the loss function. The calculation process of the loss function is: ， in, Represents the calculated value of the loss function, is the number of data samples, Expressed as the actual photovoltaic power value, Indicates that the sample is The photovoltaic power value predicted by the model at time .

10. Distributed photovoltaic power generation charging pile system, characterized in that: Including photovoltaic power generation module, energy storage module, power monitoring module, power distribution control module, charging pile module, communication module and power management module; The output end of the photovoltaic power generation module is connected to the input end of the energy storage module. The input end of the power monitoring module is connected; the output end of the power monitoring module is connected to the input end of the power distribution control module, the output end of the energy storage module is connected to the input end of the charging pile module, the input end of the communication module is respectively connected to the output end of the power monitoring module and the output end of the power distribution control module, and the output end of the communication module is connected to the cloud; the output end of the power distribution control module is connected to the user's power consumption system, the input end of the charging pile module is also connected to the output end of the municipal power grid, the input end of the power consumption management module is connected to the output end of the power distribution control module and the output end of the municipal power grid, and the output end of the power consumption management module is connected to the input end of the charging pile module and the power consumption system; wherein, the power distribution control module is used to execute the solar photovoltaic power generation prediction method described in any one of claims 1-9.

Images