Regional adaptive photovoltaic light shortage risk prediction method, device, equipment, and medium

Abstract

The present invention proposes a regional adaptive photovoltaic light shortage risk prediction method, device, equipment, and medium, which belongs to the technical field of new energy power generation risk prediction; including: step 1, construction of core and buffer analysis area data framework; respectively constructing data sets of core analysis area and buffer analysis area to form a data framework for solar radiation prediction; step 2, constructing a solar radiation time series prediction model of the buffer analysis area mask, adaptively aggregating feature information of adjacent areas according to the spatial relationship of the prediction area, and optimizing the prediction model to achieve end-to-end solar radiation prediction; step 3, designing a parameterized algorithm to screen extreme light shortage and risk analysis; the parameterized algorithm is used to screen the spatiotemporal distribution of extreme light shortage, and analyze the extreme light shortage risk faced by each photovoltaic panel in combination with the spatial distribution and elevation information of photovoltaic panels. The present invention can provide more accurate risk assessment for the operation and maintenance of photovoltaic power stations.

Description

Technical Field

[0001] The present invention belongs to the technical field of risk prediction of renewable energy power generation, and in particular, relates to a method, device, equipment and medium for regional adaptive photovoltaic low light risk prediction. Background Art

[0002] Solar energy is a renewable clean energy that does not directly emit carbon dioxide or air pollutants during operation. It has great potential to improve climate resilience, energy security and social justice. Photovoltaic power generation is a technology that directly converts solar energy into electrical energy by using the photovoltaic effect at the interface of semiconductors. The efficient use of photovoltaics is an important driving force for the production and development of new energy.

[0003] Due to the influence of factors such as precipitation and cloud cover, solar energy resources are intermittent, volatile, and random, resulting in high uncertainty in power generation in different regions and at different times. Among the many factors that interfere with photovoltaic power generation, the unknown risks brought about by random extreme weather events are one of the important reasons for the uncertainty in the current use of solar energy resources. At present, mainstream photovoltaic facilities are limited to solar radiation below 100W / m 2 According to the output characteristics of photovoltaic power generation, the incident solar short-wave radiation on the ground is less than 100W / m 2 The situation is regarded as low light, and an extreme low light event is defined. An extreme low light event refers to an event in which the solar radiation deviates seriously from its average state and has a great impact on photovoltaic power generation due to long-term, large-scale extreme weather such as continuous rain, sandstorms, haze, hail, etc., which is a low-probability event that is not easy to occur. An extreme low light event is defined according to the following judgment conditions: (1) When the number of consecutive extreme low light days is equal to 3 days, it is regarded as the beginning of an extreme low light event; (2) When there are no extreme low light days for 2 consecutive days, it is regarded as the end of an extreme low light event; (3) Non-continuous normal radiation days are allowed during an extreme low light event.

[0004] The prediction of extreme low light events depends on the prediction of solar radiation. Various mathematical methods have been used to predict solar radiation, such as numerical weather prediction (NWP) models. NWP models apply to fields of atmospheric variables. To predict the future state of the field, the model is initialized with observational data and a set of mathematical equations of atmospheric physics and dynamics are solved. Another method is model output statistics (MOS). MOS is a statistical method applied to the output of NWP models to predict other variables or improve the forecast.

[0005] Recently, a considerable number of studies have shown that the accuracy of predictions can be improved by using artificial intelligence models. The existing literature carefully reviews the methods of solar radiation prediction using artificial neural network technology. Other methods such as least squares support vector machine (LS-SVM), fuzzy and wavelet-based models have also been used as solar irradiation prediction tools. However, these studies often fail to combine the prediction results with the geospatial distribution of photovoltaic panels for more specific and detailed risk analysis. Therefore, these prediction results are not sufficient to be directly used for energy planning and scheduling under extreme low light conditions.

[0006] Some studies focus on the power generation or power generation of photovoltaic power plants. These studies provide more thorough analysis in energy planning, but they have some limitations. Some studies treat photovoltaic panels as isolated units, ignoring the possible spatiotemporal lag effects of risks between them. While other studies consider this spatiotemporal propagation, they mainly focus on the interaction between photovoltaic panels and ignore the characteristics of neighboring geographical areas where photovoltaic panels are not installed and their impact on power generation. This one-sidedness will lead to insufficient prediction accuracy. Specifically:

[0007] (1) In the prior art, many studies have shown that the use of artificial intelligence models can significantly improve the accuracy of predictions. In particular, an in-depth review of methods for solar radiation prediction using artificial neural network technology has been conducted. In addition, other methods such as least squares support vector machine (LS-SVM), fuzzy logic, and wavelet-based models have also been widely explored as tools for solar irradiance prediction. However, these studies usually do not combine the prediction results with the geographic spatial distribution of photovoltaic panels to conduct a more specific and detailed risk analysis. Therefore, these prediction results are not sufficient to be directly used for energy planning and scheduling under extreme light conditions. The purpose of the present invention is to overcome the limitations of the prior art by combining the prediction results with the distribution and elevation of photovoltaic panels to conduct a more specific and detailed risk analysis, thereby providing more accurate prediction results for energy planning and scheduling under extreme low light conditions.

[0008] (2) In the existing technology, some studies focus on the power generation or power generation of photovoltaic power stations, providing a more thorough analysis for energy planning. However, these studies have some limitations. Some studies regard photovoltaic panels as isolated units, ignoring the possible spatiotemporal lag effects of risks between them. While other studies consider this spatiotemporal propagation, they mainly focus on the interaction between photovoltaic panels and ignore the characteristics of neighboring geographical areas where photovoltaic panels are not installed and their impact on power generation efficiency. This one-sidedness will lead to insufficient prediction accuracy. Summary of the invention

[0009] In view of the deficiencies and problems of the prior art, the present invention proposes a regional adaptive photovoltaic low light risk prediction method, device, equipment, and medium. The core of the present invention is to construct a solar radiation prediction data framework, which consists of a core analysis area and a buffer analysis area. In the prediction stage, the method will extract data related to the prediction area from the data framework, and according to the analysis process, adaptively aggregate the feature information of the adjacent areas according to the spatial relationship of the prediction area, and optimize the prediction model. Based on the above prediction results, the present invention designs a set of parameterized algorithms for screening out the spatiotemporal distribution of extreme low light. Combined with the spatial distribution and elevation information of photovoltaic panels, this method can further refine the analysis of the extreme low light risk faced by each photovoltaic panel, thereby providing a more accurate risk assessment for the operation and maintenance of photovoltaic power stations.

[0010] The present invention not only considers the interaction between photovoltaic panels, but can also adaptively aggregate the solar radiation characteristics of adjacent areas and their impact on power generation based on the spatial relationship of the prediction area, so as to improve the risk prediction accuracy of photovoltaic power stations under extreme low-light conditions.

[0011] The present invention provides a regional adaptive photovoltaic extreme light shortage risk prediction method, which specifically comprises the following steps:

[0012] Step 1: Constructing the data framework of the core and buffer analysis areas; constructing the data sets of the core analysis area and the buffer analysis area respectively to form a data framework for solar radiation prediction;

[0013] Step 2: Build a solar radiation time series prediction model for the buffer analysis area mask, adaptively aggregate the feature information of neighboring areas according to the spatial relationship of the prediction area, and optimize the prediction model to achieve end-to-end solar radiation prediction;

[0014] Step 3: Design a parameterized algorithm to screen extreme light shortage and risk analysis. The parameterized algorithm is used to screen the spatiotemporal distribution of extreme light shortage, and analyze the extreme light shortage risk faced by each photovoltaic panel in combination with the spatial distribution and elevation information of the photovoltaic panels.

[0015] The present invention also provides a regional adaptive photovoltaic light shortage risk prediction device, comprising:

[0016] Core and buffer analysis area data framework construction module: construct the data sets of the core analysis area and buffer analysis area respectively to form a data framework for solar radiation prediction;

[0017] The solar radiation time series prediction model of the buffer analysis area mask adaptively aggregates the feature information of the neighboring areas according to the spatial relationship of the prediction area, and optimizes the prediction model to achieve end-to-end solar radiation prediction;

[0018] Parameterized algorithm module; used to screen the spatiotemporal distribution of extreme light shortage, and analyze the extreme light shortage risk faced by each photovoltaic panel based on the spatial distribution and elevation information of the photovoltaic panels.

[0019] The present invention also provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method.

[0020] The present invention also provides a computer-readable storage medium on which executable instructions are stored. When the instructions are executed by a processor, the processor implements the above method.

[0021] The present invention has the following beneficial effects:

[0022] The present invention proposes a regional adaptive photovoltaic light shortage risk prediction method, device, equipment, and medium. First, by constructing a data framework containing core and buffer analysis areas, the present invention comprehensively considers the geographical distribution of solar radiation and its interaction. In particular, for the geographical units at the boundaries of the core analysis area, the present invention introduces the concept of buffer analysis areas, so that these units can more comprehensively aggregate the influence of neighboring geographical units, whether they are located within or outside the analysis range, thereby significantly improving the comprehensiveness and accuracy of the prediction. Next, the present invention adopts the graph convolutional neural network (GCN) technology, which is a deep learning technology that can focus on the graph structure, effectively aggregate the information of neighboring geographical units, and simulate the complex associations in the geographic space. By combining the feature representation method and GCN, the present invention constructs a model for adaptively aggregating the feature information of neighboring areas according to the spatial relationship of the prediction area, and optimizing the prediction model to achieve end-to-end solar radiation prediction. In particular, in the back propagation stage of the model error, the present invention performs mask processing on the buffer analysis area, so that the model is more focused on the optimization fitting of the core analysis area, thereby further improving the prediction accuracy of the model. Finally, based on the prediction results, the present invention designs a set of parameterized algorithms for screening the spatiotemporal distribution of extreme light shortage. Combined with the spatial distribution and elevation information of photovoltaic panels, this method can further refine the analysis of the extreme low light risk faced by each photovoltaic panel, providing a more accurate risk assessment for the operation and maintenance of photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Technical flow chart of building the core and buffer analysis area data framework;

[0024] Figure 2 Schematic diagram of constructing a buffer zone based on the core analysis area;

[0025] Figure 3 Schematic diagram of rectangular enveloping the buffer zone of the core analysis area;

[0026] Fig. 4 Schematic diagram of the rectangular buffer zone subdivided into a fishing net-like grid;

[0027] Figure 5 Schematic diagram of spatial intersection analysis between the buffer zone fishing net and the core analysis area;

[0028] Fig. 6 Schematic diagram of the technical flow chart for constructing a regional adaptive solar radiation prediction model;

[0029] Fig. 7 Schematic diagram of mapping solar radiation values ​​of the fishing net in the buffer zone of the forecast area to node features;

[0030] Figure 8 (a) Schematic diagram of the convolutional neural network model structure for solar radiation prediction graph;

[0031] Fig. 8(b) Schematic diagram of the graph convolution module in the solar radiation prediction graph convolutional neural network model structure. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in each embodiment of the present invention described below can be combined with each other as long as they do not conflict with each other. To achieve the above-mentioned purpose, the present invention adopts the following technical scheme.

[0033] The present invention proposes a regional adaptive photovoltaic low light risk prediction method, which specifically includes the following steps:

[0034] Step 1, construction of the core and buffer analysis area data framework. In order to comprehensively consider the geographical distribution of solar radiation and its interaction, the present invention constructs data sets of the core analysis area and the buffer analysis area respectively to form a data framework for solar radiation prediction. Existing studies often fail to fully consider the interaction between geographical units at the boundaries of the core analysis area and geographical units outside the analysis range when dealing with them, which leads to inaccurate predictions. In order to overcome this limitation, the present invention introduces the concept of buffer analysis areas, so that geographical units at the boundaries can more comprehensively aggregate the influence of neighboring geographical units, whether they are within or outside the analysis range, thereby improving the comprehensiveness and accuracy of the prediction.

[0035] Step 2, construct a solar radiation time series prediction model for the buffer analysis area mask; the present invention adopts a graph convolutional neural network (GCN), which is a deep learning technology that can focus on the graph structure, effectively aggregate the information of neighboring geographic units, and simulate the complex associations in the geographic space. By combining the feature representation method and GCN, the present invention constructs a model for adaptively aggregating the feature information of neighboring areas according to the spatial relationship of the prediction area, and optimizing the prediction model to achieve end-to-end solar radiation prediction. The innovation of the present invention is that in the back propagation stage of the model error, the buffer analysis area is masked, so that the model is more focused on the optimization fitting of the core analysis area, thereby improving the prediction accuracy of the model.

[0036] Step 3: Design a parameterized algorithm to screen for extreme light shortage and risk analysis: Based on the prediction results, the present invention designs a parameterized algorithm to screen for the spatiotemporal distribution of extreme light shortage. Combined with the spatial distribution and elevation information of photovoltaic panels, this method can further refine the analysis of the extreme light shortage risk faced by each photovoltaic panel.

[0037] More specifically, step 1, construction of the core and buffer analysis area data framework; the specific flow chart is as follows Figure 1 As shown, including:

[0038] Buffers are constructed based on the core analysis area to generate buffers for the core analysis area, such as Figure 2 As shown. Assume that the spatial resolution of the solar radiation data used is On this basis, the radius of the buffer zone is further defined as .

[0039] The buffer of the core analysis area is enclosed in a rectangular shape to form a core analysis area buffer rectangular envelope, which is a buffer of the core analysis area in a regular rectangular shape, such as Figure 3 This rectangular envelope processing method is intended to simplify the geometric shape of the buffer zone to facilitate subsequent data processing and analysis.

[0040] The core analysis area buffer rectangular enveloping fishnet grid includes: The present invention further introduces a fishnet-like subdivision method to subdivide the rectangular buffer into fishnet-like grids. The resolution of each grid is set to ,like Figure 4 shown.

[0041] Historical solar radiation data were collected using the core analysis area buffer fishing net as a unit to form a data framework that includes the core and buffer analysis areas.

[0042] Perform spatial intersection analysis on the buffer fishnet and the core analysis area to identify the overlap between the two, such as Figure 5The intersecting grid cells are used as the core analysis area fishnet.

[0043] More specifically, step 2, constructing a regional adaptive solar radiation prediction model for buffer analysis regional masks; the technical flow chart is as follows Figure 6 As shown, including:

[0044] The prediction area grid ID is selected from the fishing net in the core analysis area, and a radius of The buffer is processed into a rectangular envelope for subsequent analysis. The rectangular envelope of the buffer zone in the prediction area is intersected with the rectangular envelope of the buffer zone in the core analysis area, and the grid ID of the intersection area is recorded as the grid ID of the buffer zone in the prediction area. The intersection area is defined as the buffer zone fishnet in the prediction area.

[0045] Create an integer attribute column 'Buffer' based on the prediction area buffer grid ID. Traverse the prediction area buffer grid IDs and assign values ​​of 0 or 1 depending on whether they belong to the prediction area grid ID.

[0046] The prediction area buffer fishnet ( The solar radiation values ​​of each geographic unit are mapped to node features to provide spatial information and features for subsequent prediction models, such as Figure 7 shown.

[0047] Defining the spatial dependency weights of the spatial relationships between nodes in the adaptive prediction area buffer includes: taking a single grid as a node in the graph model, further defining the spatial dependency weights between nodes in the prediction area buffer by a Gaussian kernel weighting function, and forming a matrix representing the spatial dependency weights of the spatial relationship of the adaptive prediction area buffer If the spatial dependency weight between two nodes is less than or equal to the set threshold, it is considered that when predicting the solar radiation of one of the nodes, the spatial dependency between the other node and it needs to be considered, as shown in Formula 1.

[0048] (1)

[0049] in, is the spatial dependency weight between node i and node j, which is recorded in No. Line List, is the Euclidean distance between node i and node j, is the standard deviation of the distance, is the threshold value.

[0050] The threshold is set so that More sparse, with fewer direct connections between nodes. This helps reduce the complexity of the model and allows the learning process to focus more on important node relationships.

[0051] The attribute column 'Buffer' information is used to limit the spatial scope of the graph convolution model to calculate the loss function; the graph convolution neural network model is optimized and predicted; including: the integer attribute column 'Buffer' records whether each grid ID belongs to the prediction area or the buffer area. If it is the prediction area, the value of 'Buffer' is 0, and if it is the buffer area, the value of 'Buffer' is 1. This information will be used in the back propagation of the subsequent graph convolution neural network model optimization to limit the spatial scope of the model analysis when calculating the loss function.

[0052] The graph convolutional neural network model outputs the solar radiation forecast time series of the forecast area buffer grid; the model output includes: Geographical units in (Length is Solar radiation time series) Geographical units in (Length is time series of solar radiation.

[0053] The structure of the solar radiation prediction graph convolutional neural network model is shown in Figure 8(a). The structure of the solar radiation prediction graph convolutional neural network model includes:

[0054] Input layer: The input layer is the first layer of the neural network and is responsible for receiving the raw input data. The input data is a time series of solar radiation. The input layer does not perform any calculations, but simply passes the data to the next layer.

[0055] Graph Convolution Module: The graph convolution module is the core of the model and is used to process and extract features from graph-structured data. As shown in Figure 8(b), this module contains three main layers: Temporal Convolution Layer: First, the temporal convolution layer processes the input time series data and extracts temporal correlations. This layer is able to capture the pattern of solar radiation values ​​changing over time. Graph Convolution Layer (using Aspace): Next, the graph convolution layer processes the graph-structured data and captures the correlation between different geographic units. This layer updates the representation of nodes by aggregating and transforming the neighbor information of nodes, thereby learning the mutual influence between geographic units. Temporal Convolution Layer: Finally, another temporal convolution layer further processes the data that has passed through the graph convolution layer to extract deeper temporal features.

[0056] Fully connected layer: After the graph convolution module, the fully connected layer maps the output of the previous layer to a high-dimensional space and performs more complex nonlinear transformations. This layer integrates the features extracted by the temporal convolution layer and the graph convolution layer to provide richer information for the final prediction task.

[0057] Forecast results: Finally, the model outputs the forecast results, namely the solar radiation forecast series, which is the final forecast of the solar radiation of the input data time series.

[0058] The entire model process is as follows: the input data first passes through the input layer, and then through two graph convolution modules for feature extraction. Each graph convolution module contains two temporal convolution layers and one graph convolution layer. The extracted features then pass through the fully connected layer to finally get the prediction result.

[0059] The calculation formula of graph convolution in the graph convolution layer is shown in formula (2)-formula (4):

[0060] (2)

[0061] (3)

[0062] (4)

[0063] in, Indicates that the node is The feature vector of the layer, It means that after convolution, the node is The feature vector of the layer, Indicates The parameters of the layer convolution, represents the activation function. And the matrix The component is a Laplacian matrix. In is the identity matrix, that is, a matrix with 1s on the diagonal and 0s elsewhere. is a degree matrix, only the main diagonal elements are not zero; express Main diagonal elements; express In each element; i represents the row number, j represents the column number. In the present invention, Using a matrix of spatially dependent weights .

[0064] The model uses the L2 loss function, as shown in formula (5):

[0065] (5)

[0066] in, : is the loss function, which measures the difference between the model’s predicted value and the actual value. In this case, the L2 loss function is used, also known as Euclidean loss or Mean Squared Error (MSE). : are the training parameters of the model, including the weights and biases in the convolutional neural network. These parameters are adjusted by the optimization algorithm during the training process to minimize the loss function. : is the model's predicted value sequence, representing the time point arrive The predicted solar radiation value. : is the actual observed solar radiation value sequence, corresponding to the predicted value sequence, and is used to calculate the loss function. : is the starting point of the time series. : is the predicted time range, i.e. arrive The number of time points. : For each From 1 to , Represents the model at a point in time The predicted solar radiation value. : For each From 1 to , Represents the model at a point in time The actual observed solar radiation value.

[0067] In the present invention, the buffer analysis area mask is used in the error back propagation process to improve the prediction accuracy. Specifically, first identify the grids in the data set with the value of 1 in the 'Buffer' column. These grids will not be considered in the error back propagation process. and predicted values are all set to 0, effectively eliminating their influence on the model parameter optimization. This approach allows the model to focus more on the grids within the prediction area, thereby optimizing the model's performance in these key areas.

[0068] More specifically, step 3, designing a parameterized algorithm for screening extreme light shortage and risk analysis, including:

[0069] The extreme low light screening algorithm of the present invention is designed for the solar radiation prediction sequence output in step 2, and combines it with a preset extreme low light threshold, aiming to identify and screen out the spatiotemporal distribution of extreme low light events occurring in the prediction time series.

[0070] The extreme low light screening algorithm is used to screen out extreme low light events in the solar radiation forecast time series. The specific process of the algorithm is as follows:

[0071] Initialization: The algorithm receives three input parameters: solar radiation forecast time series , the preset low light threshold , and the ID of the grid.

[0072] Traversing the sequence: The algorithm traverses the forecast time series Each element in .

[0073] Detect the start of a lack of light: For each element in the sequence , the algorithm checks whether the following conditions are met:

[0074] (The solar radiation value at the current time point is less than or equal to the threshold);

[0075] (The solar radiation value at the next time point is also less than or equal to the threshold);

[0076] (The solar radiation value at the next time point is also less than or equal to the threshold);

[0077] Record the start of the light shortage: If the above conditions are met, the algorithm records As the beginning of an extreme low light event, Set to i.

[0078] Detecting the end of low light: Next, the algorithm checks from arrive The remaining sequence , looking for the end point of the extreme low-light event. Each element in , algorithm check:

[0079] (The solar radiation value at the current time point is greater than the threshold);

[0080] (The solar radiation value at the next time point is also greater than the threshold);

[0081] Record the end of the light: If you find a , then record for , indicating that the extreme low-light event The time point ends.

[0082] Output: The algorithm outputs the grid ID of the extreme low light events from Start until It's the last day.

[0083] No light detected: If no light satisfying the condition is found during the traversal of the entire sequence, , then there is no low-light event in the predicted sequence of the algorithm output grid ID.

[0084] This algorithm determines the start of a low-light event when the solar radiation values ​​are below a threshold for three consecutive days, and the end of a low-light event when the solar radiation values ​​are above a threshold for two consecutive days.

[0085] Based on the calculation results of the extreme low light screening algorithm, the grid IDs where extreme low light events occurred are collected. Using these grid IDs, the spatial position of the corresponding grid in the core analysis area can be indexed, and spatial overlay analysis can be performed with the photovoltaic panels in the area, so as to accurately retrieve the photovoltaic panels within the spatial range of these grids.

[0086] Next, we further analyze the spatial distribution of these photovoltaic panels and overlay them with the elevation data to retrieve the elevation corresponding to the photovoltaic panels. Based on the power generation principle of photovoltaic cells, it can be seen that the open circuit voltage of photovoltaic modules will decrease with increasing temperature, resulting in a decrease in power generation, and its typical temperature coefficient is -0.35% / ℃. Since the temperature decreases with increasing altitude, when facing extreme low light events, the areas with lower elevations will have a greater decrease in power generation than areas with higher elevations due to higher temperatures, so the risks they face are relatively higher.

[0087] Based on this analysis, photovoltaic panels at altitudes of 0-1500 meters, 1500-3000 meters, and above 3000 meters are defined as areas facing extremely high risk, high risk, and relatively high risk, respectively, thereby achieving more specific and detailed risk analysis and providing more accurate prediction results for energy planning and arrangements under extreme low light conditions.

Claims

1. A regional adaptive photovoltaic light shortage risk prediction method, characterized in that: The specific steps include: Step 1: Construct the data framework of the core analysis area and the buffer analysis area; construct the data sets of the core analysis area and the buffer analysis area respectively to form a data framework for solar radiation prediction; Step 2: Build a solar radiation time series prediction model based on the buffer analysis area mask, adaptively aggregate the feature information of neighboring areas according to the spatial relationship of the prediction area, and optimize the solar radiation time series prediction model to achieve end-to-end solar radiation prediction; Step 3: Design a parameterized algorithm to screen extreme light shortage and risk analysis. The parameterized algorithm is used to screen the spatiotemporal distribution of extreme light shortage, and analyze the extreme light shortage risk faced by each photovoltaic panel in combination with the spatial distribution and elevation information of the photovoltaic panels.

2. A regional adaptive photovoltaic light shortage risk prediction method according to claim 1, characterized in that: Step 1 includes: Building a buffer zone based on the core analysis area to generate a core analysis area buffer zone; Envelope the core analysis area buffer in a rectangular manner, thereby forming a core analysis area buffer rectangular envelope; Subdividing the core analysis area buffer rectangular envelope into fishnet grids to form a core analysis area buffer rectangular envelope fishnet grid; The historical solar radiation data were collected using the core analysis area buffer envelope fishing net as a unit to form a data framework that includes the core and buffer analysis areas; A spatial intersection analysis is performed between the core analysis area buffer rectangular envelope fishing net, i.e., the buffer area fishing net, and the core analysis area to identify the overlapping parts between the two, and the intersecting grid cells are used as the core analysis area fishing net.

3. A regional adaptive photovoltaic light shortage risk prediction method according to claim 2, characterized in that: Step 2 includes: By selecting a grid from the core analysis area fishing net as the grid ID of the prediction area, a prediction area buffer rectangular envelope is generated around the prediction area for subsequent analysis; by intersecting the prediction area buffer rectangular envelope with the core analysis area buffer rectangular envelope fishing net, the grid ID of the intersection area is recorded as the grid ID of the prediction area buffer; the intersection area is defined as the prediction area buffer fishing net; Create an integer attribute column Buffer based on the prediction area buffer grid ID, traverse the prediction area buffer grid ID, and assign a value of 0 or 1 according to whether it belongs to the prediction area grid ID; The solar radiation values ​​of the fishnet in the buffer zone of the prediction area are mapped into node features to provide spatial information and features for subsequent prediction models; The spatial dependency weights that define the spatial relationships between nodes in the adaptive prediction zone buffer; The integer attribute column Buffer information is used to limit the spatial range of the graph convolutional neural network model to calculate the loss function; the graph convolutional neural network model is optimized and predicted; wherein the graph convolutional neural network model is used as a solar radiation time series prediction model; The graph convolutional neural network model outputs the solar radiation forecast time series for the buffer zone grid in the forecast area.

4. A regional adaptive photovoltaic light shortage risk prediction method according to claim 3, characterized in that: Step 3 includes: The extreme low light screening algorithm is used to screen the solar radiation prediction sequence output in step 2, and the screened solar radiation prediction sequence is combined with the preset extreme low light threshold to identify and screen the spatiotemporal distribution of extreme low light events appearing in the solar radiation prediction sequence.

5. A regional adaptive photovoltaic light shortage risk prediction method according to claim 4, characterized in that: The spatial dependency weights that define the spatial relationships between nodes in the adaptive prediction zone buffer in step 2 include: A single grid is used as a node in the graph model, and the spatial dependency weights between nodes in the prediction area buffer are defined by the Gaussian kernel weighting function to form a matrix representing the spatial dependency weights of the adaptive prediction area buffer spatial relationship. , if the spatial dependency weight between two nodes is less than or equal to the set threshold, then when predicting the solar radiation of one of the nodes, the spatial dependency between the other node and it is considered, as shown in formula (1): (1) in, is the spatial dependency weight between node i and node j, which is recorded in the spatial dependency weight matrix No. Line List, is the Euclidean distance between node i and node j, is the standard deviation of the distance, is the threshold value, It is an intermediate process parameter and is not defined.

6. A regional adaptive photovoltaic light shortage risk prediction method according to claim 5, characterized in that: A solar radiation prediction graph convolutional neural network model is used. The calculation formula of the graph convolution in the graph convolution layer in the solar radiation prediction graph convolutional neural network model is shown in formula (2) to formula (4): (2) (3) (4) in, Indicates that the node is The feature vector of the layer, It means that after convolution, the node is The feature vector of the layer, Indicates The parameters of the layer convolution, Represents the activation function, represented by the matrix The component is a Laplace moment, In is the identity matrix, that is, a matrix with 1 on the diagonal and 0 on the rest. is a degree matrix, only the main diagonal elements are not zero; express Main diagonal elements; express Each element in is the intermediate process parameter matrix, undefined, i represents the row number, j represents the column number, Using a matrix of spatially dependent weights .

7. A regional adaptive photovoltaic light shortage risk prediction method according to claim 6, characterized in that: The solar radiation prediction graph convolutional neural network model adopts the L2 loss function, as shown in formula (5): (5) in, is the loss function, are the training parameters of the model, including the weights and biases in the convolutional neural network, is the model's predicted value sequence, representing the time point arrive The predicted solar radiation value is is the sequence of actually observed solar radiation values, corresponding to the sequence of predicted values, and is used to calculate the loss function. is the starting point of the time series, is the forecast time range, i.e. arrive The number of time points, for each From 1 to , Represents the model at a point in time The predicted solar radiation value for each From 1 to , Represents the solar radiation prediction graph convolutional neural network model at time point The actual observed solar radiation value.

8. A regional adaptive photovoltaic light shortage risk prediction method according to claim 7, characterized in that: The buffer analysis area mask is used in the error back propagation process, which includes: first identifying the grids in the data set whose Buffer column is assigned a value of 1. These grids are not considered in the error back propagation process. and predicted values All are set to 0 to eliminate their influence on model parameter optimization.

9. A regional adaptive photovoltaic light shortage risk prediction device, characterized in that: include: Core buffer analysis and buffer analysis regional data framework building modules; Construct data sets for the core analysis area and buffer analysis area respectively to form a data framework for solar radiation prediction; The solar radiation time series prediction model of the buffer analysis area mask adaptively aggregates the feature information of the neighboring areas according to the spatial relationship of the prediction area, and optimizes the prediction model to achieve end-to-end solar radiation prediction; Parameterized algorithm module; used to screen the spatiotemporal distribution of extreme light shortage, and analyze the extreme light shortage risk faced by each photovoltaic panel based on the spatial distribution and elevation information of the photovoltaic panels.

10. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement a regional adaptive photovoltaic low light risk prediction method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that: Executable instructions are stored thereon, and when the instructions are executed by the processor, the processor implements the regional adaptive photovoltaic low-light risk prediction method as described in any one of claims 1 to 8.

Images