Regional distributed photovoltaic power prediction method, device and equipment based on multi-site space-time correlation

Abstract

This application discloses a method, apparatus, and equipment for predicting regional distributed photovoltaic (PV) power based on the spatiotemporal correlation of multiple sites, relating to the field of photovoltaic power generation. The method includes: clustering distributed PV power stations within a target area to obtain multiple sub-clusters; determining a representative power station for each sub-cluster based on the historical power time series of the distributed PV power stations in each sub-cluster; predicting the power output of the representative power station using a graph neural network and a multilayer perceptron based on the historical power time series of the representative power station and its neighboring stations, obtaining the predicted output of the representative power station; proportionally amplifying the predicted output of the representative power station of the sub-cluster based on the total installed capacity of the distributed PV power stations within the sub-cluster and the installed capacity of the representative power station of the sub-cluster, obtaining the predicted output of the sub-cluster; and determining the predicted output of the target area based on the predicted output of each sub-cluster. This application improves the prediction accuracy of large-scale distributed PV power.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation, and in particular to a method, apparatus and equipment for predicting regional distributed photovoltaic power based on the spatiotemporal correlation of multiple sites. Background Technology

[0002] With the accelerated global transition to a low-carbon energy structure, photovoltaic (PV) power generation, as an important form of renewable energy, is playing an increasingly vital role in the energy system. Distributed PV systems, due to their advantages such as localized energy consumption, flexible installation, and environmental friendliness, have become an important component of the new power system. However, compared to centralized PV power plants, distributed PV systems are characterized by their large number, wide distribution, and significant environmental variations. Their power generation is affected by multiple factors such as geographical location, meteorological conditions, and seasonal changes, exhibiting significant spatiotemporal fluctuations. Therefore, accurately predicting the power generation of large-scale distributed PV systems within a region has become a crucial technical issue for the safe and stable operation and dispatch optimization of the power grid.

[0003] Currently, the mainstream technical approaches for regional power prediction mainly include the following three categories.

[0004] (1) Cumulative method: The output of each power station is predicted separately and then summarized. It is suitable for scenarios with a small number of power stations and complete data, but it has high computational complexity and strong dependence on data quality in large-scale systems.

[0005] (2) Extrapolation method: Based on the similarity of historical meteorological conditions, prediction is made by matching similar irradiance and meteorological scenarios to obtain regional output results. It is suitable for areas with sufficient meteorological monitoring, but it is difficult to apply in areas with missing meteorological data.

[0006] (3) Amplification method: By clustering, power plants with similar power generation characteristics are grouped together, and representative power plants are selected for prediction and then amplified according to capacity ratio. This can significantly reduce the requirements for data integrity of individual distributed photovoltaic power plants and the computational overhead.

[0007] Due to the large number and wide distribution of regional distributed photovoltaic (PV) power stations, it is impossible to guarantee the deployment of complete meteorological monitoring equipment at each station. Furthermore, data gaps are likely to be widespread. Therefore, the amplification method is generally chosen for regional distributed PV power prediction. However, existing technologies still have significant shortcomings in the following aspects:

[0008] (1) The amplification method requires clustering of regional distributed photovoltaic sites. However, most clustering methods currently use fixed spatial similarity measures, which cannot be dynamically updated over time, making it difficult to reflect the changes in photovoltaic power output patterns under different seasons or climate conditions.

[0009] (2) Traditional clustering relies on complete historical data. For the data missing problem that is common in large-scale distributed photovoltaic scenarios, interpolation will introduce additional errors and reduce the accuracy of clustering and prediction.

[0010] (3) Existing amplification methods rely mainly on historical data of a single station when making predictions for representative power plants, and fail to effectively utilize the local spatial correlation information of surrounding power plants, resulting in insufficient short-term accuracy and robustness under climate disturbances. Summary of the Invention

[0011] The purpose of this application is to provide a method, apparatus, and equipment for predicting regional distributed photovoltaic power based on the spatiotemporal correlation of multiple sites, which can effectively improve the prediction accuracy of large-scale distributed photovoltaic power.

[0012] To achieve the above objectives, this application provides the following solution:

[0013] Firstly, this application provides a regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation, including:

[0014] Distributed photovoltaic power stations within the target area are clustered to obtain multiple sub-clusters;

[0015] Based on the historical power time series of distributed photovoltaic power stations in each sub-cluster, the representative power station of each sub-cluster is determined;

[0016] For any representative power station, based on the historical power time series of the representative power station and the historical power time series of its neighboring stations, a graph neural network and a multilayer perceptron are used to predict the power output of the representative power station.

[0017] For any sub-cluster, based on the total installed capacity of the distributed photovoltaic power stations within the sub-cluster and the installed capacity of the representative power station of the sub-cluster, the predicted output of the representative power station of the sub-cluster is proportionally amplified to obtain the predicted output of the sub-cluster.

[0018] The predicted output of the target region is determined based on the predicted output of each sub-cluster.

[0019] Secondly, this application provides a regional distributed photovoltaic power prediction device based on multi-site spatiotemporal correlation, comprising:

[0020] The clustering module is used to cluster distributed photovoltaic power stations within the target area to obtain multiple sub-clusters;

[0021] The representative power station selection module is used to determine the representative power station for each sub-cluster based on the historical power time series of distributed photovoltaic power stations in each sub-cluster.

[0022] The power plant power prediction module is used to predict the power output of any representative power plant by using a graph neural network and a multilayer perceptron, based on the historical power time series of the representative power plant and the historical power time series of its neighboring power plants.

[0023] The cluster output determination module is used to, for any sub-cluster, proportionally amplify the predicted output of the representative power station of the sub-cluster based on the total installed capacity of the distributed photovoltaic power stations within the sub-cluster and the installed capacity of the representative power station of the sub-cluster, to obtain the predicted output of the sub-cluster.

[0024] The regional output determination module is used to determine the predicted output of the target region based on the predicted output of each sub-cluster.

[0025] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation.

[0026] According to the specific embodiments provided in this application, this application has the following technical effects: by clustering photovoltaic power plants in the target area and selecting representative power plants, the typicality of each sub-cluster is ensured, data redundancy is reduced, and the clustering accuracy is significantly improved; in the prediction process, the historical power time series data of representative power plants and neighboring power plants are integrated, and the spatial correlation features are captured by graph neural networks. The time series patterns are mined by multilayer perceptrons. The dual models work together to enhance the depth of data information mining, which greatly improves the accuracy of power prediction for representative power plants. Furthermore, based on the proportional scaling strategy of total installed capacity and representative power plant installed capacity, combined with the summarization of prediction results of each sub-cluster, high-precision prediction of photovoltaic power in the target area is finally achieved. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is an application environment diagram of a regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation in one embodiment of this application.

[0029] Figure 2 This is a flowchart illustrating a regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation, provided as an embodiment of this application.

[0030] Figure 3 This is a schematic diagram illustrating the process of determining the predicted force in the target region in one embodiment of this application.

[0031] Figure 4 This is a schematic diagram illustrating the execution process of a regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation, provided in an embodiment of this application.

[0032] Figure 5 This is a comparison chart of the power curves of a sub-cluster and a representative power plant in one embodiment of this application.

[0033] Figure 6 The diagram shows the power prediction results of different methods in one embodiment of this application.

[0034] Figure 7 This is a schematic diagram of the functional modules of a regional distributed photovoltaic power prediction device based on the spatiotemporal correlation of multiple sites, provided in an embodiment of this application.

[0035] Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0036] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] The regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be set up independently, integrated into server 102, or placed in the cloud or on another server. Terminal 101 can send the location and historical power time series of distributed photovoltaic power stations within the target area to server 102. After receiving the location and historical power time series of distributed photovoltaic power stations within the target area, server 102 performs power prediction to obtain the predicted power output of the target area. Server 102 can feed back the obtained predicted power output to terminal 101. Furthermore, in some embodiments, the regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation can also be implemented independently by server 102 or terminal 101.

[0039] The terminal 101 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 102 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0040] In one exemplary embodiment, such as Figures 2 to 4 As shown, a regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 102 as an example, the explanation includes the following steps 201 to 205.

[0041] Step 201: Cluster the distributed photovoltaic power stations in the target area to obtain multiple sub-clusters.

[0042] The purpose of clustering is to group power plants with more similar output patterns into one category, thereby improving the accuracy of the amplification method. Therefore, this application designs a spatiotemporal dual-scale dynamic clustering method. Since geographically close distributed photovoltaic power plants are affected by similar meteorological conditions, their output power may have a strong correlation. The dual-scale dynamic clustering method first clusters according to geographical location, dividing the entire region into several sub-regions. Within each sub-region, further clustering is performed based on historical power time series to form sub-clusters.

[0043] In a specific application example, step 201 includes steps 11 to 13.

[0044] Step 11: Based on the geographical location (latitude and longitude) of each distributed photovoltaic power station in the target area, the K-medoids method is used to cluster the distributed photovoltaic power stations in the target area to obtain multiple sub-regions. Among them, the elbow method is used to select the optimal K value.

[0045] Step 12: For any sub-region, based on the historical power time series of each distributed photovoltaic power station in the sub-region, the probability mass similarity between each pair of distributed photovoltaic power stations in the sub-region is calculated using the Probability Mass Similarity Kernel (PMK) method to obtain a similarity matrix.

[0046] Specifically, based on each sub-region, further clustering is performed according to historical power time series to identify distributed photovoltaic power plants with similar output patterns. The PMK method is used to directly cluster historical power time series containing missing values ​​without imputation. PMK measures similarity by comparing the probability quality between data points, and can robustly cluster missing data under different mechanisms.

[0047] Distributed photovoltaic power station With distributed photovoltaic power stations The probabilistic quality similarity between them is defined as:

[0048] ;

[0049] ;

[0050] in, For distributed photovoltaic power stations With distributed photovoltaic power stations Probability quality similarity between them For distributed photovoltaic power stations Historical power time series, For distributed photovoltaic power stations Historical power time series, For distributed photovoltaic power stations With distributed photovoltaic power stations Dissimilarity The total historical time window length, This represents the total number of distributed photovoltaic power stations. For the first Time steps fall and The number of data points within the range, when using After standardization, this range is approximately equal to the probability mass of these two data distributions. For the first Time-step distributed photovoltaic power station power, For the first Time-step distributed photovoltaic power station The power.

[0051] For efficient calculation Each time step Discretized into There are 10 intervals, and missing data are placed in a separate bin, with the number denoted as 1. For cases containing missing values, PMK adjusts as follows: The value.

[0052] When there is a unilateral deletion, if When missing, .

[0053] in, for The number of landing points on the left side of the corresponding compartment. for The number of missing values ​​to the right of the bin. Assuming missing values ​​might fall within a wider range, the possible range of missing values ​​has been appropriately enlarged to maintain conservatism without interpolation.

[0054] When the data on both sides are at the time step When both are missing This is equivalent to treating missing values ​​as potentially covering the entire interval and assigning the maximum dissimilarity to that time step to reflect the uncertainty caused by missing information.

[0055] The PMK value between every two distributed photovoltaic power stations is calculated using the above method, thus constructing... The PMK similarity matrix of order 1.

[0056] Step 13: Based on the similarity matrix, the distributed photovoltaic power stations within the sub-region are clustered using the K-means clustering method to obtain multiple sub-clusters. This application uses the PMK similarity matrix instead of the traditional Euclidean distance matrix to further cluster into several sub-clusters within each sub-region.

[0057] Furthermore, this application does not use all historical power data from all distributed photovoltaic power plants to generate the PMK similarity matrix, but instead uses a monthly rolling approach to dynamically update the input power time series and its corresponding PMK clustering results.

[0058] Step 202: Determine the representative power station of each sub-cluster based on the historical power time series of the distributed photovoltaic power stations in each sub-cluster.

[0059] In a specific application example, for any sub-cluster, the Pearson correlation coefficient between the historical power time series of each distributed photovoltaic power station in the sub-cluster and the total power series of the sub-cluster is calculated, and the distributed photovoltaic power station with the highest Pearson correlation coefficient is taken as the representative power station of the sub-cluster.

[0060] Because this application uses a monthly rolling dynamic clustering mechanism, the sub-clusters will change over time, and therefore the representative power plant will also be updated monthly.

[0061] Step 203: For any representative power station, based on the historical power time series of the representative power station and the historical power time series of its neighboring stations, a graph neural network (GNN) and a multilayer perceptron are used to predict the power output of the representative power station, thereby obtaining the predicted power output of the representative power station.

[0062] After identifying the representative power plant for each sub-cluster, in order to improve the prediction accuracy of the representative power plant, a graph neural network was constructed around the representative power plant to integrate neighborhood spatial information for power prediction by utilizing the spatial correlation of power plants around the representative power plant. Under short-term disturbances such as rapid changes in cloud cover, the model can better perceive changes in local space, thereby improving the accuracy of the prediction of the representative power plant.

[0063] In a specific application example, step 203 includes steps 31 to 33.

[0064] Step 31: Select a distributed photovoltaic power station that is closest to the representative power station at every 60° interval around the representative power station as a neighboring station of the representative power station, so as to ensure that the nearest power stations around the representative power station are not all from the same direction, and to ensure information gain from multiple directions.

[0065] Step 32: Based on the historical power time series of the representative power station and the historical power time series of each neighboring station, a graph neural network is used to perform multi-layer directed graph convolution operations in each time window to obtain the final hidden tensor.

[0066] Specifically, step 32 includes the following steps (1) to (5).

[0067] (1) The representative power station and each neighboring station are used as graph nodes, and each neighboring station is connected to the representative power station, with the direction from the neighboring station to the representative power station, to construct a local graph: .in, For the set of graph nodes, , As a representative power station, Distance represents power station The set of neighboring stations, Let be the set of edges of the graph.

[0068] (2) Based on the local map, the geographical locations of the representative power station and each neighboring station, determine the directed adjacency matrix:

[0069] ;

[0070] ;

[0071] in, For the distributed photovoltaic power station corresponding to the directed adjacency matrix With distributed photovoltaic power stations elements, The edge weights are based on geographic location and are calculated using distance decay and normalization within the neighborhood. The geographical distance between the two distributed photovoltaic power stations. For the distance scale, As a representative power station The set of neighboring stations. The closer the geographical location, the more similar the meteorological conditions and the more similar the power generation patterns, and the greater the weight.

[0072] (3) Add an identity matrix to the directed adjacency matrix and normalize it to obtain the directed propagation matrix:

[0073] ;

[0074] ;

[0075] ;

[0076] in, It is a directed adjacency matrix. It is the identity matrix. Given a directed adjacency matrix with self-loops, For a directed propagation matrix, This is the out-degree diagonal matrix obtained by summing the rows. This represents the number of graph nodes in the local graph. for The Line 1 Column elements.

[0077] (4) Generate a three-dimensional tensor based on the graph nodes in the local graph and the historical power time series of each graph node. The three-dimensional tensor includes graph nodes, time, and power.

[0078] (5) Based on the directed propagation matrix and the three-dimensional tensor, perform multi-layer directed graph convolution operation in each time window to obtain the final hidden tensor.

[0079] Specifically, the forward propagation formula for directed graph convolution operations is:

[0080] ;

[0081] ;

[0082] in, For time indexing, The total historical time window length, To predict the length of the historical input data, For layer index, The number of features in the three-dimensional tensor. For a three-dimensional tensor, For the first Layer hidden representation tensor, For the first Layer hidden representation tensor, For the hidden representation tensor of layer 0, This indicates a splicing operation based on the time dimension. For the first Learnable weights of layers For the first Learnable bias of the layer It is a non-linear activation function.

[0083] Step 33: Based on the final hidden tensor, a multilayer perceptron is used to predict the power output of the representative power station to obtain the predicted power output of the representative power station.

[0084] Step 204: For any sub-cluster, based on the total installed capacity of the distributed photovoltaic power stations within the sub-cluster and the installed capacity of the representative power station of the sub-cluster, the predicted output of the representative power station of the sub-cluster is proportionally amplified to obtain the predicted output of the sub-cluster.

[0085] Step 205: Determine the predicted output of the target region based on the predicted output of each sub-cluster. The prediction of the target region is obtained by adding the predicted outputs of each sub-cluster.

[0086] This application clusters photovoltaic power plants in the target area and selects representative plants, which not only ensures the typicality of each sub-cluster and reduces data redundancy, but also significantly improves the clustering accuracy. In the prediction stage, historical power time-series data of representative plants and neighboring plants are integrated. Graph neural networks are used to capture spatial correlation features, and multilayer perceptrons are used to mine time-series patterns. The dual models work together to enhance the depth of data information mining, which greatly improves the accuracy of power prediction for representative plants. Furthermore, based on the proportional scaling strategy of total installed capacity and representative plant installed capacity, the scientific nature of sub-cluster output conversion is ensured. By combining the prediction results of each sub-cluster, high-precision prediction of photovoltaic power in the target area is finally achieved, providing reliable data support for efficient power system dispatch, and balancing prediction efficiency and accuracy.

[0087] To verify the validity of this application, a simulation experiment was conducted based on 2,102 distributed photovoltaic power stations in a certain region, using historical 72-hour data to predict the power generation for the next 24 hours.

[0088] The normalized power output curves of the three representative power plants and three sub-clusters selected using the spatiotemporal dual-scale dynamic clustering method provided in this application are as follows: Figure 5 As shown, the power curves of representative power plants selected by the spatiotemporal dual-scale dynamic clustering method are highly similar to the power curves of the sub-cluster as a whole, which can effectively characterize the overall power generation characteristics of the sub-cluster.

[0089] Table 1 Comparison of regional prediction errors for different prediction methods

[0090]

[0091] To verify the advantages of the regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation proposed in this application, the prediction methods based on spatiotemporal dual-scale clustering followed by GNN prediction, based solely on spatial clustering followed by GNN prediction, and directly summing the power of all power plants without clustering and then using a Long Short-Term Memory (LSTM) network for prediction were compared. The prediction results are as follows: Figure 6 As shown, it is evident that the prediction results of this application are more accurate than others. Table 1 shows the prediction errors of various prediction methods across the entire region. Compared to the method based solely on spatial clustering, the root mean square error (RMSE) of this application is reduced by 38.23%. Compared to the method of directly summing the power of all power plants without clustering and using LSTM for prediction, the RMSE of this application is reduced by 44.31%, demonstrating the superiority of this application in large-scale regional distributed photovoltaic prediction.

[0092] Based on the same inventive concept, this application also provides an apparatus for implementing the method described above. The solution provided by this apparatus is similar to the solution described in the above method; therefore, specific limitations in one or more apparatus embodiments provided below can be found in the limitations of the method described above, and will not be repeated here.

[0093] In one exemplary embodiment, such as Figure 7 As shown, a regional distributed photovoltaic power prediction device based on the spatiotemporal correlation of multiple sites is provided, including: a clustering module 701, a representative power station selection module 702, a power station power prediction module 703, a cluster output determination module 704, and a regional output determination module 705.

[0094] Clustering module 701 is used to cluster distributed photovoltaic power stations within the target area to obtain multiple sub-clusters.

[0095] The representative power station selection module 702 is used to determine the representative power station of each sub-cluster based on the historical power time series of the distributed photovoltaic power stations in each sub-cluster.

[0096] The power plant power prediction module 703 is used to predict the power output of any representative power plant by using a graph neural network and a multilayer perceptron, based on the historical power time series of the representative power plant and the historical power time series of its neighboring power plants.

[0097] The cluster output determination module 704 is used to, for any sub-cluster, proportionally amplify the predicted output of the representative power station of the sub-cluster based on the total installed capacity of the distributed photovoltaic power stations within the sub-cluster and the installed capacity of the representative power station of the sub-cluster, to obtain the predicted output of the sub-cluster.

[0098] The regional output determination module 705 is used to determine the predicted output of the target region based on the predicted output of each sub-cluster.

[0099] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 8As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores the locations and historical power time series of distributed photovoltaic power stations within the target area. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a regional-level distributed photovoltaic power prediction method.

[0100] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0101] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0102] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0103] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0104] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0105] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, etc., and are not limited to these.

[0106] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation, characterized in that, The method includes: Distributed photovoltaic power stations within the target area are clustered to obtain multiple sub-clusters; Based on the historical power time series of distributed photovoltaic power stations in each sub-cluster, the representative power station of each sub-cluster is determined; For any representative power station, based on the historical power time series of the representative power station and the historical power time series of its neighboring stations, a graph neural network and a multilayer perceptron are used to predict the power output of the representative power station. Specifically, based on the historical power time series of the representative power station and the historical power time series of its neighboring stations, a graph neural network and a multilayer perceptron are used to predict the power output of the representative power station, resulting in the predicted power output of the representative power station. This includes: Each time a distributed photovoltaic power station is located within a 60° radius of the representative power station, it is selected as a neighboring station of the representative power station. Based on the historical power time series of the representative power plant and the historical power time series of each neighboring power plant, a graph neural network is used to perform multi-layer directed graph convolution operations in each time window to obtain the final hidden tensor. This includes: constructing a local graph by treating the representative power plant and each neighboring power plant as graph nodes and connecting each neighboring power plant to the representative power plant in a direction from the neighboring power plant to the representative power plant; determining a directed adjacency matrix based on the local graph, the geographical locations of the representative power plant and each neighboring power plant; adding an identity matrix to the directed adjacency matrix and performing normalization to obtain a directed propagation matrix; generating a three-dimensional tensor based on the graph nodes in the local graph and the historical power time series of each graph node; the three-dimensional tensor includes graph nodes, time, and power; and performing multi-layer directed graph convolution operations in each time window based on the directed propagation matrix and the three-dimensional tensor to obtain the final hidden tensor. Based on the final hidden tensor, a multilayer perceptron is used to predict the power output of the representative power station, thereby obtaining the predicted power output of the representative power station. For any sub-cluster, based on the total installed capacity of the distributed photovoltaic power stations within the sub-cluster and the installed capacity of the representative power station of the sub-cluster, the predicted output of the representative power station of the sub-cluster is proportionally amplified to obtain the predicted output of the sub-cluster. The predicted output of the target region is determined based on the predicted output of each sub-cluster.

2. The regional distributed photovoltaic power prediction method based on multi-site space-time correlation according to claim 1, characterized in that, Distributed photovoltaic power stations within the target area are clustered to obtain multiple sub-clusters, including: Based on the geographical location of each distributed photovoltaic power station in the target area, the K-medoids method is used to cluster the distributed photovoltaic power stations in the target area to obtain multiple sub-regions; For any sub-region, based on the historical power time series of each distributed photovoltaic power station in the sub-region, the probabilistic quality similarity between each pair of distributed photovoltaic power stations in the sub-region is calculated using the probabilistic quality similarity kernel method to obtain a similarity matrix; Based on the similarity matrix, the distributed photovoltaic power stations in the sub-region are clustered using the K-means clustering method to obtain multiple sub-clusters.

3. The regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation according to claim 2, characterized in that, The probability mass similarity between the distributed photovoltaic power station and the distributed photovoltaic power station is calculated by using the following formula: The probability mass similarity between the distributed photovoltaic power station and the distributed photovoltaic power station is calculated by using the following formula: The probability mass similarity between the distributed photovoltaic power station ； ； in, For distributed photovoltaic power stations With distributed photovoltaic power stations Probability quality similarity between them For distributed photovoltaic power stations Historical power time series, For distributed photovoltaic power stations Historical power time series, For distributed photovoltaic power stations With distributed photovoltaic power stations Dissimilarity The total historical time window length, This represents the total number of distributed photovoltaic power stations. For the first Time steps fall and Number of data points within the range For the first Time-step distributed photovoltaic power station power, For the first Time-step distributed photovoltaic power station The power.

4. The regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation according to claim 1, characterized in that, Based on the historical power time series of distributed photovoltaic power stations in each sub-cluster, the representative power stations of each sub-cluster are determined, including: For any sub-cluster, calculate the Pearson correlation coefficient between the historical power time series of each distributed photovoltaic power station in the sub-cluster and the total power series of the sub-cluster, and take the distributed photovoltaic power station with the highest Pearson correlation coefficient as the representative power station of the sub-cluster.

5. The regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation according to claim 1, characterized in that, The directed adjacency matrix is: ； ； in, For the distributed photovoltaic power station corresponding to the directed adjacency matrix With distributed photovoltaic power stations elements, For geographically based edge weights, For a set of graph nodes in a local graph, The geographical distance between the two distributed photovoltaic power stations. For the distance scale, As a representative power station The set of neighboring stations.

6. The regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation according to claim 1, characterized in that, The directed propagation matrix is ​​determined using the following formula: ； ； ； in, It is a directed adjacency matrix. It is the identity matrix. Let be a directed adjacency matrix with self-loops. For a directed propagation matrix, This is the out-degree diagonal matrix obtained by summing the rows. This represents the number of graph nodes in the local graph. for The Line number Column elements.

7. A regional distributed photovoltaic power prediction device based on multi-site spatiotemporal correlation, characterized in that, The apparatus performs the regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation as described in any one of claims 1-6, and the apparatus comprises: The clustering module is used to cluster distributed photovoltaic power stations within the target area to obtain multiple sub-clusters; The representative power station selection module is used to determine the representative power station for each sub-cluster based on the historical power time series of distributed photovoltaic power stations in each sub-cluster. The power plant power prediction module is used to predict the power output of any representative power plant by using a graph neural network and a multilayer perceptron, based on the historical power time series of the representative power plant and the historical power time series of its neighboring power plants. The cluster output determination module is used to, for any sub-cluster, proportionally amplify the predicted output of the representative power station of the sub-cluster based on the total installed capacity of the distributed photovoltaic power stations within the sub-cluster and the installed capacity of the representative power station of the sub-cluster, to obtain the predicted output of the sub-cluster. The regional output determination module is used to determine the predicted output of the target region based on the predicted output of each sub-cluster.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the regional distributed photovoltaic power prediction method based on multi-site spatiotemporal correlation according to any one of claims 1-6.

Images