Power distribution network load prediction method and device based on multistage information interaction transmission, and storage medium

Abstract

The invention discloses a power distribution network load prediction method and device based on multistage information interaction transmission and a storage medium, and relates to the technical field of power systems. According to the load type historical power data of the plurality of power distribution network areas, dividing the power distribution network area to be measured to obtain a sub-area division result; constructing a multi-level graph structure system according to a sub-region division result, calculating a node association relationship of each level, and forming a corresponding level adjacent matrix according to the node association relationship of each level; and taking the adjacent matrix of the corresponding hierarchy and the load type historical power data of the to-be-tested power distribution network region as input, and outputting through the hierarchical graph convolutional neural network prediction model to obtain a target prediction result. According to the method, on the basis of distribution network area load division, the space-time relevance of each level is mined, distribution network load prediction is realized, and the prediction precision is effectively improved.

Description

Technical Field

[0001] The invention relates to a distribution network load prediction method, device and storage medium based on multi-level information interactive transmission, and belongs to the technical field of power systems. Background Art

[0002] In recent years, with the rapid development of distributed photovoltaic and electric vehicle technologies, their penetration rate in urban distribution networks has increased significantly, which has undoubtedly increased the complexity of net load forecasting and operation regulation of distribution networks. It is particularly noteworthy that the information collection obstacles faced by distributed photovoltaic power stations, such as the lack of meteorological data at some sites due to the lack of dedicated meteorological monitoring equipment, have greatly hindered the accurate prediction of photovoltaic power, making it difficult to directly apply traditional centralized photovoltaic prediction models to distributed scenarios. Electric vehicle load forecasting has a high degree of uncertainty in its charging behavior and difficult-to-capture spatiotemporal distribution characteristics, which constitute the main difficulty in the forecasting process.

[0003] There is a close and complex mutual influence relationship between distributed photovoltaic power generation, electric vehicle charging behavior and users' daily electricity load. Specifically, the output of distributed photovoltaic power generation, as a renewable energy supply source, may have a direct impact on the charging behavior of electric vehicle users due to its volatility and intermittency. When photovoltaic power generation is sufficient, electric vehicle users may be more inclined to use these clean energy sources for charging, thereby reducing dependence on and cost of traditional power grids. At the same time, the charging load of electric vehicles is also an important component of the load characteristics of distribution networks. With the popularization of electric vehicles, the distribution of their charging needs in time and space will cause additional pressure and challenges to the distribution network. The coupling effect of this charging load and distributed photovoltaic output makes the load characteristics of the distribution network more complex and changeable. Furthermore, the daily electricity load of users is also affected by the joint influence of distributed photovoltaic and electric vehicle charging loads. When photovoltaic power generation meets or even exceeds the basic electricity demand of users, the excess electricity may be used to charge electric vehicles, thereby reducing the electricity load on traditional power grids. Conversely, when photovoltaic power generation is insufficient or during the peak period of electric vehicle charging, the distribution network may need to dispatch electricity from other energy sources to meet the demand, thereby increasing the overall electricity load. In summary, distributed photovoltaics, electric vehicle charging behavior, and users' daily electricity load form an interdependent and mutually influential system. By deeply understanding the relationship between them, we can better plan and manage the power system and achieve efficient use of energy and sustainable development.

[0004] In the current field of distribution network load forecasting, two strategies are generally adopted: one is to forecast distributed photovoltaic, electric vehicle charging load and user load as independent individuals, and aggregate them to obtain the total load of the distribution network; the other is to try to include these three in the same framework for joint forecasting. However, the former strategy ignores the close connection within various load clusters, resulting in a significant waste of information resources; while the latter, although intended to integrate, may ignore the potential interactions and influences between different load types, and also fails to fully explore the forecasting potential. Both of them fail to deeply consider the hierarchical structure characteristics unique to distributed photovoltaics, thus limiting the possibility of further exploring the complex correlation between photovoltaic power stations, and ultimately affecting the overall accuracy and reliability of load forecasting. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a distribution network load prediction method, device and storage medium based on multi-level information interaction transmission. On the basis of the load division of the distribution network substation, the spatiotemporal correlation of each level is explored to realize the distribution network load prediction, thereby effectively improving the prediction accuracy.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions: In a first aspect, the present invention provides a distribution network load prediction method based on multi-level information interaction transmission, which obtains the load type historical power data of the distribution network area to be tested, wherein the distribution network area to be tested includes multiple distribution network substations, and the load types include electric vehicle charging loads, distributed photovoltaic power generation loads, and user power loads in all distribution network substations; The distribution network area to be tested is divided according to the load type historical power data of multiple distribution network areas to obtain a sub-area division result; Construct a multi-level graph structure system based on the sub-region division results; According to the multi-level graph structure system, the node association relationship of each level is calculated, and the corresponding level adjacency matrix is ​​constructed according to the node association relationship of each level; The corresponding hierarchical adjacency matrix and the historical power data of the load type in the distribution network area to be tested are taken as input, and the target prediction result is obtained through the output of the hierarchical graph convolutional neural network prediction model. The target prediction result is the predicted value of the load type in the distribution network area to be tested.

[0007] Furthermore, the sub-region division result obtained by dividing the distribution network area to be tested according to the load type historical power data of multiple distribution network areas is realized by clustering analysis using the fuzzy C-means algorithm, which specifically includes: Set the value range of the cluster number K; Traversing each cluster value within the value range, clustering and analyzing the load type historical data using the fuzzy C-means algorithm, and obtaining the fuzzy clusters under each K value and the fuzzy membership of each distribution network substation to be tested to the center of the fuzzy cluster; Calculate the DBI index of the fuzzy cluster under each K value; A fuzzy cluster with the smallest DBI index is selected as the sub-region division result, wherein the fuzzy cluster includes multiple sub-regions, and the sub-regions include multiple distribution network substations to be tested.

[0008] Furthermore, the calculation expression of the DBI index of the fuzzy cluster under the K value is: ; in, Indicates The average internal distance of sub-regions, Indicates The number of historical power data points of load type in each sub-area, Indicates One of the historical power data points of the load type in the sub-area, Indicates The central data point of each sub-region; Indicates sub-regions and The center distance of each sub-region is Indicates The central data point of each sub-region; Indicates sub-regions and The relative distance of the sub-regions, Indicates The average internal distance of each sub-region; Indicates The maximum relative distance of the sub-regions, Indicates the number of sub-regions.

[0009] Furthermore, the multi-level graph structure system is a three-level graph structure system, and the three-level graph structure system includes a second-level sub-graph, a first-level sub-graph and a global graph, wherein the second-level sub-graph has electric vehicle charging loads, distributed photovoltaic power generation loads and user power loads in distribution network substations as nodes; the first-level sub-graph has each distribution network substation in the sub-region as a node; and the global graph has each sub-region as a node.

[0010] Furthermore, the node association relationships of each level are calculated according to the multi-level graph structure system, and the corresponding level adjacency matrix is ​​formed according to the node association relationships of each level, including: The linear correlation between the electric vehicle charging load, distributed photovoltaic power generation load, and user power load in the distribution network area is calculated in the secondary subgraph, and is used as the edge weight to form a secondary subgraph adjacency matrix. The expression of the secondary subgraph adjacency matrix is: ; Among them, A represents the secondary subgraph adjacency matrix, subscript 1 represents the electric vehicle charging load, subscript 2 represents the distributed photovoltaic power generation load, and subscript 3 represents the user power load. Indicates the load type i and load type j The edge weights between i , j Take 1, 2, 3; In the first-level subgraph, the weighted average values ​​of the spatial proximity of each distribution network area and the similarity of the load curve are calculated and used as the edge weights to form the first-level subgraph adjacency matrix. The expression of the first-level subgraph adjacency matrix is: ; in, represents the first-level subgraph adjacency matrix, n represents the number of distribution network substations in the sub-region, Indicates i The distribution network area and j The edge weights between distribution network substations; The similarity between each sub-region is calculated in the global graph and used as the edge weight to form the global graph adjacency matrix. The expression of the global graph adjacency matrix is: ; in, represents the global graph adjacency matrix, m represents the number of sub-regions, Indicates i sub-regions and j The edge weights between sub-regions.

[0011] Furthermore, the linear correlation between the electric vehicle charging load, the distributed photovoltaic power generation load, and the user power load in the distribution network area, the similarity of the load curves of each distribution network area, and the similarity between each sub-area are all calculated by the Pearson correlation coefficient; The calculation expression of the spatial proximity of each distribution network area is: ; in, Indicates the distribution network area Geographical location, Indicates the distribution network area Geographical location, Indicates the distribution network area and The spatial proximity of represents the bandwidth parameter of the kernel function, represents the Euclidean distance; The calculation expression of the weighted average of the spatial proximity of the distribution network substations and the load curve similarity is: ; in, It represents the weighted average value of the spatial proximity of distribution network stations and the similarity of load curves. Indicates the similarity of load curves in distribution network areas, , are weight parameters, and .

[0012] Furthermore, the geographical location of the distribution network substation is represented by longitude and latitude or three-dimensional coordinates.

[0013] Furthermore, the structure of the hierarchical graph convolutional neural network prediction model includes a secondary sub-map level, a primary sub-map level and a global map level, the secondary sub-map level includes two serially connected graph attention network modules, the primary sub-map level includes an enhanced graph attention network module, and the global map level includes two serially connected graph attention network modules.

[0014] In a second aspect, the present invention further provides a distribution network load forecasting device based on multi-level information interactive transmission, comprising: A data acquisition module is configured to acquire historical power data of load types in a distribution network area to be tested, wherein the distribution network area to be tested includes multiple distribution network substations, and the load types include electric vehicle charging loads, distributed photovoltaic power generation loads, and user power loads in all distribution network substations; A region division module is configured to divide the distribution network area to be tested according to the load type historical power data of multiple distribution network areas to obtain a sub-region division result; A multi-level graph structure system building module is configured to build a multi-level graph structure system according to the sub-region division result; A hierarchical adjacency matrix construction module is configured to calculate the node association relationship of each level according to the multi-level graph structure system, and construct a corresponding hierarchical adjacency matrix according to the node association relationship of each level; The distribution network area load forecasting module is configured to take the corresponding hierarchical adjacency matrix and the historical power data of the load type in the distribution network area to be tested as input, and obtain the target prediction result through the output of the hierarchical graph convolutional neural network prediction model, and the target prediction result is the load type prediction value of the distribution network area to be tested.

[0015] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a distribution network load forecasting method based on multi-level information interaction transmission as described in any one of the above items.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention utilizes a distribution network area load clustering algorithm that considers similar characteristics of the total area load (electric vehicle charging load, distributed photovoltaic, user load) to divide multiple areas into multiple sub-areas with internal consistency; constructs a three-level hierarchical graph structure system of hierarchical sub-graphs and global graphs according to the clustering results; implements multi-level node association relationship representation according to the multi-level architecture graph structure system; uses a graph attention network model architecture, applies corresponding graph attention network layers at each level, deeply mines spatiotemporal features, and gradually aggregates information from local (secondary sub-graphs) to global (global graphs) to achieve distribution network load prediction; The present invention uses the hierarchical structure of the load data of the distribution network to improve the regional power prediction, while considering the correlation between the output within and between classes, capturing the temporal and spatial correlation between stations in a deeper level, overcoming the defect that the GCN prediction method is limited to the first-order neighbor information and cannot capture the hierarchical information in the graph, which is conducive to improving the prediction accuracy; When dealing with the lack of meteorological measured or forecast data, the present invention considers the deep temporal and spatial correlation of distribution network load to effectively improve the distribution network load prediction accuracy, which is beneficial to reducing the difficulty of distribution network net load prediction and distribution network operation control, and ensuring the safe and stable operation of the active distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a method for load forecasting of a distribution network based on multi-level information interaction transmission in an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a multi-level graph structure system of a distribution network load prediction method based on multi-level information interaction transmission in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention. Example 1

[0019] like Figure 1 As shown, an embodiment of the present invention provides a distribution network load forecasting method based on multi-level information interaction transmission, comprising the following steps: The load type historical power data of the distribution network area to be tested is obtained. The distribution network area to be tested includes multiple distribution network substations. The load types include electric vehicle charging loads, distributed photovoltaic power generation loads, and user power loads in all distribution network substations.

[0020] Since the load of distribution networks often exhibits characteristics that are closely related to geographical location, seasonal changes, time patterns, and socio-economic activities. Distribution network substations located in the same region or with similar socio-economic characteristics often show similar trends in load changes during the same time period. This similarity is not only reflected in the shape of the daily load curve, but also in the periodicity of weekly, monthly, and even seasonal load changes, the so-called "space-time correlation". In particular, in the absence of detailed individual user load data or accurate meteorological forecast information, the accuracy of load forecasting can be significantly improved by drawing on load data from adjacent substations or similar historical periods.

[0021] Therefore, in order to further explore the correlation between distribution network substations, in the load power prediction of distribution network substations, a large area is generally divided into several sub-areas according to the correlation of the load output power of the substations. In this embodiment, cluster analysis is performed using the fuzzy C-means algorithm to divide the distribution network area to be tested according to the load type historical power data of multiple distribution network substations to obtain the sub-area division results, specifically: Set a reasonable value range of the cluster number K. It should be noted that K is a positive integer.

[0022] Traversing each K value within the above reasonable value range, the fuzzy C-means algorithm is applied to cluster the historical power data of load types to obtain the fuzzy clusters under each K value and the fuzzy membership of each distribution network area to be tested to the center of the fuzzy cluster. The fuzzy cluster includes multiple sub-areas, and the sub-areas include multiple distribution network areas to be tested.

[0023] In order to determine the optimal clustering book K, the clustering effect evaluation index DBI index (Davies-Bouldin Index) is introduced. The DBI index measures the clustering effect by comparing the distances between different clusters and the internal distances of different clusters. Its calculation expression is as follows: ; in, Indicates The average internal distance of sub-regions, Indicates The number of historical power data points of load type in each sub-area, Indicates One of the historical power data points of the load type in the sub-area, Indicates The central data point of each sub-region; Indicates sub-regions and The center distance of each sub-region is Indicates The central data point of each sub-region; Indicates sub-regions and The relative distance of the sub-regions, Indicates The average internal distance of each sub-region; Indicates The maximum relative distance of the sub-regions, Indicates the number of sub-regions.

[0024] By evaluating the clustering results under different K values, the smaller the value of the DBI index, the better the clustering effect, and identifying the K value with the best clustering effect, that is, the highest consistency within the cluster and the greatest difference between clusters. Based on this evaluation result, the optimal K value was selected, and the FCM clustering algorithm was re-executed under this K value to obtain the sub-region division result.

[0025] A multi-level graph structure system is constructed based on the sub-region division results. In this embodiment, Figure 2 As shown, the multi-level graph structure system is a three-level graph structure system, and the three-level graph structure system includes a second-level sub-graph, a first-level sub-graph and a global graph.

[0026] The secondary sub-graph uses the electric vehicle charging load, distributed photovoltaic power generation load and user power load within a single distribution network area as nodes, deepens and refines the load characteristics, keenly captures and displays the intricate mutual influence and dependency between different load types, and provides a more detailed perspective for distribution system analysis.

[0027] The first-level subgraph uses multiple distribution network substations in the sub-region as nodes to identify areas with similar demand characteristics, effectively reducing the computational complexity and making the model more efficient when processing large-scale data. It also promotes the model's trainability and scalability.

[0028] The global map uses sub-regions as nodes, providing a macro and comprehensive observation perspective for the entire system. It can provide deep insights into the interactions and influences between different sub-communities, thereby improving the accuracy and stability of the prediction model at a higher level, and providing strong support for the overall planning and optimization decisions of the system.

[0029] Next, the node association relationships of each level are calculated according to the multi-level graph structure system, and the corresponding level adjacency matrix is ​​constructed according to the node association relationships of each level. Specifically: In the secondary subgraph hierarchy, the complex interaction mechanism between the three key load types, namely electric vehicle charging load, distributed photovoltaic power generation load, and user power load, is deeply explored. In order to quantify the degree of linear correlation between them, the Pearson correlation coefficient is used, and the linear correlation between the three load types is used as the weight of the edge to form the secondary subgraph adjacency matrix.

[0030] The secondary subgraph adjacency matrix is ​​3×3, and its expression is: ; Among them, A represents the secondary subgraph adjacency matrix, subscript 1 represents the electric vehicle charging load, subscript 2 represents the distributed photovoltaic power generation load, and subscript 3 represents the user power load. Indicates the load type i and load type j The edge weights between i , j Take 1, 2, 3.

[0031] In the first-level sub-map hierarchy, the focus is on the internal connection between substations, especially the proximity of geographical locations and the similarity of load patterns. For this purpose, a flexible algorithm, the Gaussian kernel function, is introduced to evaluate the spatial proximity between distribution network substations. The calculation expression is: ; in, Indicates the distribution network area Geographical location, Indicates the distribution network area The geographical location can be expressed as latitude, longitude or other spatial coordinates. Indicates the distribution network area and The spatial proximity of Represents the bandwidth parameter of the kernel function, which is used to control the smoothness of the function. represents the Euclidean distance.

[0032] At the same time, the Pearson correlation coefficient is also used to measure the similarity of the load curves of different distribution network areas. That is, the sum of the three load types in the distribution network area is calculated by the Pearson correlation coefficient expression. Finally, the weighted average method is used to deeply integrate the spatial proximity between the distribution network areas and the similarity of the load curves of different distribution network areas. The calculation expression is: ; in, It represents the weighted average value of the spatial proximity of distribution network stations and the similarity of load curves. Indicates the similarity of load curves in distribution network areas, , are weight parameters, and .

[0033] The fused weighted average is used as the edge weight to construct the first-level subgraph adjacency matrix.

[0034] The first-level subgraph adjacency matrix is ​​n×n, and its expression is: ; in, represents the first-level subgraph adjacency matrix, n represents the number of distribution network substations in the sub-region, Indicates i The distribution network area and j The weight value of the edge between distribution network substations.

[0035] Such a first-level sub-map can not only accurately capture the similar demands of distribution network substations, but also effectively reveal the potential mutual influence between them, ensuring the comprehensiveness and accuracy of the evaluation, and providing strong data support and decision-making basis for subsequent power system planning and scheduling.

[0036] In the global diagram, the perspective is raised to the height of the entire distribution system, focusing on the similarity between sub-regions. The similarity between sub-regions is calculated by the Pearson correlation coefficient of the sum of the three load types of all distribution network substations in the sub-region.

[0037] In this embodiment, the general formula of the Pearson correlation coefficient is as follows: ; in, represents the Pearson correlation coefficient, , Indicates the historical power data of load type, , represents the average value of historical power data of the load type, and N represents the number of historical power data points of the load type.

[0038] The similarity between sub-regions is converted into edge weights to form a global adjacency matrix. The global adjacency matrix is ​​m×m, and its expression is: ; in, represents the global graph adjacency matrix, m represents the number of sub-regions, Indicates i sub-regions and j The weight of the edge between the sub-regions.

[0039] The higher the similarity of the sub-regions, the tighter the connection between them and the higher the weight. This design not only provides macro-level insights into the distribution system, but also significantly enhances the ability of the global graph to capture the mutual influence between sub-regions.

[0040] Finally, the historical power data of the load type in the sub-area, the second-level subgraph adjacency matrix, the first-level subgraph adjacency matrix and the global graph adjacency matrix are taken as input, and the target prediction result is obtained through the output of the pre-built hierarchical graph convolutional neural network prediction model. The target prediction result is the predicted value of the load type in the distribution network area to be tested.

[0041] The construction method of the hierarchical graph convolutional neural network prediction model includes a secondary sub-map level, a primary sub-map level and a global map level. The secondary sub-map level includes two serially connected graph attention network modules, the primary sub-map level includes an enhanced graph attention network module, and the global map level includes two serially connected graph attention network modules. Example 2

[0042] Based on Example 1, this embodiment provides a distribution network load prediction device based on multi-level information interaction transmission, including: A data acquisition module is configured to acquire historical power data of load types in a distribution network area to be tested, wherein the distribution network area to be tested includes multiple distribution network substations, and the load types include electric vehicle charging loads, distributed photovoltaic power generation loads, and user power loads in all distribution network substations; A region division module is configured to divide the distribution network area to be tested according to the load type historical power data of multiple distribution network areas to obtain a sub-region division result; A multi-level graph structure system building module is configured to build a multi-level graph structure system according to the sub-region division result; A hierarchical adjacency matrix construction module is configured to calculate the node association relationship of each level according to the multi-level graph structure system, and construct a corresponding hierarchical adjacency matrix according to the node association relationship of each level; The distribution network area load forecasting module is configured to take the corresponding hierarchical adjacency matrix and the historical power data of the load type in the distribution network area to be tested as input, and obtain the target prediction result through the output of the hierarchical graph convolutional neural network prediction model, and the target prediction result is the load type prediction value of the distribution network area to be tested. Example 3

[0043] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for load forecasting of a distribution network based on multi-level information interaction and transmission as described in any one of the above items is implemented.

[0044] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0045] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0046] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0047] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0048] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A distribution network load forecasting method based on multi-level information interactive transmission, characterized in that: Obtaining historical power data of load types in a distribution network area to be tested, wherein the distribution network area to be tested includes multiple distribution network substations, and the load types include electric vehicle charging loads, distributed photovoltaic power generation loads, and user power loads; The distribution network area to be tested is divided according to the load type historical power data of multiple distribution network areas to obtain a sub-area division result; Construct a multi-level graph structure system based on the sub-region division results; According to the multi-level graph structure system, the node association relationship of each level is calculated, and the corresponding level adjacency matrix is ​​constructed according to the node association relationship of each level; The corresponding hierarchical adjacency matrix and the historical power data of the load type in the distribution network area to be tested are taken as input, and the target prediction result is obtained through the output of the hierarchical graph convolutional neural network prediction model. The target prediction result is the predicted value of the load type in the distribution network area to be tested.

2. The method for load forecasting of distribution network based on multi-level information interaction transmission according to claim 1 is characterized in that: The sub-region division result obtained by dividing the distribution network area to be tested according to the load type historical power data of multiple distribution network areas is realized by clustering analysis using the fuzzy C-means algorithm, specifically including: Set the value range of the cluster number K; Traversing each cluster value within the value range, clustering and analyzing the load type historical data using the fuzzy C-means algorithm, and obtaining the fuzzy clusters under each K value and the fuzzy membership of each distribution network substation to be tested to the center of the fuzzy cluster; Calculate the DBI index of the fuzzy cluster under each K value; A fuzzy cluster with the smallest DBI index is selected as the sub-region division result, wherein the fuzzy cluster includes multiple sub-regions, and the sub-regions include multiple distribution network substations to be tested.

3. The method for load forecasting of distribution network based on multi-level information interaction transmission according to claim 2 is characterized in that: The calculation expression of the DBI index of the fuzzy cluster under the K value is: ； in, Indicates The average internal distance of sub-regions, Indicates The number of historical power data points of load type in each sub-area, Indicates One of the historical power data points of the load type in the sub-area, Indicates The central data point of each sub-region; Indicates sub-regions and The center distance of each sub-region is Indicates The central data point of each sub-region; Indicates sub-regions and The relative distance of the sub-regions, Indicates The average internal distance of each sub-region; Indicates The maximum relative distance of the sub-regions, Indicates the number of sub-regions.

4. The method for load forecasting of distribution network based on multi-level information interaction transmission according to claim 1, characterized in that: The multi-level graph structure system is a three-level graph structure system, and the three-level graph structure system includes a second-level sub-graph, a first-level sub-graph and a global graph, wherein the second-level sub-graph takes the electric vehicle charging load, distributed photovoltaic power generation load and user power load in the distribution network substation as nodes; the first-level sub-graph takes each distribution network substation in the sub-region as a node; and the global graph takes each sub-region as a node.

5. The method for load forecasting of distribution network based on multi-level information interaction transmission according to claim 1, characterized in that: The calculation of the node association relationship of each level according to the multi-level graph structure system and the formation of the corresponding level adjacency matrix according to the node association relationship of each level include: The linear correlation between the electric vehicle charging load, distributed photovoltaic power generation load, and user power load in the distribution network area is calculated in the secondary subgraph, and is used as the edge weight to form a secondary subgraph adjacency matrix. The expression of the secondary subgraph adjacency matrix is: ； Among them, A represents the secondary subgraph adjacency matrix, subscript 1 represents the electric vehicle charging load, subscript 2 represents the distributed photovoltaic power generation load, and subscript 3 represents the user power load. Indicates load type i and load type j The edge weights between i , j Take 1, 2, 3; In the first-level subgraph, the weighted average values ​​of the spatial proximity of each distribution network area and the similarity of the load curve are calculated and used as the edge weights to form the first-level subgraph adjacency matrix. The expression of the first-level subgraph adjacency matrix is: ； in, represents the first-level subgraph adjacency matrix, n represents the number of distribution network substations in the sub-region, Indicates i The distribution network area and j The edge weights between distribution network substations; The similarity between each sub-region is calculated in the global graph and used as the edge weight to form the global graph adjacency matrix. The expression of the global graph adjacency matrix is: ； in, represents the global graph adjacency matrix, m represents the number of sub-regions, Indicates i sub-regions and j The edge weights between sub-regions.

6. The method for load forecasting of distribution network based on multi-level information interaction transmission according to claim 5 is characterized in that: The linear correlation between the electric vehicle charging load, distributed photovoltaic power generation load, and user power load in the distribution network area, the similarity of the load curves of each distribution network area, and the similarity between each sub-area are all calculated by the Pearson correlation coefficient; The calculation expression of the spatial proximity of each distribution network area is: ； in, Indicates the distribution network area Geographical location, Indicates the distribution network area Geographical location, Indicates the distribution network area and The spatial proximity of represents the bandwidth parameter of the kernel function, represents the Euclidean distance; The calculation expression of the weighted average of the spatial proximity of the distribution network substations and the similarity of the load curves is: ； in, It represents the weighted average value of the spatial proximity of distribution network stations and the similarity of load curves. Indicates the similarity of load curves in distribution network areas, , are weight parameters, and .

7. The method for load forecasting of distribution network based on multi-level information interaction transmission according to claim 6 is characterized in that: The geographical location of the distribution network substation is represented by longitude and latitude or three-dimensional coordinates.

8. The method for load forecasting of distribution network based on multi-level information interaction transmission according to claim 1, characterized in that: The structure of the hierarchical graph convolutional neural network prediction model includes a secondary sub-map level, a primary sub-map level and a global map level. The secondary sub-map level includes two serially connected graph attention network modules, the primary sub-map level includes an enhanced graph attention network module, and the global map level includes two serially connected graph attention network modules.

9. A distribution network load forecasting device based on multi-level information interactive transmission, characterized in that: include: A data acquisition module is configured to acquire historical power data of load types in a distribution network area to be tested, wherein the distribution network area to be tested includes multiple distribution network substations, and the load types include electric vehicle charging loads, distributed photovoltaic power generation loads, and user power loads in all distribution network substations; A region division module is configured to divide the distribution network area to be tested according to the load type historical power data of multiple distribution network areas to obtain a sub-region division result; A multi-level graph structure system building module is configured to build a multi-level graph structure system according to the sub-region division result; A hierarchical adjacency matrix construction module is configured to calculate the node association relationship of each level according to the multi-level graph structure system, and construct a corresponding hierarchical adjacency matrix according to the node association relationship of each level; The distribution network area load forecasting module is configured to take the corresponding hierarchical adjacency matrix and the historical power data of the load type in the distribution network area to be tested as input, and obtain the target prediction result through the output of the hierarchical graph convolutional neural network prediction model, and the target prediction result is the load type prediction value of the distribution network area to be tested.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the distribution network load forecasting method based on multi-level information interaction transmission as described in any one of claims 1 to 8 is implemented.

Images