A medium voltage distribution network load forecasting method, device, terminal and medium

Abstract

The present application discloses a method, device, terminal and medium for load prediction of a medium-voltage distribution network, and relates to the technical field of medium-voltage distribution networks. The solution provided by the present application, by introducing a dynamic graph neural network combined with an adaptive adjacency matrix, can adjust the connection relationship of load nodes according to the real-time operating status, thereby improving the prediction stability and adaptability of the model in scenarios where the grid topology changes. At the same time, the dynamic graph neural network is adopted to extract the spatial correlation between load nodes from the grid topology structure, so that the prediction model can effectively learn the interaction between different nodes, thereby improving the overall accuracy of load prediction and solving the current technical problem of inaccurate load prediction in medium-voltage distribution networks.

Description

Technical Field

[0001] The present application relates to the technical field of medium-voltage distribution networks, and in particular to a method, device, terminal, and medium for predicting load in a medium-voltage distribution network. Background Art

[0002] As an important link connecting the transmission network and the low-voltage distribution system, the load forecast of the medium-voltage distribution network is crucial for grid planning, grid dispatching, equipment operation and maintenance, distributed energy access, and demand-side management.

[0003] Currently, common methods for medium-voltage distribution network load forecasting primarily include mathematical statistics, machine learning, and deep learning. Statistical methods include the autoregressive moving average (ARIMA) model, exponential smoothing, and Kalman filtering. These methods can model the temporal dependencies of loads. Machine learning methods include support vector regression (SVR), random forests (RF), and gradient boosted decision trees (GBDT). These methods can capture complex load characteristics to a certain extent. Deep learning methods include long short-term memory (LSTM), temporal convolutional networks (TCN), and Transformer architectures. These methods can better model long-term dependencies in time series and improve load forecasting accuracy.

[0004] However, due to the development of smart grids and the widespread access to new energy sources, the operation of medium-voltage distribution networks has become more complicated. The existing medium-voltage distribution network load forecasting methods have been unable to adapt to the current load forecasting needs of medium-voltage distribution networks, resulting in the current technical problem of inaccurate load forecasting of medium-voltage distribution networks. Summary of the Invention

[0005] The present application provides a medium-voltage distribution network load prediction method, device, terminal and medium, which are used to solve the current technical problem of inaccurate medium-voltage distribution network load prediction.

[0006] To solve the above technical problems, the first aspect of the present application provides a medium voltage distribution network load forecasting method, comprising:

[0007] Acquire historical operation data of the medium voltage distribution network, wherein the historical operation data includes: historical topology data, historical load data, and historical power consumption environment data;

[0008] According to the historical topology data, and in accordance with the time information contained in the historical topology data and the preset time steps, constructing a dynamic adjacency matrix corresponding to different time steps;

[0009] The load spatial feature extraction is performed on the historical load data and the dynamic adjacency matrix of the same time step through a dynamic graph neural network to obtain the spatial feature matrix corresponding to the time step, and then the spatial feature matrix is ​​input into the self-attention network to obtain the load forecast value of the future time step through the self-attention network;

[0010] Iteratively optimizing the dynamic graph neural network and the self-attention network according to the error between the historical load data and the load forecast value until a preset iterative optimization termination condition is met, thereby obtaining an optimized dynamic graph neural network and self-attention network;

[0011] A medium-voltage distribution network load forecasting model is constructed by the dynamic graph neural network and the self-attention network, so as to obtain a load forecasting result of the medium-voltage distribution network to be predicted through the operation of the medium-voltage distribution network load forecasting model.

[0012] Preferably, constructing a dynamic adjacency matrix corresponding to different time steps according to the historical topology data, time information contained in the historical topology data and a preset time step includes:

[0013] Determine, based on the historical topology data, a power grid topology distance between each pair of nodes in each time step of the historical topology data according to time information contained in the historical topology data and a preset time step;

[0014] The dynamic adjacency matrix corresponding to each time step is constructed by combining the Gaussian kernel function with the power grid topology distance.

[0015] Preferably, the expression of the load space feature extraction function in the dynamic graph neural network is specifically:

[0016]

[0017]

[0018] Where, represents the node feature matrix of layer l at time step t; represents the degree of node i at time step t, represents the degree of node j at time step t, is the trainable weight matrix of the lth layer of GCN; is the activation function, is the adjacency relationship between nodes i and j at time step t.

[0019] Preferably, the iterative optimization of the dynamic graph neural network and the self-attention network according to the error between the historical load data and the load forecast value until a preset iterative optimization termination condition is satisfied, and the optimized dynamic graph neural network and self-attention network are obtained, comprising:

[0020] According to the error between historical load data and load forecast value, combined with the preset loss function, the loss value is obtained;

[0021] When the loss value is higher than the preset loss threshold and the number of iterations does not reach the preset iteration threshold, the network parameters of the dynamic graph neural network and the self-attention network are iteratively optimized. When the loss value is lower than the preset loss threshold or the number of iterations reaches the preset iteration threshold, the optimized dynamic graph neural network and self-attention network are obtained.

[0022] Preferably, the loss function is expressed as follows:

[0023]

[0024] Where, is the loss value, N is the number of input samples, is the historical load data at time step t, is the load forecast value at time step t.

[0025] A second aspect of the present application provides a medium voltage distribution network load forecasting device, comprising:

[0026] A historical data acquisition unit is used to acquire historical operation data of the medium voltage distribution network, wherein the historical operation data includes: historical topology data, historical load data and historical power environment data;

[0027] An adjacency matrix construction unit is used to construct a dynamic adjacency matrix corresponding to different time steps according to the historical topology data, the time information contained in the historical topology data and the preset time step;

[0028] A feature extraction and prediction unit is used to extract load spatial features from historical load data and a dynamic adjacency matrix at the same time step through a dynamic graph neural network to obtain a spatial feature matrix corresponding to the time step, and then input the spatial feature matrix into a self-attention network to obtain a load forecast value for the future time step through the self-attention network;

[0029] A network training control unit, configured to iteratively optimize the dynamic graph neural network and the self-attention network according to an error between historical load data and a load forecast value, until a preset iterative optimization termination condition is met, thereby obtaining an optimized dynamic graph neural network and self-attention network;

[0030] The load forecasting model construction unit is used to construct a medium-voltage distribution network load forecasting model through the dynamic graph neural network and the self-attention network, so as to obtain the load forecasting result of the medium-voltage distribution network to be predicted through the operation of the medium-voltage distribution network load forecasting model.

[0031] Preferably, the adjacency matrix construction unit is specifically used to:

[0032] Determine, according to the historical topology data, the power grid topology distance between each pair of nodes in each time step of the historical topology data according to time information contained in the historical topology data and a preset time step;

[0033] The dynamic adjacency matrix corresponding to each time step is constructed by combining the Gaussian kernel function with the power grid topology distance.

[0034] Preferably, the network training control unit is specifically used to:

[0035] According to the error between historical load data and load forecast value, combined with the preset loss function, the loss value is obtained;

[0036] When the loss value is higher than the preset loss threshold and the number of iterations does not reach the preset iteration threshold, the network parameters of the dynamic graph neural network and the self-attention network are iteratively optimized. When the loss value is lower than the preset loss threshold or the number of iterations reaches the preset iteration threshold, the optimized dynamic graph neural network and self-attention network are obtained.

[0037] A third aspect of the present application provides a medium voltage distribution network load forecasting terminal, comprising: a memory and a processor;

[0038] The memory is used to store program code, and the program code is used to implement a medium voltage distribution network load forecasting method as provided in the first aspect;

[0039] The processor is configured to read and execute the program code.

[0040] A fourth aspect of the present application provides a computer-readable storage medium, in which a program code is stored. The program code is used to be read and executed by a processor to implement a medium-voltage distribution network load forecasting method as provided in the first aspect.

[0041] It can be seen from the above technical solutions that this application has the following advantages:

[0042] The solution provided in this application introduces a dynamic graph neural network combined with an adaptive adjacency matrix, which can adjust the connection relationship of load nodes according to the real-time operating status, thereby improving the prediction stability and adaptability of the model in scenarios where the power grid topology changes. At the same time, the dynamic graph neural network is used to extract the spatial correlation between load nodes from the power grid topology structure, so that the prediction model can effectively learn the interaction between different nodes, thereby improving the overall accuracy of load prediction and solving the current technical problem of inaccurate load prediction in medium-voltage distribution networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 This is a flow chart of an embodiment of a method for medium-voltage distribution network load forecasting provided in this application.

[0045] Figure 2 This is a structural diagram of an embodiment of a medium-voltage distribution network load forecasting device provided in this application.

[0046] Figure 3 This is a structural diagram of an embodiment of a medium-voltage distribution network load forecasting terminal provided in this application. DETAILED DESCRIPTION

[0047] Several common methods for medium-voltage distribution network load forecasting include statistical methods such as the autoregressive moving average (ARIMA) model, exponential smoothing, and Kalman filtering. These methods can model the temporal dependence of loads, but struggle to handle nonlinear load variations and are less adaptable to short-term load fluctuations. Machine learning methods include support vector regression (SVR), random forests (RF), and gradient boosted decision trees (GBDT). While these methods can capture complex load characteristics to a certain extent, they rely primarily on manual feature engineering and struggle to effectively model temporal and spatial correlations. Deep learning methods include long short-term memory (LSTM) networks, temporal convolutional networks (TCN), and Transformer architectures. These methods can better model long-term dependencies in time series and improve load forecasting accuracy. However, traditional deep learning methods typically only perform forecasts for a single load point, ignoring the distribution network topology and the interactions between load nodes, resulting in poor global consistency in forecasts.

[0048] Load nodes in distribution networks (substations, lines, and users) exhibit complex coupling relationships. However, existing forecasting methods often model loads based solely on time series data from a single load point, ignoring the impacts between adjacent nodes. For example, load fluctuations in one region may affect load patterns in adjacent areas, but traditional methods struggle to capture these spatial correlations, resulting in significant errors in forecast results.

[0049] The embodiments of the present application provide a medium-voltage distribution network load prediction method, device, terminal and medium for solving the current technical problem of inaccurate load prediction in medium-voltage distribution networks.

[0050] In order to make the purpose, features, and advantages of the invention of this application more obvious and easy to understand, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described below are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0051] First, a detailed description of an embodiment of a medium voltage distribution network load forecasting method provided by this application is as follows:

[0052] See also Figure 1 , an embodiment of the present application provides a medium voltage distribution network load forecasting method, comprising:

[0053] Step 101: Acquire historical operation data of the medium voltage distribution network.

[0054] Among them, historical operation data includes: historical topology data, historical load data and historical power consumption environment data.

[0055] It should be noted that, first of all, it is necessary to obtain the historical operating data of the medium-voltage distribution network to be predicted. These historical operating data include at least: historical topology data and historical load data. Among them, the historical topology data represents the historical topology structure data of the distribution network, the historical load data represents the historical load time series data of each distribution transformer node, and the historical power consumption environment data represents the external factors affecting the power load, such as weather, population, etc.

[0056] Step 102: construct a dynamic adjacency matrix corresponding to different time steps based on the historical topology data, according to the time information contained in the historical topology data and the preset time steps.

[0057] Step 103: Use a dynamic graph neural network to extract load spatial features from the historical load data and the dynamic adjacency matrix of the same time step to obtain a spatial feature matrix corresponding to the time step, and then input the spatial feature matrix into the self-attention network to obtain the load forecast value of the future time step through the self-attention network.

[0058] It should be noted that in steps 102 to 103, based on the historical operation data obtained in the previous step, a dynamic adjacency matrix corresponding to different time steps is constructed according to the time information contained in the historical topology data and the preset time steps to reflect the topological structure relationship of the distribution network at different times.

[0059] Next, the load data of the current time step t is processed through the dynamic graph convolutional network. and dynamic adjacency matrix After processing, the spatial feature matrix H(t) obtained not only contains the load information of the node itself, but also comprehensively considers the topological correlation characteristics of its neighboring nodes. In order to further model the dynamic dependency of load changes over time, the spatial feature matrices {H1, H2,…, H t The data is input into the self-attention network, and its self-attention mechanism is used to deeply model the time dependence of the load, thereby improving the ability to predict future load trends.

[0060] More specifically, the specific implementation of steps 102 to 103 of this embodiment includes:

[0061] 1) At each time step, a Gaussian kernel function combined with the power grid topology distance is used to construct a dynamic adjacency matrix.

[0062] The mathematical expression of the Gaussian kernel function is:

[0063] (1)

[0064] Where: i and j are two data points; Represents the topological distance between two data points; is the scale parameter of the Gaussian kernel function, which is used to control the range of the similarity measure.

[0065] The value range of this function is between (0, 1], representing the similarity weight. The closer the distance, the higher the similarity. In the distribution network topology, the Gaussian kernel function can be used to calculate the similarity between grid nodes (the weight of the adjacency matrix), ensuring that close nodes have a larger connection weight and farther nodes have a smaller weight.

[0066] The topological distance between two data points The definition of is: based on the topological structure of the power grid, the shortest path length between nodes. Its calculation formula is:

[0067] (2)

[0068] Where: is the line weight on the passing path.

[0069] The calculation formula of the dynamic adjacency matrix is:

[0070] (3)

[0071] Where: is the adjacency relationship between nodes i and j at time t, and k is any other node in the distribution network that is associated with node i.

[0072] 2) Using dynamic graph convolutional networks for feature extraction:

[0073] (4)

[0074] Where: Represents the node feature matrix of layer l at time t; represents the degree of node i at time t, which changes over time; is the trainable weight matrix of the lth layer of GCN; is the activation function.

[0075] degree The calculation formula is as follows:

[0076] (5)

[0077] Since the adjacency matrix It changes dynamically over time, so the topological relationship and influence weight between load nodes are time-dependent.

[0078] Next, we combine the self-attention network to perform load time series prediction, including:

[0079] 1) Use the self-attention mechanism to calculate the key features of the load time series and generate query Q, key K and value V matrices to capture the mutual influence between different time steps:

[0080] (6)

[0081] (7)

[0082] Where: H is the load feature matrix extracted by dynamic graph neural network; , , is the trainable parameter of Transformer; is the dimension of the key vector K.

[0083] 2) Predict future load through self-attention network, the expression is as follows:

[0084] (8)

[0085] Where, is the load forecast value at the next time step t+1, is the load data of the current time step t. Among them, DGNN represents the dynamic graph convolutional network, which is based on the load data of the current time step t. The dynamic adjacency matrix corresponding to this time step , extract the spatial feature representation of distribution network load nodes.

[0086] Step 104: Based on the error between the historical load data and the load forecast value, the dynamic graph neural network and the self-attention network are iteratively optimized until the preset iterative optimization termination condition is met, thereby obtaining the optimized dynamic graph neural network and self-attention network.

[0087] Step 105: construct a medium-voltage distribution network load forecasting model through a dynamic graph neural network and a self-attention network, so as to obtain a load forecasting result of the medium-voltage distribution network to be predicted through the operation of the medium-voltage distribution network load forecasting model.

[0088] It should be noted that in steps 104 and 105, based on the load forecast value obtained in step 103, the error between the load forecast value and the historical load data at the same time step is calculated, and the dynamic graph neural network and the self-attention network are iteratively optimized until the preset iterative optimization termination condition is met, and the optimized dynamic graph neural network and self-attention network are obtained. The optimized dynamic graph neural network and self-attention network are then integrated to obtain a medium-voltage distribution network load forecast model. The load forecast result of the medium-voltage distribution network to be predicted can be obtained through this medium-voltage distribution network load forecast model.

[0089] More specifically, the mean square error is used as the loss function, and the loss value calculated by the loss function measures the error between the predicted value and the true value. The expression of the loss function is as follows:

[0090] (9)

[0091] Where, is the loss value, N is the number of input samples, is the historical load data at time step t, is the load forecast value at time step t.

[0092] Then, when the loss value is higher than the preset loss threshold and the number of iterations does not reach the preset iteration threshold, that is, the iterative optimization termination condition is not met, the Adam optimizer is used to iteratively optimize the dynamic graph neural network and the self-attention network. When the loss value is lower than the preset loss threshold or the number of iterations reaches the preset iteration threshold, the optimized dynamic graph neural network and self-attention network are obtained, and then the optimized dynamic graph neural network and self-attention network are used to construct a medium-voltage distribution network load forecasting model, so as to obtain the load forecast result of the medium-voltage distribution network to be predicted through the operation of the medium-voltage distribution network load forecasting model.

[0093] The medium-voltage distribution network load forecasting method proposed in this application, which integrates dynamic graph neural networks and Transformer, uses dynamic graph neural networks to extract the spatial correlation between load nodes from the grid topology, enabling the prediction model to effectively learn the interaction between different nodes, thereby improving the overall accuracy of load forecasting. At the same time, the introduction of dynamic graph neural networks combined with adaptive adjacency matrices can adjust the connection relationship of load nodes according to the real-time operating status, improving the prediction stability and adaptability of the model in scenarios where the grid topology changes. Utilizing its global self-attention mechanism, it can capture the long-term dependencies of loads and improve the ability to model long-term trends, thereby improving the stability and reliability of load forecasting.

[0094] The above is a detailed description of an embodiment of a medium-voltage distribution network load prediction method provided by the present application. The following is a detailed description of an embodiment of a medium-voltage distribution network load prediction device provided by the present application.

[0095] See also Figure 2 In a second aspect, the present application provides a medium voltage distribution network load forecasting device, comprising:

[0096] The historical data acquisition unit 201 is used to acquire historical operation data of the medium voltage distribution network, wherein the historical operation data includes: historical topology data, historical load data and historical power environment data;

[0097] The adjacency matrix construction unit 202 is configured to construct a dynamic adjacency matrix corresponding to different time steps based on the historical topology data, the time information contained in the historical topology data, and a preset time step;

[0098] The feature extraction and prediction unit 203 is used to extract the load spatial features from the historical load data and the dynamic adjacency matrix of the same time step through a dynamic graph neural network to obtain the spatial feature matrix corresponding to the time step, and then input the spatial feature matrix into the self-attention network to obtain the load forecast value of the future time step through the self-attention network;

[0099] The network training control unit 204 is used to iteratively optimize the dynamic graph neural network and the self-attention network according to the error between the historical load data and the load forecast value until the preset iterative optimization termination condition is met, thereby obtaining the optimized dynamic graph neural network and the self-attention network;

[0100] The load forecasting model construction unit 205 is used to construct a medium-voltage distribution network load forecasting model through a dynamic graph neural network and a self-attention network, so as to obtain a load forecasting result of the medium-voltage distribution network to be predicted through the operation of the medium-voltage distribution network load forecasting model.

[0101] Furthermore, the adjacency matrix construction unit 202 is specifically configured to:

[0102] According to the historical topology data, according to the time information contained in the historical topology data and the preset time step, determining the power grid topology distance between each pair of nodes in the historical topology data at each time step;

[0103] The dynamic adjacency matrix corresponding to each time step is constructed by combining the Gaussian kernel function with the power grid topology distance.

[0104] Furthermore, the network training control unit 204 is specifically configured to:

[0105] According to the error between historical load data and load forecast value, combined with the preset loss function, the loss value is obtained;

[0106] When the loss value is higher than the preset loss threshold and the number of iterations has not reached the preset iteration threshold, the network parameters of the dynamic graph neural network and the self-attention network are iteratively optimized. When the loss value is lower than the preset loss threshold or the number of iterations reaches the preset iteration threshold, the optimized dynamic graph neural network and self-attention network are obtained.

[0107] In addition, if Figure 3 As shown, the present application also provides an embodiment of a medium-voltage distribution network load forecasting terminal. The terminal implementation types include: personal computers, industrial computers, servers and embedded intelligent devices. The main components of the terminal include: a memory 33 and a processor 31, wherein the memory 33 and the processor 31 can be connected via a communication bus 34;

[0108] The memory 33 is used to store program codes, and the program codes are used to implement a medium voltage distribution network load forecasting method provided in the above embodiment;

[0109] The processor 31 is used to read and execute program codes.

[0110] The present application provides an embodiment of a computer-readable storage medium, in which program code is stored. The program code is used to be read and executed by a processor to implement a medium-voltage distribution network load forecasting method as described in the above embodiment.

[0111] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the terminals, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0112] In the several embodiments provided in this application, it should be understood that the disclosed terminals, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0113] The terms "first," "second," "third," "fourth," and the like (if any) in the specification of the present application and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present application described herein, for example, can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0114] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0115] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0116] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0117] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0118] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting load of a medium voltage distribution network, characterized in that: include: Acquire historical operation data of the medium voltage distribution network, wherein the historical operation data includes: historical topology data, historical load data, and historical power consumption environment data; According to the historical topology data, and in accordance with the time information contained in the historical topology data and the preset time steps, constructing a dynamic adjacency matrix corresponding to different time steps; The load spatial feature extraction is performed on the historical load data and the dynamic adjacency matrix of the same time step through a dynamic graph neural network to obtain the spatial feature matrix corresponding to the time step, and then the spatial feature matrix is ​​input into the self-attention network to obtain the load forecast value of the future time step through the self-attention network; Iteratively optimizing the dynamic graph neural network and the self-attention network according to the error between the historical load data and the load forecast value until a preset iterative optimization termination condition is met, thereby obtaining an optimized dynamic graph neural network and self-attention network; A medium-voltage distribution network load forecasting model is constructed by the dynamic graph neural network and the self-attention network, so as to obtain a load forecasting result of the medium-voltage distribution network to be predicted through the operation of the medium-voltage distribution network load forecasting model.

2. A method for predicting load of a medium voltage distribution network according to claim 1, characterized in that: The constructing of dynamic adjacency matrices corresponding to different time steps according to the historical topology data, time information contained in the historical topology data and preset time steps includes: Determine, according to the historical topology data, the power grid topology distance between each pair of nodes in each time step of the historical topology data according to time information contained in the historical topology data and a preset time step; The dynamic adjacency matrix corresponding to each time step is constructed by combining the Gaussian kernel function with the power grid topology distance.

3. A method for predicting load of a medium voltage distribution network according to claim 2, characterized in that: The expression of the load space feature extraction function in the dynamic graph neural network is specifically: Where, represents the node feature matrix of layer l at time step t; represents the degree of node i at time step t, represents the degree of node j at time step t, is the trainable weight matrix of the lth layer of GCN; is the activation function, is the adjacency relationship between nodes i and j at time step t.

4. A method for predicting load of a medium voltage distribution network according to claim 1, characterized in that: The iterative optimization of the dynamic graph neural network and the self-attention network according to the error between the historical load data and the load forecast value until a preset iterative optimization termination condition is satisfied, and the optimized dynamic graph neural network and self-attention network are obtained, including: According to the error between historical load data and load forecast value, combined with the preset loss function, the loss value is obtained; When the loss value is higher than the preset loss threshold and the number of iterations does not reach the preset iteration threshold, the network parameters of the dynamic graph neural network and the self-attention network are iteratively optimized. When the loss value is lower than the preset loss threshold or the number of iterations reaches the preset iteration threshold, the optimized dynamic graph neural network and self-attention network are obtained.

5. A method for predicting load of a medium voltage distribution network according to claim 4, characterized in that: The loss function is specifically expressed as: Where, is the loss value, N is the number of input samples, is the historical load data at time step t, is the load forecast value at time step t.

6. A medium voltage distribution network load forecasting device, characterized in that: include: A historical data acquisition unit is used to acquire historical operation data of the medium voltage distribution network, wherein the historical operation data includes: historical topology data, historical load data and historical power environment data; An adjacency matrix construction unit is used to construct a dynamic adjacency matrix corresponding to different time steps according to the historical topology data, the time information contained in the historical topology data and the preset time step; A feature extraction and prediction unit is used to extract load spatial features from historical load data and a dynamic adjacency matrix at the same time step through a dynamic graph neural network to obtain a spatial feature matrix corresponding to the time step, and then input the spatial feature matrix into a self-attention network to obtain a load forecast value for the future time step through the self-attention network; A network training control unit, configured to iteratively optimize the dynamic graph neural network and the self-attention network according to an error between historical load data and a load forecast value, until a preset iterative optimization termination condition is met, thereby obtaining an optimized dynamic graph neural network and self-attention network; The load forecasting model construction unit is used to construct a medium-voltage distribution network load forecasting model through the dynamic graph neural network and the self-attention network, so as to obtain the load forecasting result of the medium-voltage distribution network to be predicted through the operation of the medium-voltage distribution network load forecasting model.

7. A medium voltage distribution network load forecasting device according to claim 6, characterized in that: The adjacency matrix construction unit is specifically used for: Determine, according to the historical topology data, the power grid topology distance between each pair of nodes in each time step of the historical topology data according to time information contained in the historical topology data and a preset time step; The dynamic adjacency matrix corresponding to each time step is constructed by combining the Gaussian kernel function with the power grid topology distance.

8. A medium voltage distribution network load forecasting device according to claim 6, characterized in that: The network training control unit is specifically used for: According to the error between historical load data and load forecast value, combined with the preset loss function, the loss value is obtained; When the loss value is higher than the preset loss threshold and the number of iterations does not reach the preset iteration threshold, the network parameters of the dynamic graph neural network and the self-attention network are iteratively optimized. When the loss value is lower than the preset loss threshold or the number of iterations reaches the preset iteration threshold, the optimized dynamic graph neural network and self-attention network are obtained.

9. A medium voltage distribution network load forecasting terminal, characterized in that: include: memory and processor; The memory is used to store program code, and the program code is used to implement a medium voltage distribution network load forecasting method according to any one of claims 1 to 5; The processor is configured to read and execute the program code.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program code, and the program code is used to be read and executed by a processor to implement a medium-voltage distribution network load forecasting method according to any one of claims 1 to 5.

Images