Power transmission line icing prediction method based on multi-node association and graph convolutional network

Abstract

A power transmission line icing thickness prediction method based on multi-node association and a graph convolutional network relates to the technical field of icing prediction, and comprises the following steps: acquiring historical forecast meteorological data, historical icing data and topographic data of monitoring points of a power transmission line section and a spatial relationship of the monitoring points; screening the data, and then carrying out data normalization processing; converting the data into a graph structure G, and constructing a distance correlation degree adjacency matrix between the monitoring points at the same time; constructing a graph convolutional network prediction model; dividing the data into a training set and a test set, and training the graph convolution network icing thickness prediction model by using the training set; and predicting the trained icing thickness prediction model by using the test set and evaluating an icing thickness prediction result at the same time. Through the improvement of the algorithm, the characteristics of information characteristics of a plurality of local nodes are aggregated, and the meteorological data and topographic parameters of the plurality of nodes are used for constructing a graph network, so that the prediction accuracy of the power transmission line is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of icing prediction, and in particular to a method for predicting icing of power transmission lines based on multi-node association and graph convolutional networks. Background Art

[0002] Overhead line icing may cause accidents such as conductor dancing, insulator flashover, line disconnection and tripping, posing a great threat to the stable operation of the power system. In order to ensure the safe and stable operation of the power grid, it is necessary to predict line icing. The current icing prediction methods are mainly divided into physical prediction and algorithm prediction. The physical model is difficult to determine the parameters when using it, so it is difficult to use in practice. At present, algorithm prediction is the main method for icing prediction.

[0003] At present, the algorithm prediction is mainly based on the meteorological forecast data and terrain factors of a single monitoring point to predict the ice thickness, and does not consider the temporal and spatial correlation between different points. In order to improve the prediction accuracy, this patent proposes a transmission line ice thickness prediction method based on multi-point meteorological data and GCN, which can use meteorological data from multiple points and take into account the connection relationship of each monitoring point, so as to more accurately predict the line ice thickness. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for predicting icing on power transmission lines based on multi-node association and graph convolutional networks. By improving the algorithm and aggregating the characteristics of information features of multiple local nodes, the meteorological data and terrain parameters of multiple nodes are constructed into a graph network, thereby avoiding the error problem caused by a single meteorological data source, thereby greatly improving the prediction accuracy of the transmission line.

[0005] The technical solution adopted by the present invention is: a method for predicting ice thickness of transmission lines based on multi-node association and graph convolutional network, comprising: S01. Obtain the historical forecast weather data, historical ice cover data, terrain data and spatial relationship of each monitoring point of the transmission line section; S02. The data obtained in step S1 is screened and then normalized; S03. Convert the data processed in step 02 into a graph structure G, and construct a distance correlation adjacency matrix between each monitoring point; Among them, the nodes in the graph structure G are monitoring points, and the historical forecast meteorological data, historical ice cover data, and terrain data are the feature vectors of the nodes; S04. Build a graph convolutional network prediction model, determine the activation function, number of convolutional layers, and output dimension of each layer of the graph convolutional network prediction model.

[0006] S05. Divide the data processed in step 02 into a training set and a test set, and use the training set to train the graph convolutional network ice thickness prediction model, whose input is the historical forecast meteorological data and terrain data P times before the predicted time time1, and the output is the actual measured ice thickness at the predicted time time1; S06. Use the test set to predict the trained ice thickness prediction model, input the historical forecast meteorological data and terrain data of the monitoring point P times before the predicted time time2, output the ice thickness prediction result of the monitoring point at the predicted time time2, and evaluate the ice thickness prediction result; The value range of P is 2≤P≤6.

[0007] Furthermore, in step S01, the transmission line section includes K transmission lines, each transmission line includes N monitoring points, the historical forecast meteorological data, historical ice cover data, and terrain data of each monitoring point are acquired over a time span of d days, and the number of sampling points acquired each day is t, and the value range of t is 12≤t≤24; The historical forecast meteorological data include wind speed forecast value V, temperature forecast value T, humidity forecast value RH, and the terrain data include altitude h, vertical span In, windward slope α and slope β.

[0008] Furthermore, the data screening condition of step S02 is to remove the monitoring point data with ice thickness of 0 at five consecutive sampling points; In the data normalization process, 1 is used to indicate that the monitoring point has a windward slope, and 0 is used to indicate that the monitoring point does not have a windward slope.

[0009] Furthermore, the graph structure G in step S03 is expressed as: G=(W, A), Where W represents the node set, W={W k，n |1≤k≤K, 1≤n≤N}, k represents the serial number of the transmission line, and n represents the serial number of the monitoring point in the transmission line.

[0010] Furthermore, the formula of the distance association degree adjacency matrix in step S03 is: , In the above formula, represents the distance between monitoring point i and monitoring point j on line k, and λ represents the distance judgment threshold.

[0011] Furthermore, the distance judgment threshold λ has a value range of 80-140 km.

[0012] Furthermore, the number of convolutional layers of the graph convolutional network prediction model constructed in step S04 is 2, and the graph convolutional network prediction model formula is expressed as: , When k is 0, H 0 =X is the first layer input of the model, X∈R M*D , M is the number of nodes, D is the dimension of node feature vector, and f is the activation function; The expression formula of the activation function f is: F(X)=max(0,X).

[0013] Furthermore, the evaluation method of the ice thickness prediction result is as follows: comparing the predicted value of ice thickness at the time to be predicted with the actual value of ice thickness, and using mean absolute percentage error (MAPE) and root mean square error (RMSE) to evaluate the prediction performance; The mean absolute percentage error formula is: , The root mean square error formula is: , Where: B i is the actual value of ice thickness, is the predicted value of ice thickness, and NUM is the number of samples.

[0014] The beneficial effects of the present invention are as follows: by improving the algorithm, the characteristics of information features of multiple local nodes are aggregated, and the meteorological data and terrain parameters of multiple nodes are constructed into a graph network, thereby avoiding the error problem caused by a single meteorological data source, thereby greatly improving the prediction accuracy of the transmission line. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0016] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0017] See attached Figure 1 , a transmission line icing prediction method based on multi-node association and graph convolutional network, including: S01. Obtain the historical forecast weather data, historical ice cover data, terrain data and spatial relationship of each monitoring point of the transmission line section; S02. The data obtained in step S1 is screened and then normalized; S03. Convert the data processed in step 02 into a graph structure G, and construct a distance correlation adjacency matrix between each monitoring point; Among them, the nodes in the graph structure G are monitoring points, and the historical forecast meteorological data, historical ice cover data, and terrain data are the feature vectors of the nodes; S04. Build a graph convolutional network prediction model, determine the activation function, number of convolutional layers, and output dimension of each layer of the graph convolutional network prediction model.

[0018] S05. Divide the data processed in step 02 into a training set and a test set, and use the training set to train the graph convolutional network ice thickness prediction model, whose input is the historical forecast meteorological data and terrain data P times before the predicted time time1, and the output is the actual measured ice thickness at the predicted time time1; S06. Use the test set to predict the trained ice thickness prediction model, input the historical forecast meteorological data and terrain data of the monitoring point P times before the predicted time time2, output the ice thickness prediction result of the monitoring point at the predicted time time2, and evaluate the ice thickness prediction result; The value range of P is 2≤P≤6.

[0019] In step S01, the transmission line section includes K transmission lines, each transmission line includes N monitoring points, the historical forecast meteorological data, historical ice cover data, and terrain data of each monitoring point are obtained over a time span of d days, and the number of sampling points obtained each day is t, and the value range of t is 12≤t≤24; The historical forecast meteorological data include wind speed forecast value V, temperature forecast value T, humidity forecast value RH, and the terrain data include altitude h, vertical span In, windward slope α and slope β.

[0020] The data screening condition of step S02 is to remove the monitoring point data with ice thickness of 0 at five consecutive sampling points; In the data normalization process, 1 is used to indicate that the monitoring point has a windward slope, and 0 is used to indicate that the monitoring point does not have a windward slope.

[0021] The collected data were screened, and the nodes with ice cover thickness of 0 at five consecutive sampling points in the data were eliminated. The node features were normalized, with 1 indicating that the monitoring point had a windward slope, and 0 indicating that the monitoring point did not have a windward slope. The obtained data set was divided into a training set and a test set, with the training set accounting for 80% and the test set accounting for 20%.

[0022] The graph structure G in step S03 is expressed as: G=(W, A), Where W represents the node set, W={W k，n |1≤k≤K, 1≤n≤N}, k represents the serial number of the transmission line, and n represents the serial number of the monitoring point in the transmission line.

[0023] The formula of the distance association degree adjacency matrix in step S03 is: , In the above formula, represents the distance between monitoring point i and monitoring point j on line k, and λ represents the distance judgment threshold.

[0024] The value range of the distance judgment threshold λ is 80-140 km.

[0025] The data obtained in step S03 is converted into a graph structure G, where the monitoring points are nodes of the graph structure G, the historical forecast meteorological data, historical ice cover and terrain data of the monitoring points are the feature vectors of each node, and an adjacency matrix is ​​constructed based on the distance between each monitoring point. λ represents the distance judgment threshold. If the distance is less than λ, it is considered that the two nodes affect each other, and if the distance is greater than λ, it is considered that the two nodes do not affect each other.

[0026] The number of convolutional layers of the graph convolutional network prediction model constructed in step S04 is 2, and the graph convolutional network prediction model formula is expressed as: , When k is 0, H 0 =X is the first layer input of the model, X∈R M*D , M is the number of nodes, D is the dimension of node feature vector, and f is the activation function; The expression formula of the activation function f is: F(X)=max(0,X).

[0027] Construct a graph convolutional network prediction model, which consists of two layers. The first layer is the input layer. The input features of the model are the historical forecast meteorological data and terrain data of the monitoring point P times before the predicted time. Perform a graph convolution operation on the input features to extract the features of each node and convert the features into high-dimensional vectors. The second layer is the output layer, which takes the output of the first layer as input and performs a graph convolution operation on the input to further aggregate the features. At the same time, the activation function of the second layer outputs the predicted value of ice thickness.

[0028] The evaluation method of the ice thickness prediction result is as follows: the predicted value of ice thickness at the time to be predicted is compared with the actual value of ice thickness, and the prediction performance is evaluated using mean absolute percentage error (MAPE) and root mean square error (RMSE); The mean absolute percentage error formula is: , The root mean square error formula is: , Where: B i is the actual value of ice thickness, is the predicted value of ice thickness, and NUM is the number of samples.

[0029] This model shows better results than the standard prediction model in two key indicators, mean absolute percentage error (MAPE) and root mean square error (RMSE), which is manifested in smaller values ​​of these two indicators. Therefore, it can be concluded that this model performs better in ice thickness prediction. This improvement means that this model can predict ice cover more accurately, thereby providing higher quality support for related decision-making, significantly improving the reliability and accuracy of the prediction.

[0030] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A transmission line icing prediction method based on multi-node association and graph convolutional network, characterized in that: include: S01. Obtain the historical forecast weather data, historical ice cover data, terrain data and spatial relationship of each monitoring point of the transmission line section; S02. The data obtained in step S1 is screened and then normalized; S03. Convert the data processed in step 02 into a graph structure G, and construct a distance correlation adjacency matrix between each monitoring point; Among them, the nodes in the graph structure G are monitoring points, and the historical forecast meteorological data, historical ice cover data, and terrain data are the feature vectors of the nodes; S04. Build a graph convolutional network prediction model, determine the activation function, number of convolutional layers, and output dimension of each layer of the graph convolutional network prediction model; S05. Divide the data processed in step 02 into a training set and a test set, and use the training set to train the graph convolutional network ice thickness prediction model, whose input is the historical forecast meteorological data and terrain data P times before the predicted time time1, and the output is the actual measured ice thickness at the predicted time time1; S06. Use the test set to predict the trained ice thickness prediction model, input the historical forecast meteorological data and terrain data of the monitoring point P times before the predicted time time2, output the ice thickness prediction result of the monitoring point at the predicted time time2, and evaluate the ice thickness prediction result; The value range of P is 2≤P≤6.

2. The method for predicting icing of a power transmission line according to claim 1, characterized in that: In step S01, the transmission line section includes K transmission lines, each transmission line includes N monitoring points, the historical forecast meteorological data, historical ice cover data, and terrain data of each monitoring point are obtained over a time span of d days, and the number of sampling points obtained each day is t, and the value range of t is 12≤t≤24; The historical forecast meteorological data include wind speed forecast value V, temperature forecast value T, humidity forecast value RH, and the terrain data include altitude h, vertical span In, windward slope α and slope β.

3. The method for predicting icing of power transmission lines according to claim 1, characterized in that: The data screening condition of step S02 is to remove the monitoring point data with ice thickness of 0 at five consecutive sampling points; In the data normalization process, 1 is used to indicate that the monitoring point has a windward slope, and 0 is used to indicate that the monitoring point does not have a windward slope.

4. The method for predicting icing of a power transmission line according to claim 1, characterized in that: The graph structure G in step S03 is expressed as: G=(W, A), Where W represents the node set, W={W k，n |1≤k≤K, 1≤n≤N}, k represents the serial number of the transmission line, and n represents the serial number of the monitoring point in the transmission line.

5. The method for predicting icing of a power transmission line according to claim 1, characterized in that: The formula of the distance association degree adjacency matrix in step S03 is: ， In the above formula, represents the distance between monitoring point i and monitoring point j on line k, and λ represents the distance judgment threshold.

6. The method for predicting icing of a power transmission line according to claim 5, characterized in that: The value range of the distance judgment threshold λ is 80-140 km.

7. The method for predicting icing of a power transmission line according to claim 1, characterized in that: The number of convolutional layers of the graph convolutional network prediction model constructed in step S04 is 2, and the graph convolutional network prediction model formula is expressed as: ， When k is 0, H 0 =X is the first layer input of the model, X∈R M*D , M is the number of nodes, D is the dimension of node feature vector, and f is the activation function; The expression formula of the activation function f is: F(X)=max(0,X).

8. The method for predicting icing of a power transmission line according to claim 1, characterized in that: The evaluation method of the ice thickness prediction result is as follows: the predicted value of ice thickness at the time to be predicted is compared with the actual value of ice thickness, and the prediction performance is evaluated using mean absolute percentage error (MAPE) and root mean square error (RMSE); The mean absolute percentage error formula is: , The root mean square error formula is: ， Where: B i is the actual value of ice thickness, is the predicted value of ice thickness, and NUM is the number of samples.

Images