An industry electricity consumption prediction method based on an industry relationship complex network

Abstract

The application discloses an industry electricity consumption prediction method based on an industry relationship complex network, calculates a Pearson correlation coefficient between each industry by using an industry electricity consumption dataset, obtains a distance matrix by measuring a distance through the correlation coefficient, obtains a minimum spanning tree (MST) by using a Prim algorithm based on the distance matrix, and further constructs a plane maximum filtering graph (PMFG) to more accurately extract a core structure and a key connection in the industry network, so that a dependence relationship between industries can be better understood; the industries are divided into different communities by using a Louvain algorithm to perform community classification, a target industry is determined, relevant industries are selected according to a community classification result, a multi-graph convolution and long short-term memory model (MGC-LSTM) is constructed, electricity consumption is predicted by combining spatial features and time features, and the accuracy and interpretability of prediction are improved, so as to provide a beneficial reference for industry electricity consumption management and planning.

Description

Technical Field

[0001] This invention relates to the field of smart grids, and mainly to a method for predicting industry electricity consumption based on a complex network of industry relationships. Background Technology

[0002] Industry electricity consumption forecasting is an important research direction in the field of modern power system research, and it is of great significance for power system planning, electricity market trading, and industry policy formulation. Existing industry electricity consumption forecasting methods are mainly divided into two categories: methods based on the time-series development law of electricity data, mainly including regression analysis, gray prediction, exponential smoothing method and differential autoregressive moving average method, such as CN114444397A "A Short-term Electricity Consumption Forecasting Algorithm Based on User Profile Data Model"; and forecasting methods that combine external factors with historical data, mainly including fuzzy prediction, LASSO model, support vector machine, and artificial neural network, such as CN115619025A "Electric Load Triangular Fuzzy Forecasting Method Based on MEEMD and Optimal Combination Integration"; furthermore, patent CN114331101A "Industry Electricity Consumption Forecasting Method" discloses "an industry electricity consumption forecasting method, which uses correlation coefficients to screen industry electricity consumption factors, retaining only factors with high correlation, and on this basis, uses a clustering model to cluster the screened data, thereby further compressing the data volume." Finally, the invention obtains industry electricity consumption forecast results through a prediction model. However, the steps for screening and obtaining relevant variables are relatively simple and cannot accurately characterize the relationship between relevant factors and electricity consumption data. CN116777049A, "A User Electricity Consumption Forecasting Method and System," discloses "a user electricity consumption forecasting method and system, involving the field of electricity consumption forecasting. It takes the influencing factors of electricity consumption as characteristics, obtains historical electricity consumption under different characteristic values, takes the electricity consumption and characteristic values ​​of a time period as a sample, and obtains time series sample datasets of multiple scales by adjusting the duration of the time period; constructs a multi-scale electricity consumption forecasting model, trains the model based on the time series sample datasets of multiple scales; and uses the trained multi-scale electricity consumption forecasting model to predict user electricity consumption in the next time period." This invention is based on Long Short-Term Memory Network (LSTM) and Hybrid Density Network (MDN), considers the uncertainty of electricity consumption and the diversity of influencing factors, and constructs a multi-scale electricity consumption forecasting model. However, this invention only considers the temporal characteristics and ignores the influence of the correlation of spatial characteristics on the forecasting model, which may lead to inconsistent electricity consumption forecasting results for different locations and limited accuracy. Summary of the Invention

[0003] To address the aforementioned problems in existing technologies, this application provides a method for predicting industry electricity consumption based on complex industry relationship networks.

[0004] The technical solution of this application is as follows:

[0005] A method for predicting industry electricity consumption based on a complex network of industry relationships, the method comprising:

[0006] Step 1: Obtain industry electricity consumption data and perform data cleaning to obtain a cleaned dataset;

[0007] Step 2: Using the dataset, construct a distance matrix based on the correlation coefficients between various industries. Based on the distance matrix, use the Prim algorithm to obtain the minimum spanning tree (MST). Based on the MST, construct the corresponding algorithm to generate the planar maximum filter graph (PMFG).

[0008] Step 3: Based on the PMFG, use the Louvain algorithm to classify the industries into different communities to obtain the community classification results;

[0009] Step 4: Determine the target industry, select relevant industries based on the community classification results, construct a multi-graph convolutional and long short-term memory model MGC-LSTM, and use the MGC-LSTM model to predict the industry's electricity consumption.

[0010] Preferably, the data cleaning specifically involves using interpolation methods to process outliers and missing values ​​in the industry's electricity consumption data.

[0011] Preferably, the Pearson correlation coefficients between various industries are calculated using the dataset. A distance matrix is ​​obtained by measuring the distance using the Pearson correlation coefficient, wherein the Pearson correlation coefficient measures the distance. Expressed as a formula:

[0012] .

[0013] Preferably, the minimum spanning tree (MST) obtained using Prim's algorithm based on the distance matrix is ​​as follows:

[0014] Initialize the node set Vnew and the edge set Enew. Take any node as the starting node x and add the starting node x to Vnew. Enew is an empty set.

[0015] Iterate through the rows or columns of the distance matrix, find the node u that is closest to node v, and pair the nodes with the smallest distance.<u, v> The set of edges Enew added to the minimum spanning tree, where u∈Vnew, v Vnew: Add vertex v to node set Vnew. Repeat this step until node set Vnew contains all nodes.

[0016] Construct the structure of the minimum spanning tree using Vnew and Enew.

[0017] Preferably, the planar maximum filter graph (PMFG) generated according to the corresponding algorithm based on MST is as follows:

[0018] The weights of all edges in the minimum spanning tree (MST) edge set are arranged in ascending order to form a weight vector W1;

[0019] The vector W2 is obtained by calculating the minimum value of each row in the distance matrix;

[0020] Merge W1 and W2 to generate a new vector A. The vector A is compared with the weights in the edge set to determine the edges corresponding to the vector A. Connect the corresponding edges to generate a planar graph.

[0021] Arrange the distance matrix in ascending or descending order to obtain the distance sorting vector B;

[0022] Determine whether an element in sorted vector B belongs to vector A. If not, add the edge containing that element to the planar graph. If the planar graph remains planar after adding the edge containing that element, then add the edge to the planar graph. Otherwise, remove it. Repeat this step until the number of edges in the planar graph is greater than or equal to 3M-6, where M is the number of industries.

[0023] Obtain the final planar maximum filtered image (PMFG).

[0024] Preferably, based on the PMFG, the Louvain algorithm is used to classify industries into different communities for community classification, and the community classification results are as follows:

[0025] Each node in the planar maximum filter graph PMFG is considered as an original community. The modularity index increment ΔQ of each original community is calculated when each node is added to different neighbor communities. The node is then selected to join the original community with the largest increment ΔQ. The modularity index increment ΔQ is calculated as follows:

[0026] ; , ;

[0027] In the formula, This is the sum of the total degrees of all nodes within a neighboring community after a node has been added to different neighboring communities; This represents the number of edges connected to the current node. is the internal weight of the edge connected to the current node; m is the total connection weight of the planar maximum filtered graph PMFG; The connection weights from node i to node j;

[0028] The original community after the node is added is taken as the new node, and the process of adding new nodes to the community is iterated until the modular exponential increment ΔQ is less than the preset target value. The iteration ends and the community classification result is obtained.

[0029] Preferably, the target industry is determined, and relevant industries are selected based on the community classification results, specifically:

[0030] If the number of industries in the community to which the target industry belongs is less than the preset value, then select the top four industries with the highest correlation coefficient with the target industry in the community classification results;

[0031] If the number of industries in the community to which the target industry belongs is greater than a preset value, then the industries adjacent to the target industry in the community classification results will be used as the initial industry list, and the top four industries with the highest correlation coefficient with the initial industry list will be selected.

[0032] Preferably, the method further includes obtaining the geographical locations of the target industry and related industries, as well as the historical electricity consumption data of the industries, and constructing an adjacency matrix based on the geographical distances between the cities. An adjacency matrix is ​​constructed based on the correlation coefficients of historical electricity consumption data in the industry. ;

[0033] Adjacency matrix and adjacency matrix Merging to obtain the adjacency matrix Expressed as a formula:

[0034] ;

[0035] ;

[0036] ;

[0037] In the formula, p and q are learnable parameters. The distance between the two cities. The correlation coefficient of historical electricity consumption data of the same industry in the two cities;

[0038] Preferably, a multi-graph convolutional and long short-term memory model MGC-LSTM is constructed, and the prediction of industry electricity consumption using the MGC-LSTM model is specifically as follows:

[0039] Using adjacency matrix The two-layer graph convolutional network GCN extracts spatial features from historical electricity consumption data of the industry. The data extracted by GCN is passed to the long short-term memory model LSTM to extract the temporal features of historical electricity consumption data of the industry.

[0040] By using the hidden state vector of the LSTM at the last moment as the input of the linear layer, the electricity consumption prediction value of the target industry in the corresponding city can be obtained.

[0041] Preferably, the method further includes using mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) as metrics to measure the performance of the multi-graph convolutional and long short-term memory model MGC-LSTM, wherein:

[0042] ;

[0043] ;

[0044] ;

[0045] In the formula, N represents the number of historical electricity consumption data points for the industry. This represents the industry's historical electricity consumption figures. This represents the industry's electricity consumption forecast.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] 1) This invention provides a method for predicting industry electricity consumption based on a complex network of industry relationships. It uses an industry electricity consumption dataset to calculate the Pearson correlation coefficient between various industries, and obtains a distance matrix by measuring the distance through the correlation coefficient, which more accurately characterizes the similarity and correlation between industries.

[0048] 2) This invention provides a method for predicting industry electricity consumption based on complex industry relationship networks. Based on the distance matrix, the Prim algorithm is used to obtain the minimum spanning tree (MST), and then a planar maximum filter graph (PMFG) is constructed to more accurately extract the core structure and key connections in the industry network, thereby better understanding the dependencies between industries.

[0049] 3) This invention provides a method for predicting industry electricity consumption based on a complex network of industry relationships. It uses the Louvain algorithm to classify industries into different communities, grouping similar industries together and placing industries with high correlation in the same community to enhance the similarity and connection between industries.

[0050] 4) This invention provides a method for predicting industry electricity consumption based on complex industry relationship networks. It constructs a multi-graph convolutional and long short-term memory model MGC-LSTM, which can combine spatial and temporal features to predict electricity consumption, improve the accuracy and interpretability of the prediction, and thus provide a useful reference for industry electricity consumption management and planning. Attached Figure Description

[0051] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0052] Figure 2 This is the minimum spanning tree of electricity consumption in 31 manufacturing industries in a certain city from 2020 to 2023, according to an embodiment of the present invention.

[0053] Figure 3 This is a planar maximum filtering map of electricity consumption in 31 manufacturing industries in a certain city from 2020 to 2023, according to an embodiment of the present invention.

[0054] Figure 4 This is a community classification map of electricity consumption in 31 manufacturing industries in a certain city from 2020 to 2023, according to an embodiment of the present invention.

[0055] Figure 5 This is a prediction result diagram of a certain city from 2020 to 2023, according to an embodiment of the present invention. The left side of the dashed line represents the training data and prediction results, and the right side represents the test data and prediction results. Detailed Implementation

[0056] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0057] The present invention provides the following technical solution: an industry electricity consumption prediction method based on a complex network of industry relationships.

[0058] Example 1

[0059] This embodiment provides a method for predicting industry electricity consumption based on a complex network of industry relationships. The specific steps include:

[0060] S1. Obtain industry electricity consumption data and perform data cleaning to obtain a cleaned dataset;

[0061] Obtain relevant information such as energy statistics or electricity consumption data from energy-related institutions, government departments, and large enterprises, and extract industry electricity consumption data from the relevant information;

[0062] After obtaining industry electricity consumption data, interpolation methods are used to process outliers and missing values ​​in the industry electricity consumption data. Based on the existing data points, possible values ​​at the missing or outlier locations are inferred, thereby filling in the missing values ​​or correcting the outliers.

[0063] Furthermore, the industry electricity consumption data is first preprocessed, including removing duplicate data, checking whether the data type meets the requirements, and checking the validity of the data. Then, interpolation methods are used to remove missing values ​​and outliers. Before interpolation, the industry electricity consumption data is subjected to stationarity tests and residual sequence tests, and then appropriate interpolation methods and parameters are selected. The interpolation methods described in this embodiment include linear interpolation, spline interpolation, and Kriging interpolation.

[0064] In this embodiment, the manufacturing industry of a certain city is selected as the target industry, and the electricity consumption data of the manufacturing industry in that city from 2020 to 2023 is obtained as the raw data.

[0065] S2. Using the dataset, construct a distance matrix based on the correlation coefficients between various industries, and use the Prim algorithm to obtain the minimum spanning tree (MST) based on the distance matrix. Then, construct the corresponding algorithm based on the MST to generate the plane maximum filter graph (PMFG).

[0066] S21. Using the dataset, calculate the Pearson correlation coefficient between the various industries. A distance matrix is ​​obtained by measuring distance using the Pearson correlation coefficient. The correlation coefficient reflects the degree of linear relationship between two industries, and the similarity between industries can be measured based on the magnitude of the correlation coefficient. The Pearson correlation coefficient measures distance. Expressed as a formula:

[0067] .

[0068] S22. Based on the distance matrix, use Prim's algorithm to obtain the minimum spanning tree (MST):

[0069] A minimum spanning tree is a spanning tree of a connected undirected graph that contains all vertices of the graph but only enough edges to connect them, and has the minimum total weight. Figure 2 As shown, each vertex represents an industry, and the weight of the edge can be represented by the correlation coefficient. Prim's algorithm is a greedy algorithm that starts from a vertex and gradually expands the spanning tree, always selecting the edge with the minimum weight connecting the vertices of the spanning tree and non-spanning tree, until the spanning tree contains all vertices. Through Prim's algorithm, a connected spanning tree with the minimum total weight is obtained, which reflects the degree of correlation between industries.

[0070] Furthermore, initialize the node set Vnew and the edge set Enew, take any node as the starting node x, add the starting node x to Vnew, and Enew is an empty set;

[0071] Iterate through the rows or columns of the distance matrix, find the node u that is closest to node v, and pair the nodes with the smallest distance.<u, v> The set of edges Enew added to the minimum spanning tree, where u∈Vnew, v Vnew: Add vertex v to node set Vnew. Repeat this step until node set Vnew contains all nodes.

[0072] Construct the structure of the minimum spanning tree using Vnew and Enew;

[0073] S23. Generate the planar maximum filter graph PMFG based on the corresponding algorithm according to MST:

[0074] PMFG is a method for transforming a spanning tree into a planar graph, requiring that the graph formed after adding new edges is planar and that the number of edges varies. In PMFG, each node is connected to at least two other nodes, thus preserving the effective information of the distance network graph to the maximum extent. Because PMFG considers all possible connectivity information in the region surrounding a node and restricts the result to a planar shape, its solution is more closely related to and integrated with the original dataset, reflecting the true degree of connection between data points better than MST. Figure 3 As shown, each node in the graph corresponds to an industry, and edges represent the relationships between industries. The PMFG construction algorithm performs a series of transformations on the spanning tree until it satisfies the properties of a planar graph, i.e., there are no intersecting edges, thus obtaining a planar graph. Therefore:

[0075] The weights of all edges in the minimum spanning tree (MST) edge set are arranged in ascending order to form a weight vector W1;

[0076] The vector W2 is obtained by calculating the minimum value of each row in the distance matrix;

[0077] Merge W1 and W2 to generate a new vector A. The vector A is compared with the weights in the edge set to determine the edges corresponding to the vector A. Connect the corresponding edges to generate a planar graph.

[0078] Arrange the distance matrix in ascending or descending order to obtain the distance sorting vector B;

[0079] Determine whether an element in sorted vector B belongs to vector A. If not, add the edge containing that element to the planar graph. If the planar graph remains planar after adding the edge containing that element, then add the edge to the planar graph; otherwise, remove it. Repeat this step until the number of edges in the planar graph is greater than or equal to 3M-6, where M is the number of industries.

[0080] Obtain the final planar maximum filter map PMFG;

[0081] S3. Based on the PMFG, the Louvain algorithm is used to classify the industry into different communities to obtain the community classification results.

[0082] Louvain's algorithm is a fast, iterative algorithm for detecting community structure. It divides nodes in a graph into different communities by optimizing modularity, making connections between nodes within the same community denser and connections between nodes in different communities sparser. For example... Figure 4 As shown, each industry corresponds to a node. The Louvain algorithm divides the industry nodes into different communities, thus obtaining a complex network that includes the correlations between different communities and different industries.

[0083] Each node in the planar maximum filter graph PMFG is considered as an original community. The modularity index increment ΔQ of each original community is calculated when each node is added to different neighbor communities. The node is then selected to join the original community with the largest increment ΔQ. The modularity index increment ΔQ is calculated as follows:

[0084] ; , ;

[0085] In the formula, This is the sum of the total degrees of all nodes within a neighboring community after a node has been added to different neighboring communities; This represents the number of edges connected to the current node. is the internal weight of the edge connected to the current node; m is the total connection weight of the planar maximum filtered graph PMFG; The connection weights from node i to node j;

[0086] The original community after the node is added is taken as the new node. The process of adding new nodes to the community is iterated until the modular exponential increment ΔQ is less than the preset target value. The iteration ends and the community classification result is obtained.

[0087] S4. Determine the target industry, select relevant industries based on the community classification results, construct a multi-graph convolutional and long short-term memory model MGC-LSTM, and use the MGC-LSTM model to predict the industry's electricity consumption.

[0088] S41. If the number of industries in the community to which the target industry belongs is less than the preset value, then select the top four industries with the highest correlation coefficient with the target industry in the community classification results.

[0089] If the number of industries in the community to which the target industry belongs is greater than the preset value, then the industries adjacent to the target industry in the community classification results will be used as the initial industry list, and the top four industries with the highest correlation coefficient with the initial industry list will be selected.

[0090] S42. Construct an adjacency matrix based on the geographical distances between cities. An adjacency matrix is ​​constructed based on the correlation coefficients of historical electricity consumption data in the industry. The distance matrix reflects the spatial relationship between cities, while the correlation coefficient matrix reflects the interrelationship between industries.

[0091] Adjacency matrix and adjacency matrix Merging to obtain the adjacency matrix This adjacency matrix reflects the degree of correlation between cities and industries, expressed by the formula:

[0092] ;

[0093] ;

[0094] ;

[0095] In the formula, p and q are learnable parameters. The distance between the two cities. The correlation coefficient of historical electricity consumption data of the same industry in the two cities;

[0096] S43. Using an adjacency matrix The two-layer graph convolutional network GCN extracts spatial features from historical electricity consumption data in the industry. GCN aggregates and updates the information of adjacent nodes through iterative message passing and aggregation operations to obtain richer feature representations. The data extracted by GCN is passed to the Long Short-Term Memory (LSTM) model to extract the temporal features of historical electricity consumption data in the industry. LSTM has memory units and gating mechanisms, which can effectively capture long-term dependencies in time series data.

[0097] Using the hidden state vector of the LSTM at the last moment as the input of the linear layer, the electricity consumption prediction value of the target industry in the corresponding city is obtained.

[0098] The mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are used as metrics to measure the performance of the multi-graph convolutional and long short-term memory model MGC-LSTM.

[0099] ;

[0100] ;

[0101] ;

[0102] In the formula, N represents the number of historical electricity consumption data points for the industry. This represents the industry's historical electricity consumption figures. This represents the industry's electricity consumption forecast.

[0103] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting industry electricity consumption based on a complex network of industry relationships, characterized in that, The method includes: Step 1: Obtain industry electricity consumption data and perform data cleaning to obtain a cleaned dataset; Step 2: Using the dataset, construct a distance matrix based on the correlation coefficients between various industries. Based on the distance matrix, use the Prim algorithm to obtain the minimum spanning tree (MST). Based on the MST, construct the corresponding algorithm to generate the planar maximum filter graph (PMFG). Step 3: Based on the PMFG, use the Louvain algorithm to classify the industries into different communities to obtain the community classification results; Step 4: Determine the target industry. Select relevant industries based on the community classification results, specifically: If the number of industries in the community to which the target industry belongs is less than the preset value, then select the top four industries with the highest correlation coefficient with the target industry in the community classification results; If the number of industries in the community to which the target industry belongs is greater than the preset value, then the industries adjacent to the target industry in the community classification results will be used as the initial industry list, and the top four industries with the highest correlation coefficient with the initial industry list will be selected. Obtain the geographical locations of cities and historical electricity consumption data for the target industry and related industries, and construct an adjacency matrix based on the geographical distances between the cities. An adjacency matrix is ​​constructed based on the correlation coefficients of historical electricity consumption data in the industry. ; Adjacency matrix and adjacency matrix Merging to obtain the adjacency matrix Expressed as a formula: ； ； ； In the formula, p and q are learnable parameters. The distance between the two cities. The correlation coefficient of historical electricity consumption data of the same industry in the two cities; A multi-graph convolutional and long short-term memory model, MGC-LSTM, is constructed. This MGC-LSTM model is then used to predict industry electricity consumption. Specifically: Using adjacency matrix The two-layer graph convolutional network GCN extracts spatial features from historical electricity consumption data of the industry. The data extracted by GCN is passed to the long short-term memory model LSTM to extract the temporal features of historical electricity consumption data of the industry. By using the hidden state vector of the LSTM at the last moment as the input of the linear layer, the electricity consumption prediction value of the target industry in the corresponding city can be obtained.

2. The method for predicting industry electricity consumption based on a complex network of industry relationships according to claim 1, characterized in that, The data cleaning specifically involves using interpolation methods to process outliers and missing values ​​in industry electricity consumption data.

3. The method for predicting industry electricity consumption based on a complex network of industry relationships according to claim 1, characterized in that, Using the dataset, calculate the Pearson correlation coefficients between various industries. A distance matrix is ​​obtained by measuring the distance using the Pearson correlation coefficient, wherein the Pearson correlation coefficient measures the distance. Expressed as a formula: 。 4. The method for predicting industry electricity consumption based on a complex network of industry relationships according to claim 3, characterized in that, The minimum spanning tree (MST) is obtained using Prim's algorithm based on the distance matrix as follows: Initialize the node set Vnew and the edge set Enew. Take any node as the starting node x and add the starting node x to Vnew. Enew is an empty set. Iterate through the rows or columns of the distance matrix, find the node u that is closest to node v, and pair the nodes with the smallest distance.<u, v> The set of edges Enew added to the minimum spanning tree, where u∈Vnew, v Vnew: Add vertex v to node set Vnew. Repeat this step until node set Vnew contains all nodes. Construct the structure of the minimum spanning tree using Vnew and Enew.

5. The method for predicting industry electricity consumption based on a complex network of industry relationships according to claim 4, characterized in that, The specific method for generating the planar maximum filter graph (PMFG) based on the corresponding algorithm constructed according to MST is as follows: The weights of all edges in the minimum spanning tree (MST) edge set are arranged in ascending order to form a weight vector W1; The vector W2 is obtained by calculating the minimum value of each row in the distance matrix; Merge W1 and W2 to generate a new vector A. The vector A is compared with the weights in the edge set to determine the edges corresponding to the vector A. Connect the corresponding edges to generate a planar graph. Arrange the distance matrix in ascending or descending order to obtain the distance sorting vector B; Determine whether an element in sorted vector B belongs to vector A. If not, add the edge containing that element to the planar graph. If the planar graph remains planar after adding the edge containing that element, then add the edge to the planar graph. Otherwise, remove it. Repeat this step until the number of edges in the planar graph is greater than or equal to 3M-6, where M is the number of industries. Obtain the final planar maximum filtered image (PMFG).

6. The method for predicting industry electricity consumption based on a complex network of industry relationships according to claim 5, characterized in that, Based on the aforementioned PMFG, the Louvain algorithm is used to classify industries into different communities for community classification. The specific community classification results are as follows: Each node in the planar maximum filter graph PMFG is considered as an original community. The modularity index increment ΔQ of each original community is calculated when each node is added to different neighbor communities. The node is then selected to join the original community with the largest increment ΔQ. The modularity index increment ΔQ is calculated as follows: ； ， ； In the formula, This is the sum of the total degrees of all nodes within a neighboring community after a node has been added to different neighboring communities; This represents the number of edges connected to the current node. is the internal weight of the edge connected to the current node; m is the total connection weight of the planar maximum filtered graph PMFG; The connection weights from node i to node j; The original community after the node is added is taken as the new node, and the process of adding new nodes to the community is iterated until the modular exponential increment ΔQ is less than the preset target value. The iteration ends and the community classification result is obtained.

7. The method for predicting industry electricity consumption based on a complex industry relationship network according to claim 6, characterized in that, The method also includes using mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) as metrics to measure the performance of the multi-graph convolutional and long short-term memory model MGC-LSTM, wherein: ； ； ； In the formula, N represents the number of historical electricity consumption data points for the industry. This represents the industry's historical electricity consumption figures. This represents the industry's electricity consumption forecast.

Images