Industrial user electricity consumption behavior prediction method and system based on heterogeneous network analysis

Abstract

In the industrial user electricity consumption behavior prediction method and system based on heterogeneous network analysis and the storage medium provided in the application, user characteristics of an industrial user are acquired, and a multi-source heterogeneous data table is constructed according to the user characteristics, wherein the user characteristics include basic characteristics, behavior characteristics and environmental characteristics. The multi-source heterogeneous data table is subjected to disambiguation matching and data fusion to obtain structured data. Then, explicit relationship associations between the basic characteristics, the behavior characteristics and the environmental characteristics in the structured data are mined. Then, based on the explicit relationship associations of the structured data, a graph network structure of the structured data is obtained by using a graph convolutional neural network. Then, the electricity consumption of the industrial user is predicted by using a graph attention mechanism according to the graph network structure. Thus, in the embodiments of the application, the accuracy of the prediction result is improved, and the power grid can effectively provide targeted power resources to ensure production efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to an industrial user electricity consumption behavior prediction method and system based on heterogeneous network analysis and a storage medium. BACKGROUND

[0002] Industrial electricity consumption involves many fields and large scale, and has a large proportion in social electricity consumption and is closely related to economic development. Therefore, industrial electricity consumption behavior needs to be predicted to prepare for social electricity consumption in advance and ensure safe and economic operation of the power grid.

[0003] However, in the related art, the statistical and autoregressive methods, Boosting and Ensemble algorithms used in the time series of industrial electricity consumption prediction need to artificially synthesize a large number of features in the time period, so the features may have noise caused by subjective judgment. At the same time, in the related art, the correlation features between a large number of users and behaviors are not effectively utilized, which reduces the accuracy of the prediction result. SUMMARY

[0004] The present application provides an industrial user electricity consumption behavior prediction method and system based on heterogeneous network analysis and a storage medium to at least solve the technical problem of low prediction result accuracy in the related art.

[0005] An industrial user electricity consumption behavior prediction method based on heterogeneous network analysis is provided in a first aspect of the present application, comprising:

[0006] S1, obtaining user features of an industrial user, and constructing a multi-source heterogeneous data table according to the user features, wherein the user features include basic features, behavior features and environmental features;

[0007] S2, performing disambiguation matching and data fusion on the multi-source heterogeneous data table to obtain structured data;

[0008] S3, mining explicit relationship correlations between the basic features, behavior features and environmental features in the structured data;

[0009] S4, obtaining a graph network structure of the structured data by using a graph convolutional neural network based on the explicit relationship correlations of the structured data;

[0010] S5, predicting industrial user electricity consumption by using a graph attention mechanism according to the graph network structure.

[0011] An industrial user electricity consumption behavior prediction system based on heterogeneous network analysis is provided in a second aspect of the present application, comprising:

[0012] The acquisition module is configured to acquire user features of the industrial user, and construct a multi-source heterogeneous data table according to the user features, wherein the user features include basic features, behavior features and environment features;

[0013] The first processing module is configured to perform disambiguation matching and data fusion on the multi-source heterogeneous data table to obtain structured data.

[0014] The mining module is configured to mine explicit relationship associations between the basic features, the behavior features and the environment features in the structured data.

[0015] The second processing module is configured to obtain a graph network structure of the structured data by using a graph convolutional neural network based on the explicit relationship associations of the structured data.

[0016] The prediction module is configured to predict the power consumption of the industrial user by using an attention mechanism according to the graph network structure.

[0017] The computer device provided in the third aspect of the present application comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, and the processor can implement the method in the first aspect of the present application when executing the program.

[0018] The non-transitory computer readable storage medium provided in the fourth aspect of the present application stores a computer program, and the computer program is executed by a processor to implement the method in the first aspect of the present application.

[0019] The technical scheme provided by the embodiments of the present application at least brings the following beneficial effects:

[0020] In the industrial user power consumption behavior prediction method, system and storage medium based on heterogeneous network analysis provided in the present application, user features of an industrial user are acquired, and a multi-source heterogeneous data table is constructed according to the user features, wherein the user features include basic features, behavior features and environment features. Disambiguation matching and data fusion are performed on the multi-source heterogeneous data table to obtain structured data. Explicit relationship associations between the basic features, the behavior features and the environment features in the structured data are then mined. A graph network structure of the structured data is obtained by using a graph convolutional neural network based on the explicit relationship associations of the structured data. The power consumption of the industrial user is predicted by using a graph attention mechanism according to the graph network structure. Thus, in the embodiments of the present application, explicit association relationships and graph structure networks in user data can be mined by using a heuristic rule method and a graph convolutional neural network, so that the correlation between user power consumption behaviors can be obtained. The power consumption of the industrial user is iteratively predicted by using a graph attention mechanism, and the predicted power consumption of the industrial user is continuously corrected, thereby improving the accuracy of the prediction result, and effectively providing targeted power resources for the power grid to ensure production efficiency.

[0021] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0022] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0023] Figure 1 This is a flowchart illustrating a method for predicting the electricity consumption behavior of industrial users based on heterogeneous network analysis according to an embodiment of this application.

[0024] Figure 2 This application provides an embodiment of a heuristic rule-based method for mining explicit relationship graphs in structured data.

[0025] Figure 3 This is a schematic diagram of the structure of an industrial user electricity consumption behavior prediction system based on heterogeneous network analysis according to an embodiment of this application. Detailed Implementation

[0026] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0027] The proposed method, system, and storage medium for predicting industrial user electricity consumption behavior based on heterogeneous network analysis in this application acquires user characteristics of industrial users and constructs a multi-source heterogeneous data table based on these characteristics. The user characteristics include basic features, behavioral features, and environmental features. Disambiguation matching and data fusion are performed on the multi-source heterogeneous data table to obtain structured data. Then, explicit relationships between basic features, behavioral features, and environmental features in the structured data are mined. Based on these explicit relationships, a graph convolutional neural network is used to obtain the graph network structure of the structured data. Finally, a graph attention mechanism is used to predict the electricity consumption of industrial users based on the graph network structure. Therefore, in the embodiments of this application, explicit relationships and graph structure networks in user data can be mined using heuristic rule methods and graph convolutional neural networks, thereby obtaining the correlation between user electricity consumption behaviors. The graph attention mechanism iteratively predicts the electricity consumption of industrial users and continuously corrects the predicted electricity consumption, improving the accuracy of the prediction results. This allows the power grid to efficiently perform its intended function, effectively providing targeted power resources and ensuring production efficiency.

[0028] The following describes, with reference to the accompanying drawings, a method and system for predicting the electricity consumption behavior of industrial users based on heterogeneous network analysis, according to embodiments of this application.

[0029] Example 1

[0030] Figure 1 This is a flowchart illustrating a method for predicting industrial user electricity consumption behavior based on heterogeneous network analysis according to an embodiment of this application. Figure 1 As shown, it may include:

[0031] S1. Obtain the user characteristics of industrial users and construct a multi-source heterogeneous data table based on the user characteristics.

[0032] It should be noted that, in the embodiments of this application, the user characteristics of industrial users can be obtained from data sources such as annual reports, corporate websites, corporate business information data platforms, smart meter feedback, internet news, data from China Meteorological Ground Stations, and publicly available financial data.

[0033] In the embodiments of this application, the user characteristics of industrial users may include basic characteristics, behavioral characteristics, and environmental characteristics.

[0034] Specifically, in the embodiments of this application, the basic features are static attribute information, which may include geographical location, scale, industry, and power bureau; behavioral features are industrial user electricity consumption behavior information, which may include historical electricity consumption, multiphase active power rate, reactive power, etc.; environmental features are external environmental information, which may include enterprise-related news, weather, and macroeconomic data on the Internet.

[0035] S2. Perform disambiguation matching and data fusion on multi-source heterogeneous data tables to obtain structured data.

[0036] In the embodiments of this application, the method for disambiguation matching and data fusion of multi-source heterogeneous data tables includes the following steps:

[0037] S21. Calculate the similarity of relationships of the same attributes among different electricity consumption nodes using Formula 1, where Formula 1 is:

[0038]

[0039] Where A1 and A2 are two different electricity consumption nodes, r n Let r′ be the sequence of all attributes that appear in A1. n This represents the complete sequence of attributes that appear in A2. Is node A1 in attribute r n The relational attribute value on r. n and r′ n To ensure the same attributes across different electricity consumption nodes. Let A1 and A2 be the similarity of their corresponding attributes in terms of electricity consumption nodes.

[0040] Furthermore, in the embodiments of this application, different Similarity calculation methods are used for different types of text.

[0041] Specifically, in the embodiments of this application, for ordinary text, similarity based on edit distance and COSINE similarity based on TF-IDF can be calculated respectively; for numeric text, 7 significant digits can be retained, with each significant digit being 1 if equal and 0 if not equal; for date text, it can be normalized to the form YYYY-MM-DD, with unknown parts filled with "X", and then an indicator function is used as the judgment condition.

[0042] S22. Based on the similarity of the same attributes of different electricity consumption nodes, calculate the final similarity of different electricity consumption nodes through a similarity aggregation function.

[0043] Specifically, in the embodiments of this application, the similarity aggregation function is:

[0044]

[0045] In the embodiments of this application, ω i It is attribute r i The weights. Furthermore, in embodiments of this application, the method for determining the weight vector ω of different electricity consumption nodes may include the following steps:

[0046] Step a: Generate a similarity feature vector for different electricity consumption nodes.

[0047] In the embodiments of this application, the similarity feature vectors of different electricity consumption nodes A1 and A2 are as follows:

[0048]

[0049] Step b: Based on the similarity feature vectors of different electricity consumption nodes, use the Sigmoid function to calculate the equivalence probability of different electricity consumption nodes.

[0050] In the embodiments of this application, the probability that A1 and A2 are equivalent is calculated using the Sigmoid function with the similarity feature vector from step a above:

[0051]

[0052] In the embodiments of this application, A1≡A2 represents the equivalence relationship between A1 and A2, and when P(A1≡A2)≥0.5, it indicates that A1 and A2 have a high degree of similarity and are equivalent, otherwise they are not equivalent.

[0053] Furthermore, in the embodiments of this application, based on the equivalence relationship of step b above, the weight vector ω of different electricity consumption nodes is obtained by using the maximum likelihood estimation method of logistic regression.

[0054] Furthermore, by performing disambiguation matching and data fusion on heterogeneous data, the structural relationships between different nodes can be obtained, thus producing structured data.

[0055] S3. Mining explicit relationships between basic features, behavioral features, and environmental features in structured data.

[0056] In the embodiments of this application, heuristic rule methods can be used to mine explicit relationships between basic features, behavioral features, and environmental features in structured data. Specifically, in the embodiments of this application, various ontologies are used as master nodes, and the display relationship between result-ontology-result is obtained through one-step direct association node matching.

[0057] For example, in an embodiment of this application, Figure 2 In one embodiment of this application, a heuristic rule-based approach is used to mine explicit relationship graphs between basic features, behavioral features, and environmental features in structured data.

[0058] refer to Figure 2 It can be concluded that industrial user 2 uses power grid 1 for power supply, and industrial user 2 is associated with building A1, building B1, and building C1.

[0059] S4. Based on the display relationship association of structured data, the graph network structure of the structured data is obtained by using a graph convolutional neural network.

[0060] In the embodiments of this application, the method for obtaining the graph network structure of structured data based on the display relationship association of structured data using a graph convolutional neural network may include the following steps:

[0061] Step S41: Initialize the embedding vector V for each power consumption node. i Where i = 1, 2, 3, ..., N, and N is the number of electricity consumption nodes;

[0062] In the implementation of this application, the embedding vector V of each electricity consumption node... i It is initialized randomly and determined as the model is trained.

[0063] Step S42: Collect the relation nodes for each electricity consumption node to form a candidate relation set C for each electricity consumption node. i ;

[0064] In the embodiments of this application, candidate relation nodes for each electricity consumption node can be collected through the above-mentioned explicit relation association, and all candidate relation nodes for each electricity consumption node can be used to form a candidate relation set for each electricity consumption node.

[0065] For example, in the embodiments of this application, each candidate relation set for electricity consumption nodes can be Where i is the number of each electricity consumption node.

[0066] Step S43: Calculate the similarity between each electricity consumption node and each candidate vector in the candidate set of electricity consumption nodes;

[0067] Specifically, in the embodiments of this application, the similarity between each electricity consumption node and each candidate vector in the candidate set of the electricity consumption node can be calculated using Formula 2, where Formula 2 is:

[0068]

[0069] Step S44: Use the TopK method to select K nodes from the similarity of each electricity consumption node to form an adjacency matrix A.

[0070] In the embodiments of this application, the TopK method is the e calculated in step S43 above. ji K values ​​are selected from high to low to form the adjacency matrix A. Furthermore, in the embodiments of this application, K is the desired sparsity.

[0071] Furthermore, in the embodiments of this application, the adjacency matrix A is a relational representation matrix of the graph network structure, which can be used to represent the obtained graph network structure.

[0072] S5. Based on the graph network structure, use the graph attention mechanism to predict the electricity consumption of industrial users.

[0073] In one embodiment of this application, the method for predicting industrial user electricity consumption using a graph attention mechanism based on a graph network structure may include the following steps:

[0074] S51. Obtain data from the historical sliding time window;

[0075] In the embodiments of this application, the data based on a historical sliding time window of size w in the past time t is as follows:

[0076] X (t) =[S (t-w) ,S (t-w+1) ,…,S (t-1) ]

[0077] S52. Calculate the aggregated representation of each electricity consumption node using the data and adjacency matrix A;

[0078] In the embodiments of this application, the method for calculating the aggregate representation of each electricity consumption node using data and adjacency matrix A may include calculating the aggregate representation of each electricity consumption node using Formula 3, which is:

[0079]

[0080] Among them, Z i It is an aggregate representation of the electricity consumption node i. It is a trainable weight matrix applied to each electricity consumption node to share a linear transformation. It is the input data of node i. α is the set of neighboring nodes of the electricity consumption node i obtained from the proximity matrix A. i,i It is the attention coefficient.

[0081] Furthermore, in the embodiments of this application, the attention coefficient α can be calculated using Formula 4. i,j Formula four is:

[0082]

[0083] in, Represents a series operation. The embedding vector v of the electricity consumption node i i With transformation structural features Connection, specific Furthermore, in the embodiments of this application, LeakyReLU is used as the nonlinear activation calculation parameter for the attention mechanism, and the softmax method is used to normalize the attention mechanism parameter.

[0084] Furthermore, in the embodiments of this application, the aggregated representation for all N electricity consumption nodes is obtained through the above steps:

[0085] S53. Predict the number of users for each electricity consumption node based on the aggregated representation of each electricity consumption node.

[0086] In one embodiment of this application, the method for predicting the number of users for each electricity consumption node based on the aggregated representation of each electricity consumption node includes using Formula 5 to output a predicted electricity consumption vector of all electricity consumption nodes at time t with dimension N. Formula 5 is:

[0087]

[0088] Furthermore, in the embodiments of this application, the output is predicted. With electricity consumption feedback data s (t)The mean squared error between the two sides is used as the minimum loss equation. The minimum loss equation is then input into the stacked fully connected layer to correct the predicted power consumption vector.

[0089]

[0090] The proposed method for predicting industrial user electricity consumption behavior based on heterogeneous network analysis involves acquiring user characteristics of industrial users and constructing a multi-source heterogeneous data table based on these characteristics. These user characteristics include basic features, behavioral features, and environmental features. The multi-source heterogeneous data table undergoes disambiguation matching and data fusion to obtain structured data. Then, explicit relationships between basic features, behavioral features, and environmental features within the structured data are mined. Based on these explicit relationships, a graph convolutional neural network is used to obtain the graph network structure of the structured data. Finally, a graph attention mechanism is used to predict the electricity consumption of industrial users based on this graph network structure. Therefore, in the embodiments of this application, heuristic rule methods and graph convolutional neural networks can be used to mine explicit relationships and graph structure networks in user data, thereby obtaining the correlation between user electricity consumption behaviors. The graph attention mechanism iteratively predicts the electricity consumption of industrial users and continuously corrects the predicted electricity consumption, improving the accuracy of the prediction results. This allows the power grid to efficiently perform its intended function, effectively providing targeted power resources and ensuring production efficiency.

[0091] Example 2

[0092] Furthermore, Figure 3 This is a schematic diagram of the structure of an industrial user electricity consumption behavior prediction system based on heterogeneous network analysis according to an embodiment of this application, as shown below. Figure 3 As shown, it may include:

[0093] The acquisition module 301 is used to acquire user characteristics of industrial users and construct a multi-source heterogeneous data table based on user characteristics. The user characteristics include basic characteristics, behavioral characteristics and environmental characteristics.

[0094] The first processing module 302 is used to perform disambiguation matching and data fusion on multi-source heterogeneous data tables to obtain structured data;

[0095] Mining module 303 is used to mine explicit relationships between basic features, behavioral features, and environmental features in structured data;

[0096] The second processing module 304 is used to obtain the network structure by using a graph convolutional neural network to associate the display relationships based on structured data.

[0097] The prediction module 305 is used to predict the electricity consumption of industrial users based on the network structure and using an attention mechanism.

[0098] In one embodiment of this application, the mining module 303 is used to obtain the display relationship association of result-ontology-result by using various ontologies as main nodes and directly associating nodes in one step.

[0099] To implement the above embodiments, this disclosure also proposes a computer device.

[0100] The computer device provided in this application includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it can achieve the following: Figure 1 The method for predicting industrial user electricity consumption behavior based on heterogeneous network analysis is shown.

[0101] To implement the above embodiments, this disclosure also proposes a non-transitory computer-readable storage medium.

[0102] The non-transitory computer-readable storage medium provided in this disclosure embodiment stores a computer program; when the computer program is executed by a processor, it can achieve the following: Figure 1 The method for predicting industrial user electricity consumption behavior based on heterogeneous network analysis is shown.

[0103] The proposed method, system, and storage medium for predicting industrial user electricity consumption behavior based on heterogeneous network analysis in this application acquires user characteristics of industrial users and constructs a multi-source heterogeneous data table based on these characteristics. The user characteristics include basic features, behavioral features, and environmental features. Disambiguation matching and data fusion are performed on the multi-source heterogeneous data table to obtain structured data. Then, explicit relationships between basic features, behavioral features, and environmental features in the structured data are mined. Based on these explicit relationships, a graph convolutional neural network is used to obtain the graph network structure of the structured data. Finally, a graph attention mechanism is used to predict the electricity consumption of industrial users based on the graph network structure. Therefore, in the embodiments of this application, explicit relationships and graph structure networks in user data can be mined using heuristic rule methods and graph convolutional neural networks, thereby obtaining the correlation between user electricity consumption behaviors. The graph attention mechanism iteratively predicts the electricity consumption of industrial users and continuously corrects the predicted electricity consumption, improving the accuracy of the prediction results. This allows the power grid to efficiently perform its intended function, effectively providing targeted power resources and ensuring production efficiency.

[0104] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0105] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0106] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for predicting the electricity consumption behavior of industrial users based on heterogeneous network analysis, characterized in that, Includes the following steps: S1. Obtain the user characteristics of industrial users and construct a multi-source heterogeneous data table based on the user characteristics, wherein the user characteristics include basic characteristics, behavioral characteristics and environmental characteristics; S2. Perform disambiguation matching and data fusion on the multi-source heterogeneous data table to obtain structured data; S3. Mine the explicit relationships between the basic features, behavioral features, and environmental features in the structured data; S4. Based on the display relationship association of the structured data, obtain the graph network structure of the structured data using a graph convolutional neural network; S5. Based on the graph network structure, use the graph attention mechanism to predict the electricity consumption of industrial users; The disambiguation matching and data fusion of heterogeneous data include: Formula 1 is used to calculate the similarity of relationships of the same attributes among different electricity consumption nodes, where Formula 1 is: Where A1 and A2 are two different electricity consumption nodes, r n For all attribute sequences appearing in A1, r n ′ represents the complete sequence of attributes appearing in A2. Is node A1 in attribute r n The relational attribute value, r n and r n ' represents the same attribute across different electricity consumption nodes. The similarity of the relationship between the electricity consumption nodes A1 and A2 on the corresponding attributes; Based on the similarity of the same attributes of different electricity consumption nodes, the final similarity of different electricity consumption nodes is calculated by a similarity aggregation function. The method of mining explicit relationships in the structured data using heuristic rules includes using various ontologies as master nodes and obtaining explicit relationship associations of result-ontology-result through one-step direct association node matching; The process of establishing display relationships based on the structured data and obtaining the graph network structure of the structured data using a graph convolutional neural network includes the following steps: Initialize the embedding vector V for each electricity consumption node i Where i = 1, 2, 3, ..., N, and N is the number of electricity consumption nodes; Collect the relation nodes of each electricity consumption node to form a candidate relation set C for each electricity consumption node. i ; Calculate the similarity between each electricity consumption node and each candidate vector in the candidate set of the electricity consumption node; The TopK method is used to select K nodes from the similarity of each electricity consumption node to form an adjacency matrix A.

2. The method as described in claim 1, characterized in that, The step of predicting industrial user electricity consumption using a graph attention mechanism based on the graph network structure includes the following steps: Retrieve data from historical sliding time windows; The aggregated representation of each electricity consumption node is calculated using the data and the adjacency matrix A; The number of users for each power consumption node is predicted based on the aggregated representation of each power consumption node.

3. An industrial user electricity consumption behavior prediction system based on heterogeneous network analysis, characterized in that, Includes the following modules: The acquisition module is used to acquire user characteristics of industrial users and construct a multi-source heterogeneous data table based on the user characteristics, wherein the user characteristics include basic characteristics, behavioral characteristics and environmental characteristics; The first processing module is used to perform disambiguation matching and data fusion on the multi-source heterogeneous data table to obtain structured data; The mining module is used to mine the explicit relationships and associations between the basic features, behavioral features, and environmental features in the structured data; The second processing module is used to obtain a graph network structure by using a graph convolutional neural network based on the display relationship association of the structured data. The prediction module is used to predict the electricity consumption of industrial users based on the graph network structure and using an attention mechanism. The disambiguation matching and data fusion of heterogeneous data include: Formula 1 is used to calculate the similarity of relationships of the same attributes among different electricity consumption nodes, where Formula 1 is: Where A1 and A2 are two different electricity consumption nodes, r n For all attribute sequences appearing in A1, r n ′ represents the complete sequence of attributes appearing in A2. Is node A1 in attribute r n The relational attribute value, r n and r n ' represents the same attribute across different electricity consumption nodes. Let r1 and A2 be the similarity of their corresponding attributes in terms of electricity consumption nodes; Based on the similarity of the same attributes of different electricity consumption nodes, the final similarity of different electricity consumption nodes is calculated by a similarity aggregation function. The method of mining explicit relationships in the structured data using heuristic rules includes using various ontologies as master nodes and obtaining explicit relationship associations of result-ontology-result through one-step direct association node matching; The process of establishing display relationships based on the structured data and obtaining the graph network structure of the structured data using a graph convolutional neural network includes the following steps: Initialize the embedding vector V for each electricity consumption node i Where i = 1, 2, 3, ..., N, and N is the number of electricity consumption nodes; Collect the relation nodes of each electricity consumption node to form a candidate relation set C for each electricity consumption node. i ; Calculate the similarity between each electricity consumption node and each candidate vector in the candidate set of the electricity consumption node; The TopK method is used to select K nodes from the similarity of each electricity consumption node to form an adjacency matrix A.

4. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the method as described in any one of claims 1-2.

5. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-2.

Images