Power consumer group portraying method and system based on depth map neural network

Abstract

The invention discloses a power consumer group portraying method and system based on a depth map neural network, and relates to the technical field of power consumer portraying, and the method comprises the steps: collecting first data, and carrying out the preprocessing of the first data; constructing a first relation graph based on the second data, and performing feature processing on the first relation graph through a first model; and generating a second group portrait through a first algorithm based on the first group feature, dynamically adjusting the first relation graph based on the first data, and adaptively adjusting the second group portrait through a first machine learning method. According to the method, a user group portrait is updated in real time through an incremental updating mechanism of a graph structure and a GNN model; by fusing multi-source heterogeneous data, the accuracy and integrity of the group portrait are improved; modeling is performed on a complex relationship between users by using a graph structure, so that the accuracy of a group portrait is improved; the user group portrait is continuously optimized through reinforcement learning, and the effectiveness and accuracy of the group portrait are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power user portraits, and specifically to a method and system for power user group portraits based on a deep graph neural network. Background Art

[0002] With the rapid development of big data and artificial intelligence technologies, the power industry is also gradually moving towards digitalization and intelligence. Power user portraits, as an important means for power enterprises to provide personalized services and improve demand response management capabilities, the traditional power user portrait methods mainly conduct simple analyses based on the power consumption behavior characteristics of users to generate individual user portraits, which are difficult to effectively capture the complex relationships between users, unable to dynamically reflect the changes of user groups, and mostly rely on a single data source, resulting in limitations in the accuracy and practicality of the portraits.

[0003] On the one hand, most of the existing technologies construct user portraits based on static data, unable to timely reflect the changes in user behaviors and group characteristics, resulting in insufficient response to the rapidly changing user demands. On the other hand, the existing technologies cannot effectively capture the complex relationships between users, and only cluster users through static features, making it difficult to reveal complex associations such as social relationships and geographical similarities between users, affecting the accuracy of the portraits. In addition, the existing technologies cannot dynamically adjust according to user behaviors and the accuracy of the portraits, resulting in difficulties in ensuring the effectiveness and accuracy of the portraits.

[0004] In view of the deficiencies of the existing technologies, the present invention proposes a method and system for power user group portraits based on a deep graph neural network, which utilizes the characteristics of the graph neural network, combines multi-source heterogeneous data for user group portrait modeling, and adopts a graph structure incremental update and reinforcement learning feedback mechanism to achieve dynamic update of user portraits. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the technical problems solved by the present invention are: the existing power user portrait methods have problems such as lack of dynamic update ability, insufficient data fusion ability, insufficient complex relationship network modeling ability, and lack of an adaptive optimization mechanism.

[0007] To solve the above technical problems, the present invention provides the following technical solution: A method for power user group portraits based on a deep graph neural network, including collecting first data, preprocessing the first data to obtain second data; constructing a first relationship graph based on the second data, performing feature processing on the first relationship graph through a first model to generate first group features; generating a second group portrait based on the first group features through a first algorithm, dynamically adjusting the first relationship graph based on the first data, and adaptively adjusting the second group portrait through a first machine learning method.

[0008] As a preferred solution of the method for constructing power user group portraits based on a deep graph neural network according to the present invention, wherein: the preprocessing includes cleaning the first data through a first processing method and standardizing the data through a second processing method.

[0009] As a preferred solution of the method for constructing power user group portraits based on a deep graph neural network according to the present invention, wherein: the alignment includes aligning the preprocessed data with different frequencies.

[0010] As a preferred solution of the method for constructing power user group portraits based on a deep graph neural network according to the present invention, wherein: constructing the first relational graph based on the second data includes defining the first relational graph based on the second data and constructing the first relational graph through weight calculation.

[0011] As a preferred solution of the method for constructing power user group portraits based on a deep graph neural network according to the present invention, wherein: performing feature processing on the first relational graph through the first model includes initializing the features of the nodes in the first relational graph through the first model, and aggregating and updating the node features.

[0012] As a preferred solution of the method for constructing power user group portraits based on a deep graph neural network according to the present invention, wherein: generating the second group portrait based on the first group features through the first algorithm includes performing group division on the first group features through the use of the first algorithm and generating the second group portrait.

[0013] As a preferred solution of the method for constructing power user group portraits based on a deep graph neural network according to the present invention, wherein: dynamically adjusting the first relational graph based on the first data includes dynamically adjusting the first relational graph when the first data is newly added or changed.

[0014] Another object of the present invention is to provide a power user group portrait system based on a deep graph neural network, which can solve the technical problems of the current power user portrait method lacking dynamic update ability, insufficient data fusion ability, insufficient complex relational network modeling ability, and lack of an adaptive optimization mechanism through deep graph neural network technology.

[0015] As a preferred solution of the power user group portrait system based on the deep graph neural network according to the present invention, it includes a data processing module, a feature aggregation module, and an adaptive update module; the data processing module is used to collect first data and preprocess the first data to obtain second data; the feature aggregation module is used to construct a first relationship graph based on the second data, perform feature processing on the first relationship graph through a first model, and generate first group features; the adaptive update module is used to generate a second group portrait through a first algorithm based on the first group features, dynamically adjust the first relationship graph based on the first data, and adaptively adjust the second group portrait through a first machine learning method.

[0016] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the power user group portrait method based on the deep graph neural network are implemented.

[0017] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the steps of the power user group portrait method based on the deep graph neural network are implemented.

[0018] The beneficial effects of the present invention: The power user group portrait method based on the deep graph neural network provided by the present invention can update the user group portrait in real time through the incremental update mechanism of the graph structure and the GNN model, timely reflect the changes in user behavior and group characteristics, and improve the response ability of power enterprises to user needs; by integrating multi-source heterogeneous data and using the graph structure input ability of GNN, it describes the group characteristics of users from multiple dimensions, improving the accuracy and integrity of the group portrait; by using the graph structure to model the complex relationships between users, it can effectively capture various user-related features such as geographical location and social relationships, improving the accuracy of the group portrait; by continuously optimizing the user group portrait through reinforcement learning, the portrait can adapt to the changes in user behavior, ensuring the effectiveness and accuracy of the group portrait, thereby providing more accurate demand prediction and personalized service support for power companies. The present invention effectively solves the deficiencies of the prior art from the perspectives of dynamic update ability, multi-dimensional data fusion, complex relationship modeling, and adaptive optimization mechanism, and achieves good results in terms of the accuracy and integrity of the group portrait and the demand prediction of power companies. Description of the Drawings

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 The overall flowchart of a method for constructing power user group portraits based on a deep graph neural network provided in the first embodiment of the present invention.

[0021] Figure 2 The overall flowchart of a system for constructing power user group portraits based on a deep graph neural network provided in the third embodiment of the present invention. Detailed implementation manners

[0022] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following provides a detailed description of the specific implementation manners of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0023] Embodiment 1. Refer to Figure 1 , which is an embodiment of the present invention, providing a method for constructing power user group portraits based on a deep graph neural network, including:

[0024] S1: Collect first data, preprocess the first data to obtain second data.

[0025] Furthermore, the preprocessing includes cleaning the first data through a first processing method and standardizing the data through a second processing method.

[0026] The first data includes, but is not limited to, power user electricity consumption behavior data, social data, and user basic attributes collected from multi-source heterogeneous data.

[0027] The function of the cleaning is to remove outliers.

[0028] The method adopted by the first processing method can be the box plot method, the DBSCAN clustering method, or other suitable data cleaning methods.

[0029] In the embodiments of the present application, the first processing method adopted is the box plot method. The first data collected is removed of outliers through the box plot method. The specific steps include calculating the quartiles Q1, Q3 of the data and the interquartile range IQR = Q3 - Q1, setting the upper and lower limits of the outliers, and regarding the data points outside the upper and lower limits as outliers and removing them.

[0030] In an alternative embodiment, the first processing method uses the DBSCAN clustering method. The first data collected is first normalized to eliminate the dimension difference. After normalization, outliers are removed through the DBSCAN clustering method. The specific steps include setting clustering parameters, defining the neighborhood radius in combination with the data distribution, defining the minimum neighborhood sample data of the core points according to the data density, applying the DBSCAN algorithm to the normalized data, dividing the data points into three categories, namely core points, boundary points, and noise points. The isolated points in the sparse area are marked as -1, and the noise points with the label -1 are extracted from the DBSCAN clustering results and regarded as outliers. The original data records corresponding to the noise points are removed, and the core point and boundary point data are retained to ensure the consistency of the density of the data set.

[0031] The second processing method includes, but is not limited to, the mean-standard deviation normalization formula, min-max normalization, max absolute value normalization, etc.

[0032] In the embodiment of the present application, the second processing method uses the mean-standard deviation normalization formula to normalize the first data.

[0033] It should be noted that alignment includes aligning the preprocessed data with different frequencies.

[0034] In this embodiment, the preprocessed data with different frequencies is aligned. Specifically, it is aligned according to a fixed time window, and the missing data is processed by interpolation or filling to obtain the second data, ensuring the consistency of the data in the time dimension.

[0035] It should also be noted that the power user electricity consumption behavior data is collected in real time through smart meters, and the collected data includes, but is not limited to, daily average electricity consumption, peak-valley difference, seasonal electricity consumption patterns, etc.

[0036] It should also be noted that the social data includes, but is not limited to, being obtained from the customer interaction records of power enterprises and is used to describe the social relationships between users; the basic user attributes include the geographical location of the user and the type of electricity consumption, and the type of electricity consumption includes, but is not limited to, residential electricity consumption, industrial electricity consumption, and commercial electricity consumption.

[0037] S2: Based on the second data, construct a first relationship graph, and perform feature processing on the first relationship graph through a first model to generate first group features.

[0038] Furthermore, constructing the first relationship graph based on the second data includes defining the first relationship graph based on the second data and constructing the first relationship graph through weight calculation.

[0039] The first relationship graph includes, but is not limited to, a graph constructed through a graph neural network with power users as nodes and geographical similarity and electricity consumption behavior similarity as edge weights.

[0040] It should be noted that in the embodiments of the present application, constructing the first relational graph includes node definition, edge definition, and weight calculation.

[0041] The node definition includes that each user node represents a power user, and the features of the node are composed of electricity consumption behavior features and basic attributes.

[0042] The edge definition and weight calculation include geographical similarity calculation and electricity consumption behavior similarity calculation.

[0043] The geographical similarity calculation includes calculating the geographical similarity between user nodes using a Gaussian kernel function based on the geographical distance between users, and the formula is expressed as:

[0044]

[0045] where d(i,j) represents the geographical distance between user i and user j, and σ0 represents the smoothing parameter.

[0046] The electricity consumption behavior similarity includes calculating the electricity consumption behavior similarity between users through cosine similarity, and the formula is expressed as:

[0047]

[0048] The user relationship graph formed by nodes and edges is represented as an adjacency matrix A, where A ij represents the edge weight between node i and node j.

[0049] It should also be noted that the feature processing of the first relational graph by the first model includes initializing the features of the nodes in the first relational graph by the first model, and aggregating and updating the node features.

[0050] The first model is a graph neural network model.

[0051] In the embodiments of the present application, the first model is a graph neural network model based on a graph convolutional network and a graph attention network.

[0052] In the embodiments of the present application, the node feature initialization includes that the initial feature vector of each node is X i , and all node features form a feature matrix X, which is expressed as:

[0053]

[0054] where N represents the number of nodes and d represents the feature dimension.

[0055] In the embodiments of the present application, aggregating and updating the node features includes aggregating and updating the node features using a graph convolutional layer, and the update formula is expressed as:

[0056]

[0057] Among them, represents the adjacency matrix plus self-loops, and I represents the identity matrix. represents the degree matrix of, and W (l) represents the learnable weight matrix of the l-th layer, and σ1 represents the activation function.

[0058] The attention mechanism is used to weighted aggregate the features of neighbor nodes, and the weights are determined by the attention coefficient α ij The weight aggregation formula is expressed as:

[0059]

[0060] The attention coefficient α ij , and the calculation formula is expressed as:

[0061]

[0062] Among them, a represents the learnable parameter vector, and || represents the vector concatenation operation.

[0063] It should also be noted that after passing through the multi-layer graph convolutional network and graph attention network of the first model, the node features have already fused the information of neighbor nodes to form a high-order feature representation. However, to generate the first group feature, the node features still need to be read and aggregated again.

[0064] In the embodiments of the present application, generating the first group feature includes passing through multi-layer feature reading, aggregating neighbor node information, feature aggregation strategy, global pooling layer and feature aggregation, and multi-layer perceptron processing of the group feature. Normalization and regularization are used to re-read and aggregate the high-order features in order to generate the overall feature representation of the entire group.

[0065] For multi-layer feature reading, specifically, after passing through the propagation of multi-layer GCN and GAT, the features of each node at each layer contain different information levels. The feature representation of the last layer contains the information fusion of the node and its neighbors. By aggregating the features of different levels, a more comprehensive node representation is formed. For the node feature H (l) of each layer, it is further processed through a fully connected layer to reduce the feature dimension and improve the calculation efficiency. The features of each layer are concatenated to retain the information of different levels, so that the node features have more dimensional expression capabilities. The multi-layer feature reading is expressed as:

[0066] H final = H (l) || H (2) || … || H (L)

[0067] Among them, H (l)Denote the node features of the l-th layer, and L denote the total number of layers.

[0068] Aggregate the information of neighbor nodes. Specifically, different aggregation methods are adopted to effectively integrate the information of neighbor nodes to generate more representative node features.

[0069] In the embodiment of the present application, the adopted aggregation method is fully connected aggregation. The features of each node and its neighbors are aggregated through a fully connected layer, so as to learn the complex relationship between the node and its neighbors.

[0070] In an alternative embodiment, the adopted aggregation method is average pooling aggregation. By taking the average of the neighbor features of each node, the final node feature representation is generated. Average pooling aggregation can eliminate the influence of noise. Especially for nodes with a large number of neighbors, it can effectively capture the overall features. Average pooling aggregation is expressed as:

[0071]

[0072] where N(i) represents the set of neighbor nodes of node i, and h j represents the feature representation of the neighbor nodes.

[0073] In an alternative embodiment, the adopted aggregation method is max pooling aggregation. By taking the maximum value among the neighbor node features as the feature representation to retain the significant feature information, it has an advantage for nodes with important features.

[0074] Feature aggregation strategy. Specifically, a combination of multiple aggregation strategies is adopted to enhance the robustness of the model; a learnable weight w agg is assigned to each feature aggregation strategy. Through the feature weights, the weighted sum of multiple aggregation results is calculated to obtain the final node features, which is expressed as:

[0075]

[0076] where w agg represents the learnable parameters of different aggregation methods, represents the node features obtained by different aggregation strategies.

[0077] Global pooling layer and feature aggregation. Specifically, after generating the high-order feature representations of each node, the features of all nodes in the first relational graph are aggregated through the global pooling layer to form the feature representation of the group.

[0078] In the embodiment of the present application, the global pooling layer adopts global average pooling. Global average pooling is to take the average of the features of all nodes to generate the overall group features. Global average pooling is expressed as:

[0079]

[0080] Among them, N represents the total number of nodes in the graph, represents the final feature representation of node i.

[0081] In an alternative embodiment, the global pooling layer uses global max pooling. Global max pooling takes the maximum value of the features of all nodes in the graph, retains the most prominent group feature information, and highlights the influence of important nodes in the group features.

[0082] Processing of group features by a multi-layer perceptron. Specifically, the group features obtained through the global pooling layer may contain redundant information. To further streamline and enhance the feature representation, the group features are subjected to non-linear transformation and dimensionality reduction processing using a multi-layer perceptron, including multiple fully connected layers, each followed by an activation function for non-linearly mapping the group features, and finally generating refined group features, denoted as:

[0083] E group = MLP(H group ) MLP

[0084] Normalization and regularization. Specifically, after feature aggregation and generation, batch normalization is used to normalize the features to ensure consistent distribution of each feature dimension, thereby accelerating model convergence. In addition, dropout is used to regularize the features to prevent overfitting and improve the generalization ability of the model.

[0085] It should also be noted that the first group features include but are not limited to basic attribute features, behavioral features, and social features.

[0086] S3: Generate a second group portrait based on the first group features through a first algorithm, dynamically adjust the first relationship graph based on the first data, and adaptively adjust the second group portrait through a first machine learning method.

[0087] Furthermore, generating a second group portrait based on the first group features through a first algorithm includes performing group division on the first group features using the first algorithm and generating a second group portrait.

[0088] The first algorithm can be K-means, or mean shift clustering, or other clustering algorithms suitable for group division.

[0089] In the embodiments of the present application, the clustering algorithm used by the first algorithm is K-means to determine groups of similar users. The calculation formula for the cluster center is denoted as:

[0090]

[0091] Among them, C kDenote all user nodes in the k-th group.

[0092] In an alternative embodiment, the clustering algorithm used by the first algorithm is mean shift clustering. By calculating the mean vector of neighbor nodes within the area around each user node and using a Gaussian kernel function to estimate the density of the area around each user node, each user node is moved to the position where the corresponding mean vector is located for convergence. When the positions of the user nodes converge, nodes with similar positions are determined to belong to the same cluster, and the mean point represents each cluster.

[0093] It should be noted that after the group division by the first algorithm, feature extraction is performed to generate a second group portrait.

[0094] The second group portrait includes, but is not limited to, a basic attribute portrait, a behavior portrait, and a social portrait. Among them, the basic attribute portrait is the basic features generated based on the geographical location, electricity consumption category, etc. of the group, the behavior portrait is the electricity consumption behavior features extracted from the group, and the social portrait is the features extracted based on the interaction frequency and characteristics of the social relationship.

[0095] It should also be noted that dynamically adjusting the first relationship graph based on the first data includes dynamically adjusting the first relationship graph when the first data is newly added or changed.

[0096] In the embodiment of the present application, the first relationship graph is dynamically adjusted. Specifically, for node incremental update, when a new user joins or the features of an existing user change, the neighbor relationship is recalculated; for edge incremental update, when the user relationship changes, the edge weight is updated, and the update formula is expressed as:

[0097] W i,j =(1 - η)W i,j +ηW i, ′ j

[0098] where η represents the update speed, and W i, ′ j represents the recalculated similarity weight.

[0099] It should also be noted that in the embodiment of the present application, the second group portrait is adaptively adjusted through the first machine learning method. Specifically, based on the user's behavior feedback and the accuracy of the portrait, the group portrait is adaptively adjusted through reinforcement learning, and the reward function is defined as:

[0100] R = -|Y true -Y pred |

[0101] where Y true and Y pred respectively represent the actual and predicted electricity consumption behaviors.

[0102] Example 2 is an embodiment of the present invention, which provides a method for profiling power user groups based on a deep graph neural network. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0103] A provincial power company needs to carry out refined service management for about 500,000 power users in its jurisdiction, aiming to optimize the demand response strategy through user group portraits. The traditional method relies on static clustering analysis of single electricity consumption data, which leads to delayed portrait updates and ignores the complex relationships between users. In order to verify the effectiveness of the method of the present invention, 10 typical residential communities in a city were selected as experimental subjects, and the differences in portrait accuracy, dynamic update efficiency and multi-dimensional feature fusion capabilities between the traditional K-means clustering method and the invented method were compared.

[0104] First, data collection: electricity consumption behavior data is collected through smart meters, such as users' average daily electricity consumption, peak-to-valley difference rate, seasonal fluctuation coefficient, etc.; geographic data is obtained through users' residential coordinates and community division information; social data is obtained through the frequency of interaction with the power APP and the relevance of community forums.

[0105] The collected data were preprocessed, and the box plot method was used to eliminate outliers. The numerical data such as electricity consumption and geographical distance were normalized using the mean-standard deviation. The data collection frequency was unified to once an hour, and missing values were filled by linear interpolation.

[0106] Secondly, a user relationship graph is constructed and nodes are defined. Each user is a node, and the feature vector includes normalized electricity consumption, geographic location, and community interaction score. Edge weights are calculated. Geographic similarity is based on the Gaussian kernel function to calculate the distance similarity between users. The similarity of electricity consumption behavior is measured by cosine similarity to measure the matching degree of user electricity consumption patterns. Geographic and electricity consumption similarity are each given a 50% weight to generate an adjacency matrix.

[0107] The graph neural network model is designed by adopting a structure of alternating stacking of 3 layers of GCN and 2 layers of GAT. The GCN layer aggregates the electricity consumption behavior characteristics of neighbor nodes and retains the community-level electricity consumption pattern. The GAT layer dynamically weights the social interaction characteristics through the attention mechanism to highlight highly associated users. The training parameter settings are learning rate 0.001, batch size 256, Dropout rate 0.3, and training cycle 50 times.

[0108] Finally, K-means clustering is used on the high-order features output by GNN to extract five typical user groups for group profiling, and the group profiling is dynamically updated. That is, incremental updates are made to automatically integrate new users into the graph structure every day, and edge weights are adjusted every hour. The feature weights are adjusted according to the difference between the actual power consumption behavior of users and the predicted values to enhance learning.

[0109] Traditional K-means clustering is only based on historical electricity consumption data. The method of the present invention integrates multi-source data and a GNN model for comparative experiments, and obtains the experimental data shown in Table 1 below.

[0110]

[0111]

[0112] It can be seen from the analysis of the experimental data that the portrait accuracy of the traditional method relying only on the single feature of electricity consumption is 69.2% on average, and it is unable to distinguish user groups with similar geographical locations but different social behaviors. For example, industrial and residential users in Community B are misclassified due to similar electricity consumption. The accuracy of the method of the present invention is increased to 89.8%, and the feature dimension is extended to 8 dimensions. For example, high-interaction users (social feature > 0.8) in Community A are independently classified as the "high-demand response group", and the accuracy is increased by 17.3%.

[0113] For the traditional method relying on batch re-clustering, the update delay reaches 24 hours and it cannot adapt to real-time demand changes. For example, new users added during the peak electricity consumption period in Community D can only be included in the portrait the next day, so the portrait accuracy is relatively low. The method of the present invention reduces the delay to 0.5 hours through the incremental update mechanism, and the edge weight is adjusted once per hour, and the portrait accuracy reaches 90.3%.

[0114] The average coverage rate of the social features of the method of the present invention reaches 86%, while the traditional method completely ignores social data. The method of the present invention accurately models the geographical similarity error of users through the Gaussian kernel function, with an average of 0.3 km, which is reduced by 78.6% compared with the traditional method. However, the calculation time of the method of the present invention increases to 2.95 seconds per time, and the calculation efficiency is relatively low compared with the traditional method. However, the method of the present invention still meets the real-time requirements through model lightweight design.

[0115] Example 3, referring to Figure 2 , which is an embodiment of the present invention, provides a power user group portrait system based on a deep graph neural network, including a data processing module, a feature aggregation module, and an adaptive update module.

[0116] The data processing module is used to collect first data, preprocess the first data to obtain second data; the feature aggregation module is used to construct a first relationship graph based on the second data, perform feature processing on the first relationship graph through a first model, and generate first group features; the adaptive update module is used to generate a second group portrait based on the first group features through a first algorithm, dynamically adjust the first relationship graph based on the first data, and adaptively adjust the second group portrait through a first machine learning method.

[0117] If a function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs, etc., which are various media that can store program codes.

[0118] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device.

[0119] More specific examples (a non-exhaustive list) of computer-readable media include the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), optical fiber devices, and portable compact disc read-only memories (CDROMs). Additionally, a computer-readable medium can even be paper or other suitable media on which a program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or, if necessary, other appropriate processing, and then stored in a computer memory.

[0120] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following technologies well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logic functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc. It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

[0121] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for constructing power user group portraits based on deep graph neural networks, characterized in that, It includes: Collect the first data, preprocess the first data, and align it to obtain the second data; Construct a first relationship graph based on the second data, perform feature processing on the first relationship graph through a first model, and generate first group features; Generate a second group portrait through a first algorithm based on the first group features, dynamically adjust the first relationship graph based on the first data, and adaptively adjust the second group portrait through a first machine learning method.

2. The method for constructing power user group portraits based on a depth graph neural network according to claim 1, wherein: The preprocessing includes cleaning the first data through a first processing method and standardizing the data through a second processing method.

3. The method for constructing power user group portraits based on a deep graph neural network according to claim 2, wherein: The alignment includes aligning the preprocessed data with different frequencies.

4. The method for power user group portrait based on a deep graph neural network according to claim 3, characterized in that: The constructing of the first relationship graph based on the second data includes defining the first relationship graph based on the second data and constructing the first relationship graph through weight calculation.

5. The method for constructing power user group portraits based on a depth graph neural network according to claim 4, wherein: The performing of feature processing on the first relationship graph through the first model includes initializing the features of the nodes in the first relationship graph through the first model, aggregating and updating the node features.

6. The method for constructing power user group portraits based on a deep graph neural network according to claim 5, characterized in that: The generating of the second group portrait through the first algorithm based on the first group features includes performing group division on the first group features through the use of the first algorithm and generating the second group portrait.

7. The method for constructing power user group portraits based on a deep graph neural network according to claim 6, wherein: The dynamically adjusting of the first relationship graph based on the first data includes dynamically adjusting the first relationship graph when the first data is newly added or changed.

8. A system adopting the method for constructing power user group portraits based on a depth graph neural network as described in any one of claims 1 to 7, characterized in that: It includes a data processing module, a feature aggregation module, and an adaptive update module; The data processing module is used to collect the first data, preprocess the first data, and obtain the second data; The feature aggregation module is used to construct a first relationship graph based on the second data, perform feature processing on the first relationship graph through a first model, and generate first group features; The adaptive update module is used to generate a second group portrait through a first algorithm based on the first group features, dynamically adjust the first relationship graph based on the first data, and adaptively adjust the second group portrait through a first machine learning method.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for generating a power user group portrait based on a deep graph neural network according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for generating a power user group portrait based on a deep graph neural network according to any one of claims 1 to 7.

Images