Electric vehicle user multi-source data processing and feature extraction method and system
Through the graph neural network detection and elimination of outliers, the window interpolation method fills the data missing, and extracts features based on meta-learning, solves the problem of insufficient user charging behavior analysis in the existing technology, improves the accuracy of feature extraction, and is for electric power Car charging planning provides more accurate data support.
Patent Information
- Application Number
- CN202411675128.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-21
- Publication Date
- 2025-02-11
AI Technical Summary
The prior art has shortcomings in depicting the charging behavior of electric vehicle users and using multi-source real data for comprehensive analysis, resulting in low user characteristics accuracy and inability to provide accurate data support for electric vehicle charging planning.
Graph neural network is used to detect and eliminate abnormal points of multi-source data of electric vehicle users, fill in the missing data values using window interpolation method, and extract the user's feature based on meta-learning.
By integrating multi-source data, effectively identifying and eliminating outliers, filling in data loss, improving the accuracy of user feature extraction, and providing more accurate data support and decision-making basis for electric vehicle charging planning.
Smart Images
Figure CN119415929A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electric vehicle data analysis, and in particular to a method and system for processing and extracting multi-source data of electric vehicle users. Background Art
[0002] As the popularity of electric vehicles has triggered a significant increase in electricity demand, it has brought new challenges to the power grid system. In particular, charging electric vehicles during peak periods may put additional pressure on the distribution grid and may require expensive facility upgrades. Therefore, effectively managing demand is a key issue for grid operators. In this case, orderly charging is a promising solution because electric vehicles have the ability to provide flexibility to the grid within the framework of demand response (DR). The flexibility provided by electric vehicles can be used for various applications including grid congestion management, local solar power generation utilization and frequency response.
[0003] At present, most of the research on the charging demand of electric vehicle users is based on a single data set, mainly focusing on the trajectory data of operating vehicles or charging station data, and failing to fully consider the various factors that affect user choices. The existing user charging demand model is insufficient in describing user charging behavior and using multi-source real data for comprehensive analysis, and has certain limitations in research objects and perspectives. With the leap of big data technology, a large amount of data such as vehicle trajectories, mobile phone signals, and charging behavior records have been accumulated in the energy and transportation fields, opening up new ways to analyze user behavior, accurately predict charging demand, and optimize resource allocation. However, how to accurately and efficiently realize the multi-source data processing and feature extraction of electric vehicle users is still a technical problem that needs to be solved urgently. Summary of the invention
[0004] Technical problem to be solved by the present invention: In view of the above-mentioned problems in the prior art, a method and system for processing and extracting multi-source data of electric vehicle users are provided. The present invention aims to solve the problems that the existing user charging demand model is insufficient in describing user charging behavior and using multi-source real data for comprehensive analysis, and has certain limitations in research objects and perspectives, resulting in low accuracy of user characteristics and inability to provide accurate data support for electric vehicle charging planning.
[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: A method for processing and extracting multi-source data of electric vehicle users comprises the following steps: S1, collects multi-source data of electric vehicle users; S2, using graph neural network to detect and remove outliers from the collected multi-source data; S3, using window interpolation method to complete multi-source data of electric vehicles; S4, feature extraction of electric vehicle users based on meta-learning.
[0006] Optionally, the multi-source data of electric vehicle users collected in step S1 includes three types of data: electrical, weather and other types.
[0007] Optionally, among the multi-source data, electrical data include charging power, current, voltage and charging time, weather data include temperature, humidity and weather type, and other types of data include user travel time.
[0008] Optionally, in step S2, using a graph neural network to detect and remove outliers from the collected multi-source data includes: S2.1, for the collected multi-source data, the multi-source data at each moment is taken as a node, and the multi-source data set at each moment is taken as a generation graph structure, and any Graph Structure of Graph Neural Networks The function expression is: , In the above formula, For the A collection of graph neural network nodes that represent electrical data. For the A collection of graph neural network nodes representing weather data. For the A collection of graph neural network nodes representing other types of data; S2.2, using edge label information to initialize edge weights so that the graph neural network can perform explicit clustering of nodes through edge label information, including the support set in each sampled classification task Denoted as: , and there is , Query Set Denoted as: , And there is , ,in ~ They are No. 1~ samples, each of which consists of data and its corresponding tags constitute, is the number of categories of each meta-task in the classification task, is the number of known labeled samples in each category, For the existing basic data set, ~ Respectively ~ data points, is the number of query sets composed of samples that are not repeated in the support set. Samples With Samples , initialize the edge weights according to the following formula: , , In the above formula, For the Samples With Samples The edge label of the same type is 1, otherwise it is 0; and Respectively Samples With Samples Tags, is the initial edge weight; S2.3, the initial edge weight is obtained according to the edge label to build the initial graph structure, and the graph structure is input into the graph neural network for aggregation and propagation. When the graph neural network At this time, remember Graph Structure of Graph Neural Networks Any two node sets , For the A collection of graph neural network nodes that represent electrical data. For the A collection of graph neural network nodes representing weather data. For the The graph neural network node set representing other types of data, and the corresponding edge weight is , then The calculation function expression of the edge weight is: , , In the above formula, , For the Substitute the edge weights of the two edge weight update modules, and is the edge weight update module, and Respectively The parameters of the two edge weight update modules in the graph neural network, where the edge weight update module is the feature extraction backbone network, and Respectively Graph Structure of Graph Neural Networks No. and nodes, For the and The edge weights between nodes are and Respectively Graph Structure of Graph Neural Networks No. and nodes; S2.4, the binary cross entropy loss function BCE is used to calculate the edge weight loss of whether the predicted edge weights are connected. If the edge weight loss reaches the preset threshold or the number of iterations reaches the preset threshold, the update is stopped to obtain the label value, and more similar data are set as normal values, and fewer are set as outliers, thereby obtaining the outlier removal result; otherwise, jump to step S2.3 to continue iteration.
[0009] Optionally, when the binary cross entropy loss function BCE is used in step S2.4 to calculate the edge weight loss for the predicted edge weights, for the For graph neural network, the calculation function expression of edge weight loss is: , In the above formula, For the Edge weight loss of graph neural network, is the number of samples of the meta-task, is the number of categories for each meta-task, is the number of known labeled samples in each category, is the binary cross entropy loss function; For the Samples With Samples In the l Edge weights in graph neural networks, For the Samples With Samples The edge label is 1 if they are of the same type, otherwise it is 0.
[0010] Optionally, in step S3, the windowed interpolation method is used to fill in the electric vehicle multi-source data, including using the windowed interpolation method to fill in the missing values: for a given size k, including data and The missing values are filled in according to the following formula: , In the above formula, is the estimated value of the missing value at position i in the window, is a normalization factor, which is used to smooth the values in the window so that the estimated value is more consistent with the trend of the original data. is the data value at position j in the window.
[0011] Optionally, extracting features of electric vehicle users based on meta-learning in step S4 includes: S4.1, different node sets of the electric vehicle multi-source dataset after windowing and interpolation processing are defined as different meta-tasks; S4.2, define the optimization goal as optimizing the global model parameters of the meta-learning model Minimize the loss in the new meta-task: , In the above formula, are the optimized global model parameters, are global model parameters, is the learning rate, for The gradient of Based on the iteration direction And the global model parameters The edge weight loss obtained is: , , In the above formula, is the edge weight loss, is the number of samples of the meta-task, is the number of categories for each meta-task, is the number of known labeled samples in each category, is the binary cross entropy loss function; is the i-th sample With the jth sample In the l Edge weights in graph neural networks, is the i-th sample With the jth sample The edge label of the same type is 1, otherwise it is 0; is the edge weight loss of the i-th task, For the establishment of parameters The network model composed of S4.3, update the reinforcement learning strategy according to the following formula so that the selected action can maximize the cumulative reward: , In the above formula, for The gradient of is the loss function, is the meta-learning rate, It represents the ability of the strategy function to analyze user characteristics at the learning rate at time t. is a parameterized policy function, is the optimizer learning rate at time t, is the user feature at time t, To accumulate rewards, Indicates time; is the gradient direction of the edge weight loss; based on the optimization objective, the global model parameter 𝜃 of the meta-learning model is updated according to the following formula: , In the above formula, is the meta-learning rate, For the task overview, is the edge weight loss of the i-th batch, is the i-th batch of meta-tasks, Based on Meta-learning model, is the global model parameter after the i-th batch iteration, and the final meta-learning model is taken as the optimal meta-learning model ; S4.3, use the trained optimal meta-learning model to extract electric vehicle user features: , In the above formula, is an output vector composed of user characteristics, wherein the user characteristics include charging duration, charging frequency, charging station location and charging time, is the optimal meta-learning model, The graph structure of the input graph neural network.
[0012] In addition, the present invention also provides an electric vehicle user multi-source data processing and feature extraction system, comprising an interconnected microprocessor and a memory, wherein the microprocessor is programmed or configured to execute the electric vehicle user multi-source data processing and feature extraction method.
[0013] In addition, the present invention also provides a computer-readable storage medium, in which a computer program or instruction is stored. The computer program or instruction is programmed or configured to execute the electric vehicle user multi-source data processing and feature extraction method through a processor.
[0014] In addition, the present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the electric vehicle user multi-source data processing and feature extraction method through a processor.
[0015] Compared with the prior art, the present invention has the following advantages: Most of the existing studies on the charging demand of electric vehicle users are based on a single data set, mainly focusing on the analysis of a single dimension such as the trajectory data of operating vehicles or the charging station data, and failing to fully consider the various factors that affect user choices. The existing user charging demand model has deficiencies in depicting user charging behavior and using multi-source real data for comprehensive analysis, and has certain limitations in research objects and perspectives. The present invention can integrate multi-source data such as vehicle trajectory data, charging data, and traffic data, use a graph neural network model to remove outliers from multi-source data of electric vehicles, and use a windowed interpolation algorithm to fill in missing values; and then extract electric vehicle user features based on meta-learning, which can solve the problems of low accuracy of user features and inability to provide accurate data support for electric vehicle charging planning caused by the existing user charging demand model in depicting user charging behavior and using multi-source real data for comprehensive analysis, and the limitations in research objects and perspectives, and provide more accurate data support and decision-making basis for electric vehicle charging planning. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 Schematic diagram of the basic flow of the method of the embodiment of the present invention. DETAILED DESCRIPTION
[0017] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0018] like Figure 1 As shown, the electric vehicle user multi-source data processing and feature extraction method of this embodiment includes the following steps: S1, collects multi-source data of electric vehicle users; S2, using graph neural network to detect and remove outliers from the collected multi-source data; S3, using window interpolation method to complete multi-source data of electric vehicles; S4, feature extraction of electric vehicle users based on meta-learning.
[0019] The multi-source data of electric vehicle users collected in step S1 of this embodiment include three types of data: electrical, weather, and other types, which may include multi-source data such as vehicle trajectory data, charging data, and traffic condition data. As an optional implementation, in the multi-source data of this embodiment, the electrical data include charging power, current, voltage, and charging time, the weather data include temperature, humidity, and weather type, and the other data include user travel time.
[0020] In step S2 of this embodiment, using a graph neural network to detect and remove outliers from the collected multi-source data includes: S2.1, for the collected multi-source data, including electric data such as charging power, current, voltage, charging time, etc. of electric vehicle users, meteorological data such as weather temperature, user travel time and other data, etc., the multi-source data at each moment is used as a node, and the multi-source data set at each moment is used as a first-generation graph structure, and any Graph Structure of Graph Neural Networks The function expression is: , In the above formula, For the A collection of graph neural network nodes that represent electrical data. For the A collection of graph neural network nodes representing weather data. For the The initial node set is denoted as V0={ v 0,i}, the initial node is set to the first data point of the data sample; S2.2, using edge label information to initialize edge weights so that the graph neural network can perform explicit clustering of nodes through edge label information, including the support set in each sampled classification task Denoted as: , and there is , Query Set Denoted as: , And there is , ,in ~ They are No. 1~ samples, each of which consists of data and its corresponding tags constitute, is the number of categories of each meta-task in the classification task, is the number of known labeled samples in each category, For the existing basic data set, ~ Respectively ~ data points, is the number of query sets composed of samples that are not repeated in the support set. Samples With Samples , initialize the edge weights according to the following formula: , , In the above formula, For the Samples With Samples The edge label of the same type is 1, otherwise it is 0; and Respectively Samples With Samples Tags, is the initial edge weight; S2.3, constructing the initial graph structure by obtaining the initial edge weight according to the edge label, including: the outlier detection task passes through the initial node, constructing the initial graph structure after obtaining the initial edge weight according to the edge label; inputting the graph structure into the graph neural network for aggregation and propagation, When the graph neural network At this time, remember Graph Structure of Graph Neural Networks Any two node sets , For the A collection of graph neural network nodes that represent electrical data. For the A collection of graph neural network nodes representing weather data. For the The graph neural network node set representing other types of data, and the corresponding edge weight is , then The calculation function expression of the edge weight is: , , In the above formula, , For the Substitute the edge weights of the two edge weight update modules, and is the edge weight update module, and Respectively The parameters of the two edge weight update modules in the graph neural network, where the edge weight update module is the feature extraction backbone network, and Respectively Graph Structure of Graph Neural Networks No. and nodes, For the and The edge weights between nodes are and Respectively Graph Structure of Graph Neural Networks No. and nodes; S2.4, the binary cross entropy loss function BCE is used to calculate the edge weight loss of whether the predicted edge weights are connected. If the edge weight loss reaches the preset threshold or the number of iterations reaches the preset threshold, the update is stopped to obtain the label value, and more similar data are set as normal values, and fewer are set as outliers, thereby obtaining the outlier removal result; otherwise, jump to step S2.3 to continue iteration.
[0021] In step S2.4 of this embodiment, when the binary cross entropy loss function BCE is used to calculate the edge weight loss for the predicted edge weights, For graph neural network, the calculation function expression of edge weight loss is: , In the above formula, For the Edge weight loss of graph neural network, is the number of samples of the meta-task, is the number of categories for each meta-task, is the number of known labeled samples in each category, is the binary cross entropy loss function; For the Samples With Samples In the l Edge weights in graph neural networks, For the Samples With Samples The edge label is 1 when the data is of the same type, otherwise it is 0. Set the edge weight loss threshold or the maximum number of iterations. When the threshold is reached, stop updating and get the label value. Set more data of the same type as normal values and less data as abnormal values, so as to get the result of outlier removal.
[0022] In step S3 of this embodiment, the windowed interpolation method is used to fill in the multi-source data of electric vehicles, including using the windowed interpolation method to fill in the missing values: for a given size k, including data and The missing values are filled in according to the following formula: , In the above formula, is the estimated value of the missing value at position i in the window, is a normalization factor, which is used to smooth the values in the window so that the estimated value is more consistent with the trend of the original data. is the data value at position j in the window. Step S3 of this embodiment specifically includes: S3.1, data preparation: ensure that the data set has been processed for outliers, that is, obvious abnormal data points are deleted or replaced with reasonable values. Identify missing values in the data, which are usually represented by NaN, NULL or specific markers in the data set. S3.2, determine the window size: select a suitable window. The choice of window size should be based on the characteristics of the data and the analysis requirements, and usually needs to be determined through experiments or domain knowledge. S3.3, select the interpolation method: in the window interpolation method, weighted linear interpolation is used for completion. Consider a given size of k, including data and The window uses weighted linear interpolation: , In the above formula, is the estimated value of the missing value at position i in the window, is a normalization factor, which is used to smooth the values in the window so that the estimated value is more consistent with the trend of the original data. is the data value at position j in the window. S3.4, apply interpolation: for each missing value, use weighted linear interpolation to calculate the estimated value of the missing value based on the data points in the window before and after it. S3.5, update the data set: replace the missing values in the data set with the estimated value obtained by interpolation, complete the data set filling, and obtain the electric vehicle multi-source data set after window interpolation processing.
[0023] In step S4 of this embodiment, feature extraction of electric vehicle users based on meta-learning includes: S4.1, different node sets of the electric vehicle multi-source dataset after windowing and interpolation processing are defined as different meta-tasks; S4.2, define the optimization goal as optimizing the global model parameters of the meta-learning model Minimize the loss in the new meta-task: , In the above formula, are the optimized global model parameters, is the global model parameter, is the learning rate, for The gradient of Based on the iteration direction And the global model parameters The edge weight loss obtained is: , , In the above formula, is the edge weight loss, is the number of samples of the meta-task, is the number of categories for each meta-task, is the number of known labeled samples in each category, is the binary cross entropy loss function; is the i-th sample With the jth sample In the l Edge weights in graph neural networks, is the i-th sample With the jth sample The edge label of the same type is 1, otherwise it is 0; is the edge weight loss of the i-th task, For the establishment of parameters The network model composed of S4.3, update the reinforcement learning strategy according to the following formula so that the selected action can maximize the cumulative reward: , In the above formula, for The gradient of is the loss function, is the meta-learning rate, It represents the ability of the strategy function to analyze user characteristics at the learning rate at time t. is a parameterized policy function, is the optimizer learning rate at time t, is the user feature at time t, To accumulate rewards, Indicates time; is the gradient direction of the edge weight loss; based on the optimization objective, the global model parameter 𝜃 of the meta-learning model is updated according to the following formula: , In the above formula, is the meta-learning rate, For the task overview, is the edge weight loss of the i-th batch, is the i-th batch of meta-tasks, Based on Meta-learning model, is the global model parameter after the i-th batch iteration, and the final meta-learning model is taken as the optimal meta-learning model ; S4.3, use the trained optimal meta-learning model to extract electric vehicle user features: , In the above formula, is an output vector composed of user characteristics, wherein the user characteristics include charging duration, charging frequency, charging station location and charging time, is the optimal meta-learning model, The graph structure of the input graph neural network.
[0024] In order to verify the multi-source data processing and feature extraction method of electric vehicle users in this embodiment, this embodiment uses well-known indicators such as anomaly detection accuracy, false alarm rate, recall rate, F1 score, feature extraction accuracy, feature coverage, etc. to conduct experiments. In the experiment, the results of outlier detection and elimination of multi-source data of electric vehicle users collected using graph neural network are shown in Figure 1. The results of feature extraction of electric vehicle users based on meta-learning are shown in Table 2. The comparison of the experimental results of the method of this embodiment and the existing meta-learning is shown in Table 3.
[0025] Table 1 Outlier detection and removal effects
[0026] Table 2 Effects of feature extraction and user behavior analysis
[0027] Table 3 Comparison of experimental results with other existing methods
[0028] The outlier detection and elimination effects in Table 1 show that the multi-source data processing and feature extraction method for electric vehicle users in this embodiment can effectively identify and eliminate outliers in the data, and the accuracy of outlier detection reaches 95%, with a false alarm rate of only 2%. The results of feature extraction of electric vehicle users based on meta-learning shown in Table 2 show that the multi-source data processing and feature extraction method for electric vehicle users in this embodiment can extract key features reflecting user charging behavior from complex multi-source data, including charging time, charging frequency, charging station location and charging time, etc. The accuracy of feature extraction is improved by 10%, and the feature extraction accuracy of charging time series data reaches 90%, which significantly improves the accuracy and efficiency of electric vehicle charging planning, and is very helpful for mining node information and edge weight information. The experimental results of the method in this embodiment and the existing meta-learning shown in Table 3 show that the results of the multi-source data processing and feature extraction method for electric vehicle users in this embodiment exceed the existing meta-learning method, especially far exceeding the general machine learning method (feature extraction accuracy ≤ 85%, user behavior prediction accuracy ≤ 81%). These results indicate that the electric vehicle user multi-source data processing and feature extraction method of this embodiment performs well in terms of data quality and user behavior analysis, and provides strong support for electric vehicle charging management.
[0029] In summary, the multi-source data processing and feature extraction method for electric vehicle users in this embodiment is based on improved meta-learning of graph neural network to realize multi-source data processing and feature extraction for electric vehicle users. First, multi-source real data sets such as vehicle trajectory data, charging data, and traffic condition data are integrated, and the graph neural network model is used to remove outliers from multi-source data of electric vehicles. Then, the window difference algorithm is used to fill the missing values; and then the electric vehicle user features are extracted based on meta-learning to provide more accurate data support and decision-making basis for electric vehicle charging planning. The multi-source data processing and feature extraction method for electric vehicle users in this embodiment integrates multi-source data, including operating vehicle trajectories, charging station records, private car travel, traffic conditions, and interest point queries, etc., implements real data mining practices, and can accurately capture user behavior preferences through data feature extraction and analysis to better serve the popularization of electric vehicles and the construction and development of new power systems.
[0030] In addition, this embodiment also provides an electric vehicle user multi-source data processing and feature extraction system, including an interconnected microprocessor and a memory, wherein the microprocessor is programmed or configured to execute the electric vehicle user multi-source data processing and feature extraction method.
[0031] In addition, this embodiment also provides a computer-readable storage medium, in which a computer program or instruction is stored. The computer program or instruction is programmed or configured to execute the electric vehicle user multi-source data processing and feature extraction method through a processor.
[0032] In addition, this embodiment also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the electric vehicle user multi-source data processing and feature extraction method through a processor.
[0033] Those skilled in the art should understand that the technical solutions provided by the embodiments of the present application may be in the form of methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0034] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.
Claims
1. A method for processing and extracting multi-source data of electric vehicle users, characterized in that: The steps include: S1, collects multi-source data of electric vehicle users; S2, using graph neural network to detect and remove outliers from the collected multi-source data; S3, using window interpolation method to complete multi-source data of electric vehicles; S4, feature extraction of electric vehicle users based on meta-learning.
2. The electric vehicle user multi-source data processing and feature extraction method according to claim 1 is characterized in that: The multi-source data of electric vehicle users collected in step S1 includes three types of data: electrical, weather, and other.
3. The electric vehicle user multi-source data processing and feature extraction method according to claim 2 is characterized in that: Among the multi-source data, electrical data includes charging power, current, voltage and charging time, weather data includes temperature, humidity and weather type, and other types of data include user travel time.
4. The electric vehicle user multi-source data processing and feature extraction method according to claim 2 is characterized in that: In step S2, the graph neural network is used to detect and remove outliers from the collected multi-source data, including: S2.1, for the collected multi-source data, the multi-source data at each moment is taken as a node, and the multi-source data set at each moment is taken as a generation graph structure, and any Graph Structure of Graph Neural Networks The function expression is: ， In the above formula, For the A collection of graph neural network nodes representing electrical data. For the A collection of graph neural network nodes representing weather data. For the A collection of graph neural network nodes representing other types of data; S2.2, using edge label information to initialize edge weights so that the graph neural network can perform explicit clustering of nodes through edge label information, including the support set in each sampled classification task Denoted as: , and there is , Query Set Denoted as: ， And there is , ,in ~ They are No. 1~ samples, each of which consists of data and its corresponding tags constitute, is the number of categories of each meta-task in the classification task, is the number of known labeled samples in each category, For the existing basic data set, ~ Respectively ~ data points, is the number of query sets composed of samples that are not repeated in the support set. Samples With Samples , initialize the edge weights according to the following formula: ， ， In the above formula, For the Samples With Samples The edge label of the same type is 1, otherwise it is 0; and Respectively Samples With Samples Tags, is the initial edge weight; S2.3, the initial edge weight is obtained according to the edge label to build the initial graph structure, and the graph structure is input into the graph neural network for aggregation and propagation. When the graph neural network At this time, remember Graph Structure of Graph Neural Networks Any two node sets , For the A collection of graph neural network nodes representing electrical data. For the A collection of graph neural network nodes representing weather data. For the The graph neural network node set representing other types of data, and the corresponding edge weight is , then The calculation function expression of the edge weight is: ， ， In the above formula, , For the Substitute the edge weights of the two edge weight update modules, and is the edge weight update module, and Respectively The parameters of the two edge weight update modules in the graph neural network, where the edge weight update module is the feature extraction backbone network, and Respectively Graph Structure of Graph Neural Networks No. and nodes, For the and The edge weights between nodes are and Respectively Graph Structure of Graph Neural Networks No. and nodes; S2.4, the binary cross entropy loss function BCE is used to calculate the edge weight loss of whether the predicted edge weights are connected. If the edge weight loss reaches the preset threshold or the number of iterations reaches the preset threshold, the update is stopped to obtain the label value, and more similar data are set as normal values, and fewer are set as outliers, thereby obtaining the outlier removal result; otherwise, jump to step S2.3 to continue iteration.
5. The electric vehicle user multi-source data processing and feature extraction method according to claim 4 is characterized in that: In step S2.4, when the binary cross entropy loss function BCE is used to calculate the edge weight loss of the predicted edge weights, for the For graph neural network, the calculation function expression of edge weight loss is: ， In the above formula, For the Edge weight loss of graph neural network, is the number of samples of the meta-task, is the number of categories for each meta-task, is the number of known labeled samples in each category, is the binary cross entropy loss function; For the Samples With Samples In the l Edge weights in graph neural networks, For the Samples With Samples The edge label is 1 if they are of the same type, otherwise it is 0.
6. The electric vehicle user multi-source data processing and feature extraction method according to claim 1 is characterized in that: In step S3, the windowed interpolation method is used to fill in the multi-source data of electric vehicles, including using the windowed interpolation method to fill in the missing values: for a given size k, including data and The missing values are filled in according to the following formula: ， In the above formula, is the estimated value of the missing value at position i in the window, is a normalization factor, which is used to smooth the values in the window so that the estimated value is more consistent with the trend of the original data. is the data value at position j in the window.
7. The electric vehicle user multi-source data processing and feature extraction method according to claim 1 is characterized in that: The feature extraction of electric vehicle users based on meta-learning in step S4 includes: S4.1, different node sets of the electric vehicle multi-source dataset after windowing and interpolation processing are defined as different meta-tasks; S4.2, define the optimization goal as optimizing the global model parameters of the meta-learning model Minimize the loss in the new meta-task: ， In the above formula, are the optimized global model parameters, are global model parameters, is the learning rate, for The gradient of Based on the iteration direction And the global model parameters The edge weight loss obtained is: ， ， In the above formula, is the edge weight loss, is the number of samples of the meta-task, is the number of categories for each meta-task, is the number of known labeled samples in each category, is the binary cross entropy loss function; is the i-th sample With the jth sample In the l Edge weights in graph neural networks, is the i-th sample With the jth sample The edge label of the same type is 1, otherwise it is 0; is the edge weight loss of the i-th task, For the establishment of parameters The network model composed of S4.3, update the reinforcement learning strategy according to the following formula so that the selected action can maximize the cumulative reward: ， In the above formula, for The gradient of is the loss function, is the meta-learning rate, It represents the ability of the strategy function to analyze user characteristics at the learning rate at time t. is a parameterized policy function, is the optimizer learning rate at time t, is the user feature at time t, To accumulate rewards, Indicates time; is the gradient direction of the edge weight loss; based on the optimization objective, the global model parameter 𝜃 of the meta-learning model is updated according to the following formula: ， In the above formula, is the meta-learning rate, For the task overview, is the edge weight loss of the i-th batch, is the i-th batch of meta-tasks, Based on Meta-learning model, is the global model parameter after the i-th batch iteration, and the final meta-learning model is taken as the optimal meta-learning model ; S4.3, use the trained optimal meta-learning model to extract electric vehicle user features: ， In the above formula, is an output vector composed of user characteristics, wherein the user characteristics include charging duration, charging frequency, charging station location and charging time, is the optimal meta-learning model, The graph structure of the input graph neural network.
8. An electric vehicle user multi-source data processing and feature extraction system, comprising an interconnected microprocessor and a memory, characterized in that: The microprocessor is programmed or configured to execute the electric vehicle user multi-source data processing and feature extraction method as described in any one of claims 1 to 7.
9. A computer-readable storage medium having a computer program or instruction stored therein, characterized in that: The computer program or instruction is programmed or configured to execute the electric vehicle user multi-source data processing and feature extraction method described in any one of claims 1 to 7 through a processor.
10. A computer program product comprising a computer program or instructions, characterized in that The computer program or instruction is programmed or configured to execute the electric vehicle user multi-source data processing and feature extraction method described in any one of claims 1 to 7 through a processor.