Method and system for identifying abnormal electricity consumption behaviors of multiple types of users and electronic equipment

Abstract

The invention belongs to the technical field of electronic information, and relates to a method and system for identifying abnormal electricity consumption behaviors of multiple types of users and electronic equipment. The method comprises the following steps: according to the difference of power consumption behaviors among different types of users and the space-time relevance of the same type of users in the power consumption behaviors, statistically analyzing user power consumption data to obtain characteristic indexes representing the power consumption behaviors of multiple types of users, and performing reduction on the characteristic indexes by utilizing a limit gradient lifting tree algorithm to form an optimal data set; the method comprises the following steps: establishing a user power consumption behavior preliminary identification model based on a graph convolutional neural network according to spatial relevance among power consumption behaviors of different users and time relevance of power consumption behaviors of the same user, and obtaining an optimized user power consumption behavior identification model by utilizing Bayesian optimization graph convolutional neural network parameters; the method achieves the quick and accurate recognition of the power consumption behaviors of the users, and solves a problem that the recognition precision is low during the recognition because of the high difference between the power consumption behaviors of different types of users in an existing method.

Description

Technical Field

[0001] The present invention belongs to the field of electronic information technology, and specifically relates to a method, system and electronic equipment for identifying abnormal electricity usage behaviors of multiple types of users. Background Art

[0002] Electricity is an indispensable resource in modern society, providing convenience for industry, commerce and residents. With the continuous growth of demand for electricity, one of the increasingly serious challenges facing the power system is the increase in electricity theft. Not only are the means and methods of electricity theft different, but high-tech means of electricity theft are more covert than traditional means of electricity theft. Electricity theft not only leads to a waste of energy resources, but also poses an economic threat to legitimate users and energy suppliers, and even causes damage to the power transmission and transformation equipment of the power grid. In serious cases, it will affect the safe operation of the power grid. Therefore, effectively detecting and combating electricity theft is crucial to ensure the healthy operation and economic benefits of the power system.

[0003] The nature of electricity consumption of power users can be divided into residential users, commercial users, industrial users and mixed residential and commercial users. Among them, residential users use single-phase power supply, industrial users use three-phase power supply, and commercial and mixed users use single-phase or three-phase power supply. At present, domestic and foreign scholars' research on user electricity consumption behavior identification mainly focuses on single-phase or three-phase power supply users, while there are relatively few studies on the electricity consumption behavior of users with both single-phase and three-phase power supply. There are multiple types of users in the substation area, and there are differences in electricity consumption behavior between different types of users, while the same type of users have temporal and spatial correlation in electricity consumption behavior. Therefore, accurate identification of the electricity consumption behavior of multiple types of users is crucial to assisting the operation decision-making of the power grid and optimizing the operation efficiency. Summary of the invention

[0004] The purpose of the present invention is to provide a method, system and electronic device for identifying abnormal electricity usage behaviors of multiple types of users, so as to improve the accuracy of identifying the electricity usage behaviors of multiple types of users in the prior art.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A method for identifying abnormal electricity consumption behaviors of multiple types of users, comprising: S1: Based on the differences in electricity consumption behaviors of different types of users and the temporal and spatial correlation of electricity consumption behaviors of users of the same type, the actual data of users' electricity consumption are statistically analyzed to obtain characteristic indicators that characterize the electricity consumption behaviors of multiple types of users; S2: Use the extreme gradient boosting tree to reduce the feature index and form the optimal feature set; S3: Based on the optimal feature set, the adjacency matrix is ​​established by using the spatiotemporal correlation of the optimal features. The optimal feature set and the adjacency matrix are input into the graph convolutional neural network to build a preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users based on the graph convolutional neural network. S4: obtaining and recording the evaluation index of the preliminary recognition model of abnormal electricity consumption behavior of multiple types of users, judging whether the accuracy of the preliminary recognition model of abnormal electricity consumption behavior of multiple types of users meets the preset requirements according to the evaluation index; if so, outputting the model as the abnormal electricity consumption behavior recognition model of multiple types of users; if not, optimizing the graph convolutional neural network parameters in the preliminary recognition model of abnormal electricity consumption behavior of multiple types of users by using Bayesian optimization to obtain a hyperparameter combination, taking the hyperparameter combination as input, repeating S3 to S4 to obtain the abnormal electricity consumption behavior recognition model of multiple types of users; S5: Input the optimal feature set into the abnormal power consumption behavior recognition model of multiple types of users to identify the abnormal power consumption behavior.

[0006] Preferably, in said S1, the multiple types of users include single-phase users and three-phase users; The characteristic indicators of the single-phase user's electricity consumption behavior include: the user's daily frozen electricity, daily average current, daily average voltage, zero-live current difference, voltage imbalance and current imbalance; The characteristic indicators of the three-phase user's electricity consumption behavior include: the user's daily frozen electricity, daily average three-phase voltage, daily average three-phase current, zero-live three-phase current difference, current imbalance and voltage imbalance; The characteristic indicators characterizing the electricity consumption behavior of multiple types of users include: daily frozen electricity consumption; daily average three-phase current includes: average current of phase A, average current of phase B, average current of phase C; daily average three-phase voltage includes: average voltage of phase A, average voltage of phase B, average voltage of phase C; zero-live current difference, voltage imbalance and current imbalance.

[0007] Preferably, in S2, the feature indicators are reduced using the extreme gradient boosting tree, the importance scores of different feature indicators are calculated, they are sorted from large to small according to the importance scores, and the feature indicators whose importance scores are less than a preset value are reduced.

[0008] Preferably, in S3, the adjacency matrix is ​​constructed by taking the optimal feature set as a node and connecting it with K nearest neighbor nodes in space through undirected edges to generate an adaptive adjacency matrix , the adjacency matrix calculation formula is as follows:

[0009]

[0010] in, is the adjacency matrix At the node With Node The element values ​​between For Node The set of K nearest neighbors determined by distance, For Node The set of K nearest neighbors determined by distance, is a parameter used to adjust the effect of distance, selected through experience or cross-validation; The Euclidean distance between feature vectors is used to represent the similarity between nodes; For the The node eigenvalues; For the The node eigenvalues; is the total number of characteristic indicators.

[0011] Preferably, the optimal feature set and the adjacency matrix are input into a graph convolutional neural network to construct a preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users based on a graph convolutional neural network, specifically:

[0012] In the formula, Indicates +1 layer of nodes The features of the optimal feature set of input for ; Representation Node The set of adjacent nodes of ; is the activation function; is the normalization factor, for nodes The degree and node The product of the degrees of is used to normalize the contribution of adjacent nodes; is the adjacency matrix At the node With Node The element values ​​between Indicates The weight matrix of the layer has self-learning function; Indicates Layer Node characteristics.

[0013] Preferably, the graph convolutional neural network parameters in the multi-type user power consumption behavior abnormal preliminary identification model are optimized by Bayesian to obtain a hyperparameter combination, the principle of which is expressed by the following formula:

[0014] In the formula, is the evaluation index value under the optimal hyperparameter combination; is the model evaluation indicator; Search space for hyperparameter combinations; Indicates the maximum model evaluation index value; is a combination of hyperparameters; is the optimal hyperparameter combination.

[0015] Preferably, the evaluation indicators include: accuracy, precision, recall, F1 score and area under the curve.

[0016] In a second aspect, the present application also discloses a system for constructing a model for identifying abnormal electricity consumption behavior of multiple types of users, including: An acquisition unit is used to perform statistical analysis on actual data of user electricity consumption according to the differences in electricity consumption behaviors of different types of users and the temporal and spatial correlation of electricity consumption behaviors of users of the same type, so as to obtain characteristic indicators characterizing the electricity consumption behaviors of multiple types of users; The reduction unit is used to reduce the feature index using the extreme gradient boosting tree to form the optimal feature set; A model building unit is used to establish an adjacency matrix based on the optimal feature set and the spatiotemporal correlation of the optimal features, input the optimal feature set and the adjacency matrix into the graph convolutional neural network, and build a preliminary recognition model for abnormal power consumption behaviors of multiple types of users based on the graph convolutional neural network; A model optimization unit is used to obtain and record evaluation indicators of the preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users, and judge whether the accuracy of the preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users meets preset requirements according to the evaluation indicators. If the requirements are met, the model is output as the abnormal electricity consumption behavior recognition model for multiple types of users; if not, the graph convolutional neural network parameters in the preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users are optimized by Bayesian optimization to obtain a hyperparameter combination, and the hyperparameter combination is used as input, and returned to the model construction unit for optimization to obtain the abnormal electricity consumption behavior recognition model for multiple types of users; The recognition unit is used to input the optimal feature set into the abnormal power consumption behavior recognition model of multiple types of users to identify the abnormal power consumption behavior.

[0017] In a third aspect, an electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above-mentioned methods for identifying abnormal electricity usage behaviors of multiple types of users when executing the computer program.

[0018] In a fourth aspect, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for identifying abnormal electricity consumption behavior of multiple types of users as described in any one of the above are implemented.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1) The present invention provides a method for identifying abnormal electricity consumption behaviors of multiple types of users, which solves the problem that the existing identification methods face the risk of low identification accuracy due to the high differences between the electricity consumption behaviors of different types of users. The method can model different users according to their spatiotemporal correlations in electricity consumption behaviors, and achieve high-accuracy identification of users with abnormal electricity consumption behaviors, thus providing a scientific basis and feasible solutions for the safe operation and abnormal warning of the distribution network; 2) The present invention can fully acquire and utilize multi-source information of user electricity consumption data, use statistical methods to extract characteristic indicators that characterize user electricity consumption behavior, and use extreme gradient boosting tree to simplify the above characteristics to form an optimal feature set. The optimal feature indicator set is used as the input of the graph convolutional neural network, and Bayesian optimization is used to update it. A multi-type user electricity consumption behavior anomaly recognition model based on the graph convolutional neural network is established, which improves the efficiency of abnormal electricity consumption behavior detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a flow chart of model optimization according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions 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 part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. 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.

[0024] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0025] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0026] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0027] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0028] The present invention is further described in detail below in conjunction with the accompanying drawings: See also Figure 1 A method for identifying abnormal electricity consumption behaviors of multiple types of users, characterized by comprising: S1: Based on the differences in electricity consumption behaviors of different types of users and the temporal and spatial correlation of electricity consumption behaviors of users of the same type, the actual data of users' electricity consumption are statistically analyzed to obtain characteristic indicators that characterize the electricity consumption behaviors of multiple types of users; S2: Use the extreme gradient boosting tree to reduce the feature index and form the optimal feature set; S3: Based on the optimal feature set, the adjacency matrix is ​​established by using the spatiotemporal correlation of the optimal features. The optimal feature set and the adjacency matrix are input into the graph convolutional neural network to build a preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users based on the graph convolutional neural network. S4: obtaining and recording the evaluation index of the preliminary recognition model of abnormal electricity consumption behavior of multiple types of users, judging whether the accuracy of the preliminary recognition model of abnormal electricity consumption behavior of multiple types of users meets the preset requirements according to the evaluation index; if so, outputting the model as the abnormal electricity consumption behavior recognition model of multiple types of users; if not, optimizing the graph convolutional neural network parameters in the preliminary recognition model of abnormal electricity consumption behavior of multiple types of users by using Bayesian optimization to obtain a hyperparameter combination, taking the hyperparameter combination as input, repeating S3 to S4 to obtain the abnormal electricity consumption behavior recognition model of multiple types of users; S5: Input the optimal feature set into the abnormal power consumption behavior recognition model of multiple types of users to identify the abnormal power consumption behavior.

[0029] In some embodiments, in S1, the multiple types of users include single-phase users and three-phase users; among which residential users are mostly single-phase, and industrial and commercial users are mostly three-phase.

[0030] The characteristic indicators of the single-phase user's electricity consumption behavior include: the user's daily frozen electricity, daily average current, daily average voltage, zero-live current difference, voltage imbalance and current imbalance; The characteristic indicators of the three-phase user's electricity consumption behavior include: the user's daily frozen electricity, daily average three-phase voltage, daily average three-phase current, zero-live three-phase current difference, current imbalance and voltage imbalance; The characteristic indicators characterizing the electricity consumption behavior of multiple types of users include: daily frozen electricity consumption; daily average three-phase current includes: average current of phase A, average current of phase B, average current of phase C; daily average three-phase voltage includes: average voltage of phase A, average voltage of phase B, average voltage of phase C; zero-live current difference, voltage imbalance and current imbalance.

[0031] The calculation process of the statistical characteristic indicators of user electricity consumption behavior is as follows: 1) User daily power freeze For users with different electricity consumption scales, their electricity consumption behaviors are significantly different, and the modeling results of users with different electricity consumption levels will be different. The calculation formula for daily frozen electricity consumption P is as follows:

[0032] In the formula, is the displayed power consumption at time t of the day; It is the displayed power at time t of the previous day.

[0033] 2) Daily average single-phase current and single-phase voltage The daily average current and voltage of single-phase users are both recorded as phase A. The calculation formulas for the daily average single-phase current and voltage are as follows:

[0034]

[0035] In the formula, , They are the average values ​​of single-phase current and voltage respectively; n=24, corresponding to 24 sampling points per day. For three-phase users, in addition to the A-phase current and voltage, there are also B- and C-phase currents and B- and C-phase voltages. According to the above formula, we can get , , , .

[0036] 3) Zero-live current difference The calculation formula for the zero-live current difference of a single-phase user is as follows:

[0037] In the formula, Indicates live wire current; Indicates the neutral current.

[0038] 4) Voltage imbalance Voltage imbalance reflects the difference between the voltages of each phase in the system and the magnitude of the voltage fluctuation of the user. For single-phase users, the voltage generally remains relatively balanced within a certain range when in normal power consumption. The voltage imbalance calculation formula is as follows:

[0039] In the formula, It is the maximum voltage value of the day; Minimum voltage value during the day.

[0040] For three-phase users, the three-phase voltage unbalance calculation formula is as follows:

[0041] In the formula, , , is the effective value of voltage; is the average value of three-phase voltage.

[0042] 5) Current imbalance The calculation formula for single-phase user current unbalance is as follows:

[0043] For three-phase users, the three-phase current unbalance degree I B The calculation formula is as follows:

[0044] In the formula, , , is the effective value of current; is the average value of three-phase current.

[0045] Therefore, the main characteristics that characterize the user's electricity consumption behavior are [P, , , , , , , , , ].

[0046] In some embodiments, in S2, the feature indicators are reduced using an extreme gradient boosting tree, the importance scores of different feature indicators are calculated, they are sorted from large to small according to the importance scores, and the feature indicators whose importance scores are less than a preset value are reduced.

[0047] XGBoost is a recently proposed machine learning algorithm that can effectively improve model prediction accuracy and learning efficiency. It is a reinforcement learning technique used to solve various problems. The XGBoost decision tree model can be represented by the following additive model. In each iteration, XGBoost adds a new tree and displays the linear combination of these trees:

[0048] In the formula, Indicates The predicted value of the model at the iteration; Indicates the number of decision trees; Indicates input samples; represents the set of all decision trees, It is The first tree The weight of each sample is assigned to the corresponding leaf node.

[0049] The objective function of XGBoost consists of two parts: the loss function and the regularization term. The calculation formula is as follows:

[0050]

[0051] In the formula, Indicates The objective function of the iteration; Indicates Samples( −1) Iteration prediction value; It is The regular term of the iterative model represents the The complexity of each sub-model plays a role in reducing the probability of overfitting; and is the regular term coefficient, which limits the complexity of the decision tree; Indicates the number of leaf nodes in the tree model.

[0052] The quality calculation formula that can be used to evaluate the tree model is as follows:

[0053] The smaller the objective function value, the better the tree model is and the more reasonable the node division is.

[0054] When splitting each layer of nodes, the algorithm uses step-by-step splitting and greedy strategies. Before and after the node splitting, the algorithm will calculate the reduction of the loss function, which is called the gain. During the iteration process, the algorithm ensures that the result of each node splitting is maximized, which is the greedy strategy. Taking feature X1 as the node splitting example, the optimal gain calculation formula for feature X1 when splitting the node is as follows:

[0055] in:

[0056] Gain is used as the objective function in the iterative process. Whenever the maximum gain of a tree node is obtained, the best result of this node splitting can be obtained. Through iteration, the best splitting points of all feature values ​​can be obtained respectively, until the actual category of the user calibrated by the expert is used as the final leaf node, that is, the classification result, and the optimal tree structure is obtained. When the tree model reaches the optimal state, the number of splits of each feature is counted, and the feature importance score is determined according to the number of splits. , the calculation formula is as follows:

[0057] in: is the total number of node splits in the optimal decision tree, For the The number of splits for a feature.

[0058] The optimal feature set is formed using feature indices with importance scores higher than the set threshold.

[0059] In some embodiments, in S3, the adjacency matrix is ​​constructed by taking the optimal feature set as a node and connecting it with K nearest neighbor nodes in the space through undirected edges to generate an adaptive adjacency matrix , the adjacency matrix calculation formula is as follows:

[0060]

[0061] in, is the adjacency matrix At the node With Node The element values ​​between For Node The set of K nearest neighbors determined by distance, For Node The set of K nearest neighbors determined by distance, is a parameter used to adjust the effect of distance, selected through experience or cross-validation; The Euclidean distance between feature vectors is used to represent the similarity between nodes; For the The node eigenvalues; For the The node eigenvalues; is the total number of characteristic indicators.

[0062] In some embodiments, the optimal feature set and the adjacency matrix are input into a graph convolutional neural network to construct a preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users based on a graph convolutional neural network, specifically:

[0063] In the formula, Indicates +1 layer of nodes The features of the optimal feature set of input for ; Representation Node The set of adjacent nodes of ; is the activation function, commonly used ones are ReLu, Sigmoid, etc.; is the normalization factor, for nodes The degree and node The product of the degrees of is used to normalize the contribution of adjacent nodes; is the adjacency matrix At the node With Node The element values ​​between Indicates The weight matrix of the layer has self-learning function; Indicates Layer Node characteristics.

[0064] After a finite number of iterations and feature calculations, the output feature is obtained after the last layer of graph convolution, which is recorded as Label.

[0065] In some embodiments, the graph convolutional neural network parameters in the multi-type user power consumption behavior abnormal preliminary identification model are optimized by Bayesian to obtain a hyperparameter combination, and the principle is expressed by the following formula:

[0066] In the formula, is the evaluation index value under the optimal hyperparameter combination; is the model evaluation indicator; Search space for hyperparameter combinations; Indicates the maximum model evaluation index value; is a combination of hyperparameters; is the optimal hyperparameter combination.

[0067] In the analysis of user electricity consumption behavior, the selection of graph convolutional neural network hyperparameters is closely related to model performance, which is specifically reflected in performance indicators such as model complexity, computational efficiency, and accuracy. Therefore, it is necessary to use an optimization algorithm to select hyperparameters during the training process. The Bayesian optimization algorithm seeks optimization through iterative calculation of the acquisition function and the probabilistic proxy model. Based on the existing prior knowledge and evaluation results, it calculates the optimal hyperparameter combination in a short time to improve model performance. When the model is first trained, the initial hidden layer nodes and the initial learning rate are set, the inverse of the average recall rate of the test set results is used as the loss function, and the minimum loss is used as the objective function, that is, to ensure that the number of missed detections for all categories is minimized. Through the Bayesian hyperparameter iterative optimization, the evaluation indicators of each generation of models and the current hyperparameter combination are calculated, and the optimal hyperparameter combination suitable for the model problem can be selected according to the model evaluation indicators.

[0068] When the Bayesian optimization algorithm is optimized in the hyperparameter space, the parameter search space includes the hidden layer node number search space and the learning rate search space.

[0069] In some embodiments, the evaluation indicators include: accuracy, precision, recall, F1 score and area under the curve.

[0070] For the multi-class user recognition method, each class is considered as a positive class and other classes are considered as negative classes. The classification performance indicators used to evaluate the results include: Accuracy (Acc), Precision (Precision), Recall (Recall), F1-score (F1) and Area Under the Curve (AUC). Finally, based on the recognition results of the test set, the average value of these indicators is calculated as the final multi-classification indicator value.

[0071] Accuracy refers to the proportion of correctly classified samples to the total number of samples. The calculation formula is as follows:

[0072] In the formula, TP is the correctly classified user; FN refers to the misclassification of the existing category user into another category; FP is the user of another category is wrongly classified into the current category; TN is correctly classified as other category users.

[0073] The precision rate is the ratio of the number of samples predicted as positive to the number of samples predicted as positive. The closer the value is to 1, the lower the misjudgment rate. The calculation formula is as follows:

[0074] The recall rate indicates the proportion of positive samples that are predicted to be correct among the total positive samples. The closer the value is to 1, the lower the missed rate is. The calculation formula is as follows:

[0075] The F1 score is the harmonic mean of precision and recall, which takes into account the accuracy and coverage of the model on the positive category. The calculation formula is as follows:

[0076] The area under the curve (AUC) is a commonly used indicator for evaluating the performance of classification models, and is usually used in receiver operating characteristic (ROC) curve analysis. AUC represents the area under the ROC curve, and its value range is between 0 and 1. The ROC curve shows the relationship between the true positive rate and the false positive rate when the classifier threshold changes. The calculation formula is as follows:

[0077]

[0078] The present invention proposes a method for analyzing abnormal electricity consumption behavior of users based on Bayesian-graph convolutional neural network. First, the electricity consumption history data of users in the substation area and the actual operation data of the current smart meter are collected, and the statistical characteristic indicators of electricity consumption behaviors of different users are mined. The above characteristic indicators are simplified by using the extreme gradient boosting tree algorithm to form an optimal data set; according to the spatial correlation between electricity consumption behaviors of different users and the spatiotemporal correlation of electricity consumption behaviors of the same user, a user electricity consumption behavior recognition model based on graph convolutional neural network is established, and the parameters of graph convolutional neural network are optimized by Bayesian to realize the rapid recognition of electricity consumption behaviors of users in the substation area.

[0079] The present invention discloses a method and system for identifying abnormal electricity consumption behaviors of multiple types of users, and relates to the field of electronic information technology. The specific steps are as follows: according to the characteristics of the differences in electricity consumption behaviors of different types of users and the spatiotemporal correlation of electricity consumption behaviors of the same type of users, statistical analysis is performed on the actual data of electricity consumption of users in a certain substation to extract characteristic indicators that characterize the electricity consumption behaviors of users; the above characteristic indicators are reduced by using the eXtreme Gradient Boosting (XGBoost) tree to form an optimal feature set; the optimal feature indicator set is used as the input of the graph convolutional neural network to establish a preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users based on the graph convolutional neural network; the Bayesian Optimization (BO) graph convolutional neural network (GCN) hyperparameter optimization model is used, and the model can be used to realize the rapid recognition of abnormal electricity consumption behaviors of different types of users. The model construction method proposed by the present invention makes full use of the spatiotemporal correlation of electricity consumption behaviors of users in the substation, reduces the influence of differences in electricity consumption habits between users on the recognition accuracy, and provides a feasible solution for auxiliary power grid operation decision-making and optimization of operation efficiency.

[0080] In some embodiments, a method for identifying abnormal electricity usage behaviors of multiple types of users includes: Step 1: Based on the differences in electricity consumption behaviors of different types of users and the temporal and spatial correlation of electricity consumption behaviors of the same type of users, the actual electricity consumption data of users in a certain area are statistically analyzed to extract characteristic indicators that characterize the electricity consumption behaviors of users; Step 2: Use the Extreme Gradient Boosting Tree (XGBoost) to reduce the above feature indicators to form the optimal feature set; Step 3: Use the optimal feature index set as the input of the graph convolutional neural network (GCN), and establish an adjacency matrix based on the spatiotemporal correlation of electricity consumption characteristics, and then preliminarily build a multi-type user electricity consumption behavior abnormality recognition model based on the graph convolutional neural network; Graph Neural Network (GNN) is a deep learning algorithm that has recently gained popularity in the field of relationship analysis and classification prediction. In the context of electricity user behavior segmentation, GNN can construct each user's historical electricity consumption data into a multi-node interconnected graph structure according to temporal and spatial correlations, where each node represents daily user electricity consumption information and the edges represent similar relationships on the time and space scales. Graph Convolutional Neural Network (GCN) is a commonly used feature extraction algorithm in GNN, which performs convolution operations on graph structured data to extract relationship features between nodes. Graph structure , represents the node set, that is, the optimal feature set mentioned in step 2; Represents a set of edges, i.e., the node associations represented by the adjacency matrix.

[0081] The adjacency matrix is ​​constructed by taking the statistical features of the user's daily electricity consumption behavior as a node, that is, each node includes the optimal features obtained in step 2, and connects it with the K nearest neighbor (KNN) nodes in the space through undirected edges to generate an adaptive adjacency matrix , and the corresponding feature matrix Together they serve as the input of the GCN model for training the model. The adjacency matrix calculation formula is as follows:

[0082]

[0083] in, For Node The set of K nearest neighbors determined by distance, For Node The set of K nearest neighbors determined by distance, is a parameter used to adjust the effect of distance and can be selected through experience or cross-validation. The Euclidean distance between feature vectors is used to represent the similarity between nodes.

[0084] The definition of graph convolution is as follows:

[0085] In the formula, Indicates Layer Node The characteristics of for ; Representation Node The set of adjacent nodes of ; Indicates The weight matrix of the layer usually has self-learning function; is the normalization factor, usually a node The degree and node The product of the degrees of is used to normalize the contribution of adjacent nodes; is the activation function, commonly used ones are ReLu, Sigmoid, etc.

[0086] Step 4: Use Bayesian optimization (BO) of graph convolutional neural network hyperparameters to obtain a multi-type user power consumption behavior anomaly recognition model based on graph convolutional neural network with better performance, and realize rapid recognition of abnormal power consumption behaviors of different types of users; When identifying abnormal electricity usage behavior, accuracy, precision, recall rate, F1 score, and area under the curve are usually used to judge the performance of the recognition model.

[0087] Step 5: Input the optimal feature set into the abnormal power consumption behavior recognition model of multiple types of users to identify abnormal power consumption behavior.

[0088] [Example] A method for identifying abnormal electricity consumption behaviors of multiple types of users, the method comprising the following steps: S1. Based on the differences in electricity consumption behaviors of different types of users and the temporal and spatial correlation of electricity consumption behaviors of the same type of users, the actual data of electricity consumption of users in a certain area are statistically analyzed to extract characteristic indicators that characterize the electricity consumption behaviors of users. The feature extraction process is as follows: 1) According to the data type required for abnormal power consumption behavior detection, the required raw data is collected from the distribution network intelligent information collection system. The raw data includes:

[0089] 2) Process the original data and calculate the statistical characteristics of single-phase users’ electricity consumption, including daily frozen electricity, daily average current, daily average voltage, zero-live current difference, voltage imbalance, and current imbalance. The statistical characteristics of three-phase users’ electricity consumption mainly include daily frozen electricity, daily average three-phase voltage, daily average three-phase current, zero-live three-phase current difference, current imbalance, and voltage imbalance.

[0090] 3) The calculation process of the statistical characteristic indicators of user electricity consumption behavior is as follows: ①User daily power freeze For users with different electricity consumption scales, their electricity consumption behaviors are significantly different, and the modeling results of users with different electricity consumption levels will be different. The calculation formula for daily frozen electricity consumption P is as follows:

[0091] In the formula, is the displayed power consumption at time t of the day; It is the displayed power at time t of the previous day.

[0092] ② Daily average single-phase current and single-phase voltage The daily average current and voltage of single-phase users are both recorded as phase A. The calculation formulas for the daily average single-phase current and voltage are as follows:

[0093]

[0094] In the formula, , They are the average current and voltage of phase A respectively; n=24, corresponding to 24 sampling points per day. For three-phase users, in addition to the current and voltage of phase A, there are also currents and voltages of phases B and C. The average current of phase B is calculated according to the above formula , average current of phase C , average voltage of phase B , average voltage of phase C .

[0095] ③Zero-live current difference The calculation formula for the zero-live current difference of a single-phase user is as follows:

[0096] In the formula, Indicates live wire current; Indicates the neutral current.

[0097] For three-phase users, the calculation formula of the zero-live current difference is as follows:

[0098] In the formula, , , They are the A, B, and C phase live wire currents respectively; is the neutral current.

[0099] ④Voltage imbalance Voltage imbalance reflects the difference between the voltages of each phase in the system and the magnitude of the voltage fluctuation of the user. For single-phase users, the voltage generally remains relatively balanced within a certain range when in normal power consumption. The voltage imbalance calculation formula is as follows:

[0100] In the formula, It is the maximum voltage value of the day; Minimum voltage value during the day.

[0101] For three-phase users, the three-phase voltage unbalance calculation formula is as follows:

[0102] In the formula, , , is the effective value of voltage; is the average value of three-phase voltage.

[0103] ⑤Current imbalance The calculation formula for single-phase user current unbalance is as follows:

[0104] For three-phase users, the three-phase current unbalance degree I B The calculation formula is as follows:

[0105] In the formula, , , is the effective value of current; is the average value of three-phase current.

[0106] The main features characterizing the user's electricity consumption behavior calculated from the original data are:

[0107] S2. Use the Extreme Gradient Boosting Tree (XGBoost) to reduce the above feature indicators. By calculating the importance scores of different features and sorting them, the features with importance scores less than 6% are simplified to obtain the following optimal feature set:

[0108] XGBoost is a recently proposed machine learning algorithm that can effectively improve model prediction accuracy and learning efficiency. It is a reinforcement learning technique used to solve various problems. The XGBoost decision tree model can be represented by the following additive model. In each iteration, XGBoost adds a new tree and displays the linear combination of these trees:

[0109] In the formula, Indicates The predicted value of the model at iteration; Indicates the number of decision trees; Indicates input samples; represents the set of all decision trees, It is The first tree The weight of each sample is assigned to the corresponding leaf node.

[0110] The objective function of XGBoost consists of two parts: loss function and regularization term. The calculation formula is as follows:

[0111]

[0112] In the formula, Indicates The objective function of the iteration; Indicates Samples( −1) Iteration prediction value; It is The regular term of the iterative model represents the The complexity of each sub-model plays a role in reducing the probability of overfitting; and is the regular term coefficient, which limits the complexity of the decision tree; Indicates the number of leaf nodes in the tree model.

[0113] The quality calculation formula that can be used to evaluate the tree model is as follows:

[0114] The smaller the objective function value, the better the tree model is and the more reasonable the node division is.

[0115] When splitting each layer of nodes, the algorithm uses step-by-step splitting and greedy strategies. Before and after the node splitting, the algorithm will calculate the reduction of the loss function, which is called the gain. During the iteration process, the algorithm ensures that the result of each node splitting is maximized, which is the greedy strategy. Taking feature X1 as the node splitting example, the optimal gain calculation formula for feature X1 when splitting the node is as follows:

[0116] in:

[0117] Gain is used as the objective function in the iterative process. Whenever the maximum gain of a tree node is obtained, the best result of this node splitting can be obtained. Through iteration, the best splitting points of all feature values ​​can be obtained respectively, until the actual category of the user calibrated by the expert is used as the final leaf node, that is, the classification result, and the optimal tree structure is obtained. When the tree model reaches the optimal state, the number of splits of each feature is counted, and the feature importance score is determined according to the number of splits. , the calculation formula is as follows:

[0118] in: is the total number of node splits in the optimal decision tree, For the The number of splits for a feature.

[0119] The optimal feature set is formed using feature indices with importance scores higher than the set threshold.

[0120] S3. The optimal feature index set is used as the input of the graph convolutional neural network (GCN), and an adjacency matrix is ​​established according to the spatiotemporal correlation of electricity consumption characteristics, and then a preliminary recognition model for abnormal electricity consumption behavior of multiple types of users based on the graph convolutional neural network is constructed; Graph Neural Network (GNN) is a deep learning algorithm that has recently gained popularity in the field of relationship analysis and classification prediction. In the context of electricity user behavior segmentation, GNN can construct each user's historical electricity consumption data into a multi-node interconnected graph structure according to temporal and spatial correlations, where each node represents daily user electricity consumption information and the edges represent similar relationships on the time and space scales. Graph Convolutional Neural Network (GCN) is a commonly used feature extraction algorithm in GNN, which performs convolution operations on graph structured data to extract relationship features between nodes. Graph structure , represents the node set, that is, the optimal feature set mentioned in step 2; Represents a set of edges, i.e., the node associations represented by the adjacency matrix.

[0121] The adjacency matrix is ​​constructed by taking the statistical features of the user's daily electricity consumption behavior as a node, that is, each node includes the optimal features obtained in step 2, and connects it with the K nearest neighbor (KNN) nodes in the space through undirected edges to generate an adaptive adjacency matrix , and the corresponding feature matrix Together they serve as the input of the GCN model for training the model. The adjacency matrix calculation formula is as follows:

[0122]

[0123] in, For Node The set of K nearest neighbors determined by distance, For Node The set of K nearest neighbors determined by distance, is a parameter used to adjust the effect of distance and can be selected through experience or cross-validation. The Euclidean distance between feature vectors is used to represent the similarity between nodes.

[0124] The definition of graph convolution is as follows:

[0125] In the formula, Indicates Layer Node The characteristics of for ; Representation Node The set of adjacent nodes of ; Indicates The weight matrix of the layer usually has self-learning function; is the normalization factor, usually a node The degree and node The product of the degrees of is used to normalize the contribution of adjacent nodes; is the activation function, commonly used ones are ReLu, Sigmoid, etc.

[0126] After a finite number of iterations and feature calculations, the output feature is obtained after the last layer of graph convolution, which is recorded as Label.

[0127] Graph Convolutional Neural Network Modeling of Optimal Datasets,When modeling the graph convolutional neural network for user historical data, the following two points are mainly considered:,1) the temporal correlation of the electricity consumption behavior of the same user;,2) the spatial correlation between the electricity consumption behaviors of different users.,Thus, the adjacency matrix required for historical data modeling is obtained.

[0128] S4. Obtain and record the evaluation indicators of the preliminary identification model for abnormal electricity consumption behaviors of multiple types of users, and determine whether the accuracy of the preliminary identification model for abnormal electricity consumption behaviors of multiple types of users meets the preset requirements based on the evaluation indicators. If so, output the model as a model for identifying abnormal electricity consumption behaviors of multiple types of users; if not, use Bayesian optimization (BO) of graph convolutional neural network hyperparameters to obtain a multi-type user electricity consumption behavior abnormality identification model based on graph convolutional neural network with better performance, so as to realize rapid identification of abnormal electricity consumption behaviors of different types of users.

[0129] The Bayesian optimization algorithm seeks the best through iterative calculation of the acquisition function and the probability proxy model. Based on the prior knowledge and evaluation results, it calculates the optimal hyperparameter combination in a short time to improve the model performance. The specific operation process is as follows: Figure 2 As shown. Each time the graph convolutional neural network performs feature extraction and user behavior recognition, a hyperparameter optimization interval is set. Through Bayesian hyperparameter optimization, a hyperparameter combination is provided for the graph convolutional neural network at each iteration. The optimal hyperparameter combination suitable for the model problem can be selected according to the model evaluation index. The principle is expressed by the following formula:

[0130] In the formula, is the evaluation index value under the optimal hyperparameter combination; is the optimal hyperparameter combination; is the model evaluation indicator; Search space for hyperparameter combinations.

[0131] In practical applications, the hyperparameters of Bayesian optimization mainly include: the number of hidden layer neurons and the learning rate of the hidden layer of the graph convolutional neural network ,but ={optimal number of hidden layer neurons, optimal learning rate}.

[0132] S5. Input the optimal feature set into the abnormal power consumption behavior recognition model of multiple types of users to identify the abnormal power consumption behavior.

[0133] When identifying abnormal electricity consumption behavior, accuracy, precision, recall, F1 score, and area under the curve are usually used to judge the performance of the recognition model. For multi-type user recognition methods, each class is considered a positive class and other classes are considered negative classes. The classification performance indicators used to evaluate the results include: accuracy (Acc), precision (Precision), recall (Recall), F1-score (F1) and area under the curve (AUC). Finally, the average of these indicators is calculated as the final multi-classification indicator value.

[0134] Accuracy refers to the proportion of correctly classified samples to the total number of samples. The calculation formula is as follows:

[0135] In the formula, TP is the correctly classified user; FN refers to the misclassification of the existing category user into another category; FP is the user of another category is wrongly classified into the current category; TN is correctly classified as other category users.

[0136] The precision rate is the ratio of the number of samples predicted as positive to the number of samples predicted as positive. The closer the value is to 1, the lower the misjudgment rate. The calculation formula is as follows:

[0137] The recall rate indicates the proportion of positive samples that are predicted to be correct among the total positive samples. The closer the value is to 1, the lower the missed rate is. The calculation formula is as follows:

[0138] The F1 score is the harmonic mean of precision and recall, which takes into account the accuracy and coverage of the model on the positive category. The calculation formula is as follows:

[0139] The area under the curve (AUC) is a commonly used indicator for evaluating the performance of classification models, and is usually used in receiver operating characteristic (ROC) curve analysis. AUC represents the area under the ROC curve, and its value range is between 0 and 1. The ROC curve shows the relationship between the true positive rate and the false positive rate when the classifier threshold changes. The calculation formula is as follows:

[0140]

[0141] After verification of the test set, when the model performance indicators reach the predetermined standards, for example, when the recall rate and precision both reach above 90%, it means that the model meets the application requirements and can perform user behavior analysis on real-time data.

[0142] The present invention provides a method for identifying abnormal electricity consumption behaviors of multiple types of users. The model constructed by this method solves the problem that the existing identification methods face the risk of low recognition accuracy due to the high differences between the electricity consumption behaviors of different types of users. Models can be established separately according to the spatiotemporal correlations of the electricity consumption behaviors of different users, thereby achieving high-accuracy identification of users with abnormal electricity consumption behaviors, providing a scientific basis and feasible solutions for the safe operation and abnormal warning of the distribution network.

[0143] The present application discloses a system for identifying abnormal electricity consumption behavior of multiple types of users, including: An acquisition unit is used to perform statistical analysis on actual data of user electricity consumption according to the differences in electricity consumption behaviors of different types of users and the temporal and spatial correlation of electricity consumption behaviors of users of the same type, so as to obtain characteristic indicators characterizing the electricity consumption behaviors of multiple types of users; The reduction unit is used to reduce the feature index using the extreme gradient boosting tree to form the optimal feature set; A model building unit is used to establish an adjacency matrix based on the optimal feature set and the spatiotemporal correlation of the optimal features, input the optimal feature set and the adjacency matrix into the graph convolutional neural network, and build a preliminary recognition model for abnormal power consumption behaviors of multiple types of users based on the graph convolutional neural network; A model optimization unit is used to obtain and record evaluation indicators of the preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users, and judge whether the accuracy of the preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users meets preset requirements according to the evaluation indicators. If the requirements are met, the model is output as the abnormal electricity consumption behavior recognition model for multiple types of users; if not, the graph convolutional neural network parameters in the preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users are optimized by Bayesian optimization to obtain a hyperparameter combination, and the hyperparameter combination is used as input, and the optimization is returned to the model construction unit to obtain the abnormal electricity consumption behavior recognition model for multiple types of users; The recognition unit is used to input the optimal feature set into the abnormal power consumption behavior recognition model of multiple types of users to identify the abnormal power consumption behavior.

[0144] The present application discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for identifying abnormal electricity consumption behaviors of multiple types of users described in any one of the above items are implemented.

[0145] The present application discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for identifying abnormal electricity consumption behaviors of multiple types of users described in any one of the above are implemented.

[0146] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0147] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks 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 flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0148] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing 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.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying abnormal electricity consumption behavior of multiple types of users, characterized in that: include: S1: Based on the differences in electricity consumption behaviors of different types of users and the temporal and spatial correlation of electricity consumption behaviors of users of the same type, the actual data of users' electricity consumption are statistically analyzed to obtain characteristic indicators that characterize the electricity consumption behaviors of multiple types of users; S2: Use the extreme gradient boosting tree to reduce the feature index and form the optimal feature set; S3: Based on the optimal feature set, the adjacency matrix is ​​established by using the spatiotemporal correlation of the optimal features. The optimal feature set and the adjacency matrix are input into the graph convolutional neural network to build a preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users based on the graph convolutional neural network. S4: obtaining and recording evaluation indicators of the preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users, and judging whether the accuracy of the preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users meets preset requirements according to the evaluation indicators; if the accuracy meets the requirements, outputting the model as the abnormal electricity consumption behavior recognition model for multiple types of users; If not, the graph convolutional neural network parameters in the multi-type user electricity consumption behavior abnormal preliminary identification model are optimized by Bayesian to obtain a hyperparameter combination, and the hyperparameter combination is used as input to repeat S3 to S4 to obtain a multi-type user electricity consumption behavior abnormal identification model; S5: Input the optimal feature set into the abnormal power consumption behavior recognition model of multiple types of users to identify the abnormal power consumption behavior.

2. According to claim 1, a method for identifying abnormal electricity consumption behavior of multiple types of users is characterized by: In S1, the multiple types of users include single-phase users and three-phase users; The characteristic indicators of the single-phase user's electricity consumption behavior include: the user's daily frozen electricity, daily average current, daily average voltage, zero-live current difference, voltage imbalance and current imbalance; The characteristic indicators of the three-phase user's electricity consumption behavior include: the user's daily frozen electricity, daily average three-phase voltage, daily average three-phase current, zero-live three-phase current difference, current imbalance and voltage imbalance; The characteristic indicators characterizing the electricity consumption behavior of multiple types of users include: daily frozen electricity consumption; daily average three-phase current includes: average current of phase A, average current of phase B, average current of phase C; daily average three-phase voltage includes: average voltage of phase A, average voltage of phase B, average voltage of phase C; zero-live current difference, voltage imbalance and current imbalance.

3. According to claim 1, a method for identifying abnormal electricity consumption behavior of multiple types of users is characterized by: In S2, the feature indexes are reduced using the extreme gradient boosting tree, the importance scores of different feature indexes are calculated, the importance scores are sorted from large to small, and the feature indexes whose importance scores are less than a preset value are reduced.

4. According to claim 1, a method for identifying abnormal electricity consumption behavior of multiple types of users is characterized by: In S3, the adjacency matrix is ​​constructed by taking the optimal feature set as a node and connecting it with the K nearest neighbor nodes in the space through undirected edges to generate an adaptive adjacency matrix , the adjacency matrix calculation formula is as follows: in, is the adjacency matrix At the node With Node The element values ​​between For Node The set of K nearest neighbors determined by distance, For Node The set of K nearest neighbors determined by distance, is a parameter used to adjust the effect of distance, selected through experience or cross-validation; The Euclidean distance between feature vectors is used to represent the similarity between nodes; For the The node eigenvalues; For the The node eigenvalues; is the total number of characteristic indicators.

5. A method for identifying abnormal electricity consumption behavior of multiple types of users according to claim 4, characterized in that: The optimal feature set and the adjacency matrix are input into the graph convolutional neural network to construct a preliminary recognition model for abnormal electricity consumption behaviors of multiple types of users based on the graph convolutional neural network, specifically: In the formula, Indicates +1 layer of nodes The features of the optimal feature set of input for ; Representation Node The set of adjacent nodes of ; is the activation function; is the normalization factor, for nodes The degree and node The product of the degrees of is used to normalize the contribution of adjacent nodes; is the adjacency matrix At the node With Node The element values ​​between Indicates The weight matrix of the layer has self-learning function; Indicates Layer Node characteristics.

6. A method for identifying abnormal electricity consumption behaviors of multiple types of users according to claim 1, characterized in that: The Bayesian optimization is used to optimize the graph convolutional neural network parameters in the multi-type user power consumption behavior abnormal preliminary identification model to obtain a hyperparameter combination, and its principle is expressed by the following formula: In the formula, is the evaluation index value under the optimal hyperparameter combination; is the model evaluation indicator; Search space for hyperparameter combinations; Indicates the maximum model evaluation index value; is a combination of hyperparameters; is the optimal hyperparameter combination.

7. The method for identifying abnormal electricity consumption behavior of multiple types of users according to claim 1, characterized in that: The evaluation indicators include: accuracy, precision, recall, F1 score and area under the curve.

8. A system for identifying abnormal electricity consumption behavior of multiple types of users, characterized by: include: An acquisition unit is used to perform statistical analysis on actual data of user electricity consumption according to the differences in electricity consumption behaviors of different types of users and the temporal and spatial correlation of electricity consumption behaviors of users of the same type, so as to obtain characteristic indicators characterizing the electricity consumption behaviors of multiple types of users; The reduction unit is used to reduce the feature index using the extreme gradient boosting tree to form the optimal feature set; A model building unit is used to establish an adjacency matrix based on the optimal feature set and the spatiotemporal correlation of the optimal features, input the optimal feature set and the adjacency matrix into the graph convolutional neural network, and build a preliminary recognition model for abnormal power consumption behaviors of multiple types of users based on the graph convolutional neural network; A model optimization unit is used to obtain and record evaluation indicators of the preliminary recognition model of abnormal electricity consumption behaviors of multiple types of users, and judge whether the accuracy of the preliminary recognition model of abnormal electricity consumption behaviors of multiple types of users meets preset requirements according to the evaluation indicators. If the requirements are met, the model is output as the abnormal electricity consumption behavior recognition model of multiple types of users; If not, the graph convolutional neural network parameters in the multi-type user electricity consumption behavior abnormal preliminary identification model are optimized by Bayesian to obtain a hyperparameter combination, and the hyperparameter combination is used as input to return to the model construction unit for optimization to obtain a multi-type user electricity consumption behavior abnormal identification model; The recognition unit is used to input the optimal feature set into the abnormal power consumption behavior recognition model of multiple types of users to identify the abnormal power consumption behavior.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for identifying abnormal electricity consumption behaviors of multiple types of users as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the method for identifying abnormal electricity consumption behaviors of multiple types of users according to any one of claims 1 to 7.

Images