Low-voltage user electricity stealing identification method and system

Abstract

The invention discloses a low-voltage user electricity larceny identification system, which comprises the following steps: S1, collecting electricity utilization data of a low-voltage user, and extracting a feature vector from the electricity utilization data; s2, using a relevance vector machine (RVM) to carry out feature selection on the power utilization features of the power consumers to obtain an optimal feature set; s3, performing clustering analysis on the optimal feature set by using an improved K-means clustering algorithm based on information gain, automatically identifying a clustering number and an initial clustering center, and performing clustering iteration to realize abnormal power consumption detection; and S4, identifying abnormal electricity consumption behaviors according to a clustering result, and generating a report containing abnormal electricity consumption behavior information and an electricity stealing suspected user list. The invention further discloses a corresponding system. According to the invention, clustering analysis is carried out on the optimal feature set through the improved K-means clustering algorithm based on information gain, abnormal electricity consumption detection is realized, clustering performance can be improved, and electricity stealing identification efficiency and accuracy are improved.

Description

Technical Field

[0001] The invention relates to the field of electricity fee management, and in particular to a method and system for identifying electricity theft by low-voltage users. Background Art

[0002] Electricity theft, where users illegally obtain electricity without paying for it, has caused significant economic losses to power companies and society. Traditional electricity theft detection mainly relies on manual meter reading and on-site inspection, but this method is inefficient and prone to omissions. With the widespread application of smart meters and the advancement of big data analysis technology, electricity theft detection methods based on electricity data mining have gradually emerged. These methods analyze users' electricity usage patterns, construct electricity usage behavior features, and use statistical learning and machine learning algorithms to identify abnormal electricity usage behaviors to achieve automatic detection of electricity theft.

[0003] In practical applications, electricity theft detection often combines a variety of feature extraction methods, such as statistical analysis, time series analysis, cluster analysis, and machine learning, to build a comprehensive user profile and improve the accuracy and interpretability of detection. However, electricity theft patterns are diverse and time-varying, and models need to be continuously updated and optimized to adapt to the ever-changing means of electricity theft. In addition, electricity theft detection not only relies on data analysis technology, but also needs to cooperate with on-site inspections and punishment measures to form a complete anti-electricity theft system.

[0004] Although neural networks have achieved good results in electricity theft detection, they still have limitations such as poor interpretability, strong data dependence, high computational complexity, difficulty in model selection, and neglect of domain knowledge. The "black box" nature of neural networks makes its decision-making process difficult to explain, limiting its persuasiveness in practical applications. At the same time, the difficulty in obtaining electricity theft samples may lead to insufficient model generalization ability, and the distribution of electricity theft data may drift over time, requiring continuous model updates. In addition, the complex structure and large number of parameters of neural networks lead to high computational overhead, and real-time detection of large user groups may face resource bottlenecks. In terms of model selection, many architectures and hyperparameters need to be adjusted, which increases the difficulty of model design. Finally, neural networks that rely solely on data-driven may not be able to fully utilize expert knowledge and physical laws, affecting detection performance. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method and system for identifying electricity theft by low-voltage users, which adopts an improved K-means clustering algorithm based on information gain to perform cluster analysis on the optimal feature set to realize abnormal electricity consumption detection, and can automatically determine the number of clusters and the initial cluster center, thereby improving the clustering performance and improving the efficiency and accuracy of electricity theft identification.

[0006] In order to solve the above technical problems, as a first aspect of the present invention, a method for identifying electricity theft by low-voltage users is provided, which comprises the following steps:

[0007] Step S1, collecting electricity consumption data of low-voltage users and extracting feature vectors from the electricity consumption data;

[0008] Step S2, using a relevance vector machine (RVM) to perform feature selection on the power consumption characteristics of the power user to obtain an optimal feature set;

[0009] Step S3, using an improved K-means clustering algorithm based on information gain to perform cluster analysis on the optimal feature set, automatically identifying the number of clusters and the initial cluster center, and performing cluster iteration to achieve abnormal power consumption detection;

[0010] Step S4, identifying abnormal electricity usage behavior based on the clustering results, and generating a report including abnormal electricity usage behavior information and a list of suspected electricity theft users.

[0011] Optionally, the step of performing feature selection using a relevance vector machine (RVM) further comprises:

[0012] Construct the feature vector space and define the probability model of weight vector and noise component;

[0013] The model parameters are learned by maximizing the marginal likelihood estimate, achieving joint optimization of feature selection and classifier design.

[0014] Optionally, step S3 further comprises:

[0015] Step S30, determining the optimal number of clusters K based on information gain;

[0016] Step S31, selecting K samples with the largest spatial density as initial cluster centers;

[0017] Step S32, dividing the non-cluster center features into the nearest cluster center according to the Euclidean distance;

[0018] Step S33, calculating a new cluster center;

[0019] Step S34 repeats the division and calculation steps until the algorithm converges.

[0020] Optionally, the step S30 further includes:

[0021] Step S301, calculating the information entropy of the feature space; for the feature space Its information entropy is defined as:

[0022]

[0023] in, ‖·‖ l1 It means to find the l1 norm of the variables;

[0024] The information entropy of the kth subclass after clustering is:

[0025]

[0026] Among them, C k is the feature number set corresponding to the k-th subclass;

[0027] Step S302, gradually increase the number of clusters k, and calculate the information gain at each k value using the following formula:

[0028]

[0029] Step S303: Select k corresponding to the maximum information gain value as the optimal number of clusters K.

[0030] Optionally, the step S31 further includes:

[0031] Step S310, calculate each sample x as follows n The spatial density dist(x n ):

[0032]

[0033] in,‖‖ l2 It means to find the l2 norm of the variable;

[0034] Step S311, sorting by spatial density from large to small, and selecting the first K samples as initial cluster centers.

[0035] Accordingly, as another aspect of the present invention, a low-voltage user electricity theft identification system is also provided, which comprises:

[0036] A data collection and extraction unit is used to collect power consumption data of low-voltage users and extract feature vectors from the power consumption data;

[0037] A feature selection module is used to select the power consumption characteristics of power users by using a relevance vector machine (RVM) to obtain an optimal feature set;

[0038] The cluster analysis module is used to perform cluster analysis on the optimal feature set using the improved K-means clustering algorithm based on information gain, automatically identify the number of clusters and the initial cluster center, and perform cluster iteration to achieve abnormal power consumption detection;

[0039] The electricity theft identification module is used to identify abnormal electricity usage behavior based on the clustering results and generate a report containing abnormal electricity usage behavior information and a list of suspected electricity theft users.

[0040] Optionally, the feature selection module further comprises:

[0041] A feature vector space construction unit, used for constructing a feature vector space;

[0042] A probability model definition unit, used to define the probability model of the weight vector and the noise component;

[0043] The joint optimization unit is used to learn model parameters by maximizing marginal likelihood estimation to achieve joint optimization of feature selection and classifier design.

[0044] Optionally, the cluster analysis module further includes:

[0045] A cluster number determination module, used to determine the optimal cluster number K based on information gain;

[0046] The initial cluster center selection module is used to select K samples with the largest spatial density as the initial cluster centers;

[0047] A partitioning unit is used to partition non-cluster center features to the nearest cluster center according to the Euclidean distance;

[0048] A center calculation unit, used to calculate new cluster centers;

[0049] The convergence judgment unit is used to judge whether the algorithm has converged. If not, the division and calculation process is repeated until the algorithm converges.

[0050] Optionally, the cluster number determination module further includes:

[0051] An information entropy calculation unit, used to calculate the information entropy of the feature space;

[0052] An information gain calculation unit is used to gradually increase the number of clusters k and calculate the information gain under each k value;

[0053] The optimal K value determination unit is used to select the k corresponding to the maximum information gain value as the optimal cluster number K.

[0054] Optionally, the initial cluster center selection module further includes:

[0055] A spatial density calculation unit, used to calculate the spatial density of each sample;

[0056] The initial center selection unit is used to sort the samples from large to small according to the spatial density and select the first K samples as the initial clustering centers.

[0057] The implementation of the embodiments of the present invention has the following beneficial effects:

[0058] The present invention proposes a method and system for identifying electricity theft by low-voltage users, which adopts an improved K-means clustering algorithm based on information gain for identifying electricity theft by low-voltage users. Through the RVM feature selection method, the present invention realizes the joint optimization of feature selection and classifier design, determines the optimal feature set from the feature vector, and improves the classification performance. At the same time, the present invention proposes a method for determining the optimal number of clusters based on information gain, and selects the initial cluster center in combination with the sample space density, which solves the problem of subjectivity and poor adaptability of the setting of the number of clusters and the initial cluster center in the traditional K-means clustering algorithm.

[0059] The implementation of the present invention can automatically determine the number of clusters and the initial cluster center, improve clustering performance, and thus more accurately identify abnormal electricity consumption behavior. In addition, the present invention combines the feature selection capability of RVM, improves the effectiveness and sparsity of features, and reduces the interference of redundant information on detection performance. Overall, the present invention provides a more intelligent and precise solution for the identification of electricity theft by low-voltage users, which helps power companies to more effectively prevent and combat electricity theft and ensure the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying creative labor, other drawings obtained based on these drawings still belong to the scope of the present invention.

[0061] Figure 1 A schematic diagram of the main process of a method for identifying electricity theft by low-voltage users provided by the present invention;

[0062] Figure 2 A schematic diagram of the structure of a low-voltage user electricity theft identification system provided by the present invention;

[0063] Figure 3 for Figure 2 Schematic diagram of the structure of the feature selection module;

[0064] Figure 4 for Figure 2 The schematic diagram of the cluster analysis module;

[0065] Figure 5 for Figure 4 The schematic diagram of the structure of the module for determining the number of clusters;

[0066] Figure 6 for Figure 4 Schematic diagram of the structure of the initial cluster center selection module in . DETAILED DESCRIPTION

[0067] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings.

[0068] In order to understand the technical solution of the present invention, the technical terms involved in the present invention are first introduced as follows:

[0069] Electricity theft: Electricity theft refers to the illegal acquisition and use of electricity resources without settlement and payment in accordance with regulations. This behavior has seriously affected the normal power supply and operation order, and has caused a series of negative impacts on the power industry and social economy. Electricity theft includes various forms: illegal wiring of public power facilities, private wiring, tampering with electricity meters, collusion between insiders and outsiders to steal electricity, etc.

[0070] Low-voltage users: Low-voltage users refer to users with lower voltage levels in the power system. In the power system, electricity is transmitted and transformed from power plants through substations before being delivered to users. According to the different electricity demand and electricity usage locations, the power system divides users into different voltage levels, among which low-voltage users refer to users with lower voltage. The voltage level of low-voltage users is usually 220V or 230V (depending on the electricity standards of the region and country), which is a common voltage level for homes, commercial buildings, small industries, and small offices. Compared with high-voltage users, low-voltage users have a smaller power load and require relatively lower power capacity.

[0071] Anomaly detection: Anomaly detection of electricity consumption is to use data analysis and machine learning technologies to analyze the electricity consumption behavior of users and promptly detect abnormal electricity consumption, such as sudden changes in electricity consumption, electricity consumption violations, electricity theft, etc. It is of great significance in the operation and management of power systems. Electricity users include multiple categories such as industry, commerce, agriculture, and residents, and the electricity consumption behaviors of different users show diverse characteristics. Some users may have abnormal behaviors such as electricity consumption violations and electricity theft, which will cause economic losses to power companies and affect the safe operation of the power grid.

[0072] Feature extraction: Extraction of electricity theft features is a key step in detecting abnormal electricity use. It analyzes the user's electricity usage data to identify features that may be related to electricity theft. Based on the extracted features, it is determined whether a user is suspected of electricity theft, providing a basis for further investigation. Quantitative assessment of the electricity theft risks of different users helps to reasonably allocate audit resources. By analyzing the characteristics of electricity theft, the laws and means of electricity theft can be summarized to provide support for combating electricity theft.

[0073] K-means: K-means is a commonly used unsupervised machine learning algorithm that is used to divide a data set into K clusters. Its basic idea is to assign data points to the nearest cluster center through continuous iteration and update the position of the cluster center until the optimal clustering result is achieved.

[0074] RVM: RVM is the abbreviation of Relevance Vector Machine, which is a machine learning algorithm for classification and regression tasks. RVM is a sparse kernel method based on the Bayesian framework. Unlike SVM, RVM introduces a priori probability distribution and learns model parameters by maximizing marginal likelihood estimates. RVM has a wide range of applications in many fields, such as: Text classification: classifying documents into predefined categories based on text content. Image recognition: detecting and identifying targets based on image features. Bioinformatics: used for tasks such as gene expression data analysis and protein structure prediction. Time series prediction: predicting future trends and changes based on historical data.

[0075] like Figure 1 As shown, a schematic diagram of the main process of a method for identifying electricity theft by a low-voltage user provided by the present invention is shown. In this embodiment, the method at least includes the following steps:

[0076] Step S1, collecting electricity consumption data of low-voltage users and extracting feature vectors from the electricity consumption data;

[0077] Step S2, using a relevance vector machine (RVM) to perform feature selection on the power consumption characteristics of the power user to obtain an optimal feature set;

[0078] In a specific example, the step S2 further includes:

[0079] Construct the feature vector space and define the probability model of weight vector and noise component;

[0080] The model parameters are learned by maximizing the marginal likelihood estimate, achieving joint optimization of feature selection and classifier design.

[0081] It is understandable that according to the traditional anomaly detection model, after feature extraction, classifier design is required to achieve classification judgment of different electricity consumption behaviors. However, the current feature extraction process and classifier design process are independent of each other. The feature extraction process does not consider whether the proposed features are optimal for the classifier, and the subjectively extracted features usually contain a lot of redundant information. This information is not only unhelpful for abnormal electricity detection, but will mislead classifier training. Therefore, it is necessary to perform feature selection and determine the optimal feature set from the feature vector to improve classification performance. RVM is a probabilistic model based on the Bayesian framework developed on the basis of SVM. Compared with SVM, it has stronger sparsity and can achieve joint optimization of feature selection and classifier design. The model for feature selection using RVM can be expressed as:

[0082]

[0083] Among them, f m is the characteristic vector corresponding to the mth power user, K(f,f m ) is the kernel function, w=[w1,w2,...,w m ] T is the weight vector, ε is the noise part in the data. In order to build a complete Bayesian model, probability models are designed for the weight vector and noise component respectively. The reasonable probability model is Gaussian distribution, that is, w obeys the mean of zero, and the covariance matrix is ​​α -1 I has a Gaussian distribution, ε has a mean of zero, and the covariance matrix is ​​γ -1 Gaussian distribution of I. The noise part in the data, usually assumed to be Gaussian noise, represents the difference between the actual observation and the model prediction. α m is the weight vector ω m The related precision parameter, used in the Bayesian framework, represents ω m γ is the accuracy parameter related to the noise ε and is used in the Bayesian framework to represent the uncertainty of the noise.

[0084] Step S3, using an improved K-means clustering algorithm based on information gain to perform cluster analysis on the optimal feature set, automatically identifying the number of clusters and the initial cluster center, and performing cluster iteration to achieve abnormal power consumption detection;

[0085] Step S4, identifying abnormal electricity usage behavior based on the clustering results, and generating a report including abnormal electricity usage behavior information and a list of suspected electricity theft users.

[0086] In a specific example, the anomaly detection model built by the present invention is a clustering method based on K-means clustering. However, the clustering performance of traditional K-means is closely related to the number of clusters K and the setting of the initial cluster center. The currently commonly used methods based on expert knowledge base or prior information have the problem of strong subjectivity and poor adaptability. In order to solve this problem, this paper proposes a method for determining the optimal number of clusters based on information gain. At the same time, considering that K-means is a method that uses Euclidean distance as the clustering division criterion, K samples with the largest spatial density are selected as the initial cluster center, and the number of clusters and the initial cluster center of K-means clustering are automatically determined by information gain and sample spatial density, thereby improving clustering performance.

[0087] Therefore, the step S3 further includes:

[0088] Step S30, determining the optimal number of clusters K based on information gain;

[0089] In a specific example, the step S30 further includes:

[0090] Step S301, calculating the information entropy of the feature space; for the feature space Its information entropy is defined as:

[0091]

[0092] Among them, E X is the information entropy of the feature space X, which is used to measure the uncertainty or randomness of the data; For sample x n The distribution probability in the feature space is used to calculate the information entropy. In one example, ‖·‖ l1 It means to find the l1 norm of the variable, ‖·‖ l2 It means finding the l2 norm of the variable; N is the number of electricity users or the number of samples in the feature space;

[0093] The information entropy of the kth subclass after clustering is:

[0094]

[0095] Among them, C k is the feature number set corresponding to the k-th subclass;

[0096] Step S302, gradually increase the number of clusters k, and calculate the information gain I at each k value by the following formula: K :

[0097]

[0098] Step S303: Select k corresponding to the maximum information gain value as the optimal number of clusters K.

[0099] Step S31, selecting K samples with the largest spatial density as initial cluster centers;

[0100] In a specific example, the step S31 further includes:

[0101] Step S310, calculate each sample x as follows n The spatial density dist(x n ):

[0102]

[0103] Among them, the spatial density dist(x n ) is used to determine the initial cluster center;

[0104] Step S311, sorting by spatial density from large to small, and selecting the first K samples as initial cluster centers.

[0105] Step S32, dividing the non-cluster center features into the nearest cluster center according to the Euclidean distance;

[0106] Step S33, calculating a new cluster center;

[0107] Step S34 repeats the division and calculation steps until the algorithm converges.

[0108] Therefore, in the specific example, based on the above definition, the specific algorithm flow of the proposed improved K-means clustering can be summarized as follows:

[0109] Input: Optimal feature vector space after RVM feature selection N is the number of electricity users; the maximum number of clusters is K max .

[0110] Output: There are also clustering results and anomaly detection results.

[0111] Algorithm flow:

[0112] Set the number of clusters k = 1,

[0113] Step 1: First, calculate the feature vector space based on the information entropy of the improved K-means Information entropy E X ;

[0114] Step 2: Calculate the information entropy E of the kth subclass after clustering k ;

[0115] Step 3: Calculate the information gain I at this time k ;

[0116] Step 4: Let k = k + 1, and repeat steps 1 to 3 until k = K max ;

[0117] Step 5: Select k corresponding to the maximum value of all information gains recorded in step 3 as the optimal number of clusters K;

[0118] Step 6: Select K features with the largest spatial density as the initial clustering centers.

[0119] Step 7: Divide the non-cluster center features in the feature space into the cluster closest to them according to the Euclidean distance;

[0120] Step 8: Calculate the new cluster center according to the following formula

[0121]

[0122] Among them, n k is the number of features of the kth subset, which is used to calculate the new cluster center.

[0123] Step 9: Re-divide the feature space according to the new cluster center obtained in step 8. If the results of two adjacent divisions are consistent, the algorithm is considered to have converged, otherwise repeat steps 7 and 8.

[0124] The following is a specific example:

[0125] Assume that we have electricity consumption data of 1,000 low-voltage users, and each user has 10 initial features (such as average daily electricity consumption, peak-to-valley ratio, monthly electricity consumption change rate, etc.).

[0126] Data preprocessing: The electricity consumption data of 1,000 users were cleaned and normalized to obtain a standardized data set.

[0127] Feature selection: Use RVM to perform feature selection on the standardized data set. Assume that 5 features are finally selected as the optimal feature set (such as 3 of the daily average electricity consumption, peak-to-valley ratio, and monthly electricity consumption change rate, as well as 2 additional features screened out by RVM).

[0128] Improved K-means clustering:

[0129] Determine the optimal number of clusters K: By calculating the information gain, it is found that the information gain is the largest when K = 3, so K = 3 is selected as the number of clusters.

[0130] Select initial cluster centers: Calculate the spatial density of each user (only 5 features are considered now) and select the 3 users with the largest density as the initial cluster centers.

[0131] Perform clustering: divide the remaining users into three clusters based on the Euclidean distance, update the cluster centers and repeat the division process until the clustering results are stable.

[0132] Anomaly detection: By analyzing the electricity consumption characteristics of the three clusters, it is found that the electricity consumption behavior of users in one cluster is significantly different from that of the other two clusters (for example, the average daily electricity consumption is far below the average level, but the peak-to-valley ratio is abnormally high). The users in this cluster are marked as suspected electricity theft users.

[0133] Through this method, we can effectively identify electricity theft among low-voltage users and provide strong support for power companies' electricity fee management and anti-electricity theft work.

[0134] like Figure 2 FIG. 1 is a schematic diagram showing a structure of an embodiment of a low-voltage user electricity theft identification system provided by the present invention. Figures 3 to 6 As shown, in this embodiment, the system 1 at least includes:

[0135] The data collection and extraction unit 10 is used to collect the power consumption data of low-voltage users and extract feature vectors from the power consumption data;

[0136] The feature selection module 11 is used to select the power consumption characteristics of the power users by using a relevance vector machine (RVM) to obtain an optimal feature set;

[0137] The cluster analysis module 12 is used to perform cluster analysis on the optimal feature set using an improved K-means clustering algorithm based on information gain, automatically identify the number of clusters and the initial cluster center, and perform cluster iteration to achieve abnormal power consumption detection;

[0138] The electricity theft identification module 13 is used to identify abnormal electricity usage behavior according to the clustering results, and generate a report containing abnormal electricity usage behavior information and a list of suspected electricity theft users.

[0139] like Figure 3 As shown, in a specific example, the feature selection module 11 further includes:

[0140] A feature vector space construction unit 110, used to construct a feature vector space;

[0141] A probability model definition unit 111, used to define a probability model of a weight vector and a noise component;

[0142] The joint optimization unit 112 is used to learn the model parameters by maximizing the marginal likelihood estimation, so as to achieve the joint optimization of feature selection and classifier design.

[0143] like Figure 4 As shown, in a specific example, the cluster analysis module 12 further includes:

[0144] A cluster number determination module 120, configured to determine the optimal cluster number K based on information gain;

[0145] An initial cluster center selection module 121 is used to select K samples with the largest spatial density as initial cluster centers;

[0146] A division unit 122, used to divide the non-cluster center features into the nearest cluster center according to the Euclidean distance;

[0147] A center calculation unit 123, used for calculating a new cluster center;

[0148] The convergence judgment unit 124 is used to judge whether the algorithm has converged. If not, the division and calculation process is repeated until the algorithm converges.

[0149] like Figure 5 As shown, in a specific example, the cluster number determination module 120 further includes:

[0150] An information entropy calculation unit 1200, used to calculate the information entropy of the feature space;

[0151] An information gain calculation unit 1201 is used to gradually increase the number of clusters k and calculate the information gain at each k value;

[0152] The optimal K value determining unit 1202 is used to select k corresponding to the maximum information gain value as the optimal cluster number K.

[0153] like Figure 6 As shown, in a specific example, the initial cluster center selection module 121 further includes:

[0154] A spatial density calculation unit 1210, configured to calculate the spatial density of each sample;

[0155] The initial center selection unit 1211 is used to sort the samples from large to small according to the spatial density and select the first K samples as the initial clustering centers.

[0156] For more details, please refer to and combine the above Figure 1 The description is not repeated here.

[0157] The implementation of the embodiments of the present invention has the following beneficial effects:

[0158] The present invention proposes a method and system for identifying electricity theft by low-voltage users, which adopts an improved K-means clustering algorithm based on information gain for identifying electricity theft by low-voltage users. Through the RVM feature selection method, the present invention realizes the joint optimization of feature selection and classifier design, determines the optimal feature set from the feature vector, and improves the classification performance. At the same time, the present invention proposes a method for determining the optimal number of clusters based on information gain, and selects the initial cluster center in combination with the sample space density, which solves the problem of subjectivity and poor adaptability of the setting of the number of clusters and the initial cluster center in the traditional K-means clustering algorithm.

[0159] The implementation of the present invention can automatically determine the number of clusters and the initial cluster center, improve clustering performance, and thus more accurately identify abnormal electricity consumption behavior. In addition, the present invention combines the feature selection capability of RVM, improves the effectiveness and sparsity of features, and reduces the interference of redundant information on detection performance. Overall, the present invention provides a more intelligent and precise solution for the identification of electricity theft by low-voltage users, which helps power companies to more effectively prevent and combat electricity theft and ensure the safe and stable operation of the power system.

[0160] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices, 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. Moreover, 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.

[0161] 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.

[0162] The above disclosure is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for identifying electricity theft by low-voltage users, characterized in that: The steps include: Step S1, collecting electricity consumption data of low-voltage users and extracting feature vectors from the electricity consumption data; Step S2, using a relevance vector machine (RVM) to perform feature selection on the power consumption characteristics of the power user to obtain an optimal feature set; Step S3, using an improved K-means clustering algorithm based on information gain to perform cluster analysis on the optimal feature set, automatically identifying the number of clusters and the initial cluster center, and performing cluster iteration to achieve abnormal power consumption detection; Step S4, identifying abnormal electricity usage behavior based on the clustering results, and generating a report including abnormal electricity usage behavior information and a list of suspected electricity theft users.

2. The method for identifying electricity theft by low-voltage users according to claim 1, characterized in that: The step S2 further comprises: Construct the feature vector space and define the probability model of weight vector and noise component; The model parameters are learned by maximizing the marginal likelihood estimate, achieving joint optimization of feature selection and classifier design.

3. The method for identifying electricity theft by low-voltage users according to claim 2, characterized in that: The step S3 further comprises: Step S30, determining the optimal number of clusters K based on information gain; Step S31, selecting K samples with the largest spatial density as initial cluster centers; Step S32, dividing the non-cluster center features into the nearest cluster center according to the Euclidean distance; Step S33, calculating a new cluster center; Step S34 repeats the division and calculation steps until the algorithm converges.

4. The method for identifying electricity theft by low-voltage users according to claim 3 is characterized in that: The step S30 further comprises: Step S301, calculating the information entropy of the feature space; for the feature space Its information entropy is defined as: in, ‖·‖ l1 It means to find the l1 norm of the variable, ‖·‖ l2 It means to find the l2 norm of the variables; The information entropy of the kth subclass after clustering is: Among them, C k is the feature number set corresponding to the k-th subclass; Step S302, gradually increase the number of clusters k, and calculate the information gain I at each k value by the following formula: K : Step S303: Select k corresponding to the maximum information gain value as the optimal number of clusters K.

5. The method for identifying electricity theft by low-voltage users according to claim 3 is characterized in that: The step S31 further comprises: Step S310, calculate each sample x as follows n The spatial density dist(x n ): in,‖·‖ l2 It means to find the l2 norm of the variables; Step S311, sorting by spatial density from large to small, and selecting the first K samples as initial cluster centers.

6. A low-voltage user electricity theft identification system, characterized in that: include: A data collection and extraction unit is used to collect power consumption data of low-voltage users and extract feature vectors from the power consumption data; A feature selection module is used to select the power consumption characteristics of power users by using a relevance vector machine (RVM) to obtain an optimal feature set; The cluster analysis module is used to perform cluster analysis on the optimal feature set using the improved K-means clustering algorithm based on information gain, automatically identify the number of clusters and the initial cluster center, and perform cluster iteration to achieve abnormal power consumption detection; The electricity theft identification module is used to identify abnormal electricity usage behavior based on the clustering results and generate a report containing abnormal electricity usage behavior information and a list of suspected electricity theft users.

7. The low-voltage user electricity theft identification system according to claim 6 is characterized in that: The feature selection module further comprises: A feature vector space construction unit, used for constructing a feature vector space; A probability model definition unit, used to define the probability model of the weight vector and the noise component; The joint optimization unit is used to learn model parameters by maximizing marginal likelihood estimation to achieve joint optimization of feature selection and classifier design.

8. The low-voltage user electricity theft identification system according to claim 6 is characterized in that: The cluster analysis module further comprises: A cluster number determination module, used to determine the optimal cluster number K based on information gain; The initial cluster center selection module is used to select K samples with the largest spatial density as the initial cluster centers; A partitioning unit is used to partition non-cluster center features to the nearest cluster center according to the Euclidean distance; A center calculation unit, used to calculate new cluster centers; The convergence judgment unit is used to judge whether the algorithm has converged. If not, the division and calculation process is repeated until the algorithm converges.

9. The low-voltage user electricity theft identification system according to claim 8 is characterized in that: The cluster number determination module further includes: An information entropy calculation unit, used to calculate the information entropy of the feature space; An information gain calculation unit is used to gradually increase the number of clusters k and calculate the information gain under each k value; The optimal K value determination unit is used to select the k corresponding to the maximum information gain value as the optimal cluster number K.

10. The low-voltage user electricity theft identification system according to claim 8, characterized in that: The initial cluster center selection module further includes: A spatial density calculation unit, used to calculate the spatial density of each sample; The initial center selection unit is used to sort the samples from large to small according to the spatial density and select the first K samples as the initial clustering centers.

Images