CN-121997217-A - Abnormal electricity behavior identification method and device, readable storage medium and terminal equipment

CN121997217ACN 121997217 ACN121997217 ACN 121997217ACN-121997217-A

Abstract

The application provides a method and a device for identifying abnormal electricity consumption behaviors, a readable storage medium and terminal equipment, and relates to the technical field of power distribution monitoring, wherein the method comprises the steps of obtaining an electric quantity data sequence of each electric energy metering terminal in a specified period; if abnormal data exist in the current electric quantity data sequence, a target reference electric quantity data sequence is determined based on the matching relation between the electric quantity data sequences of a plurality of other electric energy metering terminals and the change trend of the current electric quantity data sequence, the abnormal data of the current electric quantity data sequence are corrected based on the target reference electric quantity data sequence, the corrected electric quantity data are taken as input, and the abnormal electricity consumption behavior recognition result of the current electric energy metering terminal is output through a preset abnormal electricity consumption behavior recognition model. The method and the device can effectively identify the users with abnormal electricity consumption, and remarkably improve the accuracy and reliability of abnormal electricity consumption detection of the users.

Inventors

SHEN ZHENDONG
ZHANG YU
YU KEXIN
HUO CHAO
BAI HUIFENG
LING YONGGANG
SHI SHUO
ZHANG GANGHONG
Yang Qinbin
TIAN YANG
LIU HAO

Assignees

北京智芯微电子科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251212

Claims (11)

1. An abnormal electricity consumption behavior identification method is characterized by comprising the following steps: acquiring an electric quantity data sequence of each electric energy metering terminal in a specified period; If abnormal data exist in the current electric quantity data sequence, determining a target reference electric quantity data sequence based on a matching relation between the electric quantity data sequences of a plurality of other electric energy metering terminals and the change trend of the current electric quantity data sequence, and correcting the abnormal data of the current electric quantity data sequence based on the target reference electric quantity data sequence; And taking the corrected electric quantity data as input, and outputting an abnormal electricity consumption behavior recognition result of the current electric energy metering terminal through a preset abnormal electricity consumption behavior recognition model, wherein the abnormal electricity consumption behavior recognition model is obtained by training a sequential neural network model and a classifier model through electric quantity data sequences of different electric energy metering terminals under different electricity consumption behaviors.
2. The abnormal electricity consumption behavior recognition method according to claim 1, wherein the time sequence neural network model is an extended long-short-term memory model, the classifier model is a related vector machine model, and the abnormal electricity consumption behavior recognition model comprises: the data input module, the extended long-short-term memory model and the related vector machine model; The data input module is used for inputting the acquired electric quantity data sequence into the extended long-period memory model; The extended long-short-term memory model is used for extracting time sequence characteristics of an input electric quantity data sequence and outputting a time sequence characteristic vector sequence of the electric quantity data sequence to the related vector machine model; The correlation vector machine model is used for mapping the received time sequence feature vector to a Gaussian process space, outputting the abnormal probability that each time point belongs to each abnormal electricity consumption behavior, and determining the abnormal electricity consumption behavior corresponding to the maximum abnormal probability as the abnormal electricity consumption behavior of the corresponding time point.
3. The abnormal electrical behavior recognition method of claim 1, wherein determining that abnormal data exists in the current electrical quantity data sequence comprises: The method comprises the steps that a missing value exists in a current electric quantity data sequence, or the value of any electric quantity data in the current electric quantity data sequence is higher than a first electric quantity data threshold value or lower than a second electric quantity data threshold value, and the first electric quantity data threshold value is higher than the second electric quantity data threshold value; the first power data threshold and the second power data threshold are determined by: Dividing a current electric quantity data sequence into a plurality of electric quantity data subsequences, determining a minimum value and a maximum value in the current electric quantity data subsequences for each electric quantity data subsequence, and determining a maximum electric quantity data difference between the maximum value and the minimum value; And adjusting the maximum electric quantity data difference by a preset adjustment factor to obtain an adjustment value of the maximum electric quantity data difference, taking the sum of the maximum value and the adjustment value as a first electric quantity data threshold value of the current electric quantity data subsequence, taking the difference between the minimum value and the adjustment value as a second electric quantity data threshold value of the current electric quantity data subsequence, and taking the appointed value as the second electric quantity data threshold value of the current electric quantity data subsequence if the difference between the minimum value and the adjustment value is smaller than the appointed value.
4. The abnormal electricity behavior recognition method according to claim 1, wherein determining the target reference electricity data sequence based on a matching relationship between the electricity data sequences of the plurality of other electricity metering terminals and the change trend of the current electricity data sequence includes: Constructing a first reference electric quantity data subsequence by using a plurality of electric quantity data before the first abnormal data in the current electric quantity data sequence, and constructing a second reference electric quantity data subsequence by using a plurality of electric quantity data after the last abnormal data in the current electric quantity data sequence; Constructing a reference electric quantity data sequence based on the first reference electric quantity data subsequence and the second reference electric quantity data subsequence according to the time sequence; acquiring electric quantity data sequences of a plurality of other electric energy metering terminals as reference electric quantity data sequences, wherein the length of each reference electric quantity data sequence is the same as that of the reference electric quantity data sequence; And matching the change trend of each reference electric quantity data sequence with the change trend of the reference electric quantity data sequence to obtain the matching degree value of each reference electric quantity data sequence and the reference electric quantity data sequence, and taking the reference electric quantity data sequence with the largest matching degree value as the target reference current data sequence.
5. The abnormal electricity consumption behavior recognition method according to claim 4, wherein the step of matching the trend of each reference electricity consumption data sequence with the trend of the reference electricity consumption data sequence to obtain a matching degree value of each reference electricity consumption data sequence and the reference electricity consumption data sequence comprises the steps of: For each reference electric quantity data sequence, calculating a local distance between each electric quantity data in the current reference electric quantity data sequence and each electric quantity data in the reference electric quantity data sequence; Taking the reference electric quantity data sequence as a column, taking the current reference electric quantity data sequence as a row, and constructing a distance matrix of local distance between each electric quantity data in the current reference electric quantity data sequence and each electric quantity data in the reference electric quantity data sequence; based on the local distances, converting the distance matrix into an accumulated cost matrix representing accumulated distances between each electric quantity data in a current reference electric quantity data sequence and each electric quantity data in the reference electric quantity data sequence; Determining the adjacent element with the smallest accumulated distance in the adjacent element of the previous row or the previous column of the initial element by taking the accumulated distance corresponding to the last row and the last column in the accumulated cost matrix as an initial element, taking the target adjacent element as a new initial element, repeating the process until the target adjacent element of the current initial element is the element of the first row or the first column of the accumulated cost matrix, and taking the current target adjacent element as an end element to obtain an optimal cost path; and determining the reciprocal of the sum of the accumulation distances corresponding to all elements on the optimal cost path as a matching degree value of the current reference electric quantity data sequence and the reference electric quantity data sequence.
6. The abnormal electrical behavior identification method of claim 5, wherein converting the distance matrix into a cumulative cost matrix representing a cumulative distance between each electrical quantity data in a current reference electrical quantity data sequence and each electrical quantity data in the reference electrical quantity data sequence based on each local distance comprises: Initializing a first row and a first column of the distance matrix to obtain accumulated distances corresponding to all elements in the first row and the first column of the distance matrix; For each other element except the first row and the first column in the distance matrix, determining that the element with the smallest accumulated distance in the adjacent element of the previous row or the previous column of the current element is a target adjacent element of the current element, and taking the sum of the local distance of the current element and the accumulated distance of the target adjacent element as the accumulated distance of the current element; initializing a first row and a first column of the distance matrix, comprising: for each element in the first row of the distance matrix, the accumulated sum of the local distances of the elements before the current element is taken as the accumulated distance of the current element, and for each element in the first column of the distance matrix, the accumulated sum of the local distances of the elements before the current element is taken as the accumulated distance of the current element.
7. The abnormal electricity consumption behavior recognition method according to claim 4, wherein acquiring the electric quantity data series of the plurality of other electric energy metering terminals as the reference electric quantity data series includes: Determining a first time period corresponding to the first reference electric quantity data sub-sequence and a second time period corresponding to the second reference electric quantity data sub-sequence; For each of a plurality of other power metering terminals: Constructing a first reference electric quantity data subsequence based on a plurality of electric quantity data of the current electric energy metering terminal in the first time period, and constructing a second reference electric quantity data subsequence based on a plurality of electric quantity data of the current electric energy metering terminal in the second time period; and constructing a reference electric quantity data sequence based on the first reference electric quantity data subsequence and the second reference electric quantity data subsequence according to the time sequence.
8. The abnormal electricity usage behavior recognition method according to claim 7, wherein correcting abnormal data of a current electricity amount data sequence based on the target reference electricity amount data sequence includes: determining time points corresponding to different data in the current electric quantity data sequence as target time points; and replacing corresponding abnormal data in the current electric quantity data sequence by the electric quantity data of the electric energy metering terminal at each target time point corresponding to the target reference electric quantity data sequence.
9. An abnormal electricity behavior recognition device, characterized by comprising: The data acquisition module is configured to acquire an electric quantity data sequence of each electric energy metering terminal in a specified period; The data correction module is configured to determine a target reference electric quantity data sequence based on a matching relation between the electric quantity data sequences of the plurality of other electric energy metering terminals and the change trend of the current electric quantity data sequence if abnormal data exist in the current electric quantity data sequence, and correct the abnormal data of the current electric quantity data sequence based on the target reference electric quantity data sequence; The power consumption behavior recognition module is configured to take corrected electric quantity data as input, and output an abnormal power consumption behavior recognition result of the current electric energy metering terminal through a preset abnormal power consumption behavior recognition model, wherein the abnormal power consumption behavior recognition model is obtained by training a time sequence neural network model and a classifier model through electric quantity data sequences of different electric energy metering terminals under different power consumption behaviors.
10. A machine-readable storage medium having instructions stored thereon, which when executed by a processor cause the processor to be configured to perform the abnormal electricity usage behavior identification method of any of claims 1-8.
11. Terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the abnormal electricity usage behavior identification method according to any of the claims 1-8 when the computer program is executed.

Description

Abnormal electricity behavior identification method and device, readable storage medium and terminal equipment Technical Field The application relates to the technical field of power distribution monitoring, in particular to an abnormal electricity consumption behavior identification method, an abnormal electricity consumption behavior identification device, a machine-readable storage medium and terminal equipment. Background Along with the continuous expansion of the power grid scale, the abnormal electricity consumption phenomenon is increased, so that not only is the electric power loss caused, but also a huge economic burden is brought to an electric company. Losses in grid operation are largely divided into technical losses (TECHNICAL LOSS, TL) and Non-technical losses (Non-TECHNICAL LOSS, NTL). Technical losses refer to the problems of resistance losses and the like when equipment and current in a power grid pass through, while non-technical losses refer to losses which are caused by illegal power utilization behaviors of power distribution network users and cannot be interpreted through technical means. The traditional NTL detection method mainly relies on means such as manual screening to prevent illegal electricity utilization behaviors such as electricity larceny, however, the method not only consumes a great deal of manpower and material resources, but also has limited effect. With the popularization of smart meters, it becomes feasible to detect anomalies for users by means of smart algorithms. At present, a detection method for NTL based on a data mining technology generally adopts a classical non-supervised learning method and an emerging theory to carry out anomaly detection analysis, such as a similarity deviation theory, a time sequence algorithm, a ridge regression model, a random matrix theory and the like. The method is characterized in that a large amount of electricity consumption data and a data mining method are utilized to design an abnormality detection system based on a big data analysis technology, or electricity consumption characteristics are extracted based on a machine learning and deep learning method, and an electricity consumption abnormality identification model is established. However, the identification method of abnormal electricity at present can realize the identification of abnormal electricity, but does not consider the problem of low upper limit of accuracy caused by gradient explosion and gradient dispersion existing in the deep neural network, and meanwhile, in the process of identifying the abnormal electricity by adopting big data, the inaccuracy of the original data also affects the identification result. Disclosure of Invention An object of an embodiment of the present application is to provide a method for identifying abnormal electricity consumption behavior, an apparatus for identifying abnormal electricity consumption behavior, a machine-readable storage medium, and a terminal device, so as to solve the above-mentioned problems. In order to achieve the above object, a first aspect of the present application provides an abnormal electricity usage behavior recognition method, including: acquiring an electric quantity data sequence of each electric energy metering terminal in a specified period; If abnormal data exist in the current electric quantity data sequence, determining a target reference electric quantity data sequence based on a matching relation between the electric quantity data sequences of a plurality of other electric energy metering terminals and the change trend of the current electric quantity data sequence, and correcting the abnormal data of the current electric quantity data sequence based on the target reference electric quantity data sequence; And taking the corrected electric quantity data as input, and outputting an abnormal electricity consumption behavior recognition result of the current electric energy metering terminal through a preset abnormal electricity consumption behavior recognition model, wherein the abnormal electricity consumption behavior recognition model is obtained by training a sequential neural network model and a classifier model through electric quantity data sequences of different electric energy metering terminals under different electricity consumption behaviors. Optionally, the time sequence neural network model is an extended long-term and short-term memory model, the classifier model is a related vector machine model, and the abnormal electricity consumption behavior recognition model comprises: the data input module, the extended long-short-term memory model and the related vector machine model; The data input module is used for inputting the acquired electric quantity data sequence into the extended long-period memory model; The extended long-short-term memory model is used for extracting time sequence characteristics of an input electric quantity data sequence and outputting a time sequence characterist