CN-121998244-A - Intermittent electricity stealing quantity evaluation method based on double-model reconstruction error comparison

CN121998244ACN 121998244 ACN121998244 ACN 121998244ACN-121998244-A

Abstract

The embodiment of the disclosure relates to an intermittent electricity larceny assessment method based on double-model reconstruction error comparison, which comprises the steps of obtaining multi-user electricity consumption sequence data, constructing standard electricity consumption sequence data according to the multi-user electricity consumption sequence data, carrying out mask processing on the standard electricity consumption sequence data, training a normal user model and an electricity larceny user model by using a training set after the mask processing, inputting a testing set and a verification set after the mask processing into the trained normal user model and the electricity larceny user model for reconstruction, determining an electricity larceny judgment threshold range according to a first normal reconstruction sequence and a first electricity larceny reconstruction sequence and based on user labels of users of the testing set, determining electricity larceny suspicion index values of users of the verification set according to a second normal reconstruction sequence and a second electricity larceny reconstruction sequence, and assessing the electricity larceny suspicion grades of all users to be tested in the verification set based on the electricity larceny judgment threshold range and the electricity larceny suspicion index values. The embodiment of the disclosure can effectively avoid misjudgment on the electricity stealing behavior.

Inventors

LUAN ZHIRONG
YAN ZHENG
WANG QIAN
WANG YUSEN
ZHANG ZHIYAN

Assignees

西安理工大学

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. An intermittent electricity stealing quantification assessment method based on double-model reconstruction error comparison is characterized by comprising the following steps: acquiring power consumption sequence data of multiple users, and constructing standard power consumption sequence data according to the power consumption sequence data of the multiple users, wherein the standard power consumption sequence data comprises a training set, a testing set and a verification set; Masking the standard power consumption sequence data; Constructing a normal user model and an electricity stealing user model, and respectively training the normal user model and the electricity stealing user model by using a training set after mask processing; inputting the test set subjected to mask processing into a trained normal user model and a trained electricity stealing user model respectively for reconstruction to obtain a first normal reconstruction sequence and a first electricity stealing reconstruction sequence; Determining a first normal comprehensive fitting index and a first electricity stealing comprehensive fitting index according to the first normal reconstruction sequence and the first electricity stealing reconstruction sequence, determining electricity stealing suspected index distribution based on user labels of all users in the test set, and determining an electricity stealing judging threshold range according to the electricity stealing suspected index distribution; Inputting the verification set subjected to mask processing into a trained normal user model and a trained electricity stealing user model respectively for reconstruction to obtain a second normal reconstruction sequence and a second electricity stealing reconstruction sequence; Determining a second normal comprehensive fitting index and a second electricity stealing comprehensive fitting index according to the second normal reconstruction sequence and the second electricity stealing reconstruction sequence, predicting the electricity stealing suspicion index value of each user to be tested in the verification set by using the second normal comprehensive fitting index and the second electricity stealing comprehensive fitting index, and evaluating the electricity stealing suspicion grade of each user to be tested in the verification set based on the electricity stealing judgment threshold range and the electricity stealing suspicion index value.
2. The intermittent electricity theft quantization evaluation method based on double model reconstruction error comparison according to claim 1, wherein the training set after mask processing trains the normal user model and the electricity theft user model, respectively, comprising: the training set after mask processing is used as an original training sample mask sequence and is respectively input into the normal user model and the electricity stealing user model for reconstruction, so as to obtain a third normal reconstruction sequence and a third electricity stealing reconstruction sequence; Calculating a first mean square error loss corresponding to the normal user model according to the original training sample mask sequence and a third normal reconstruction sequence, and calculating a second mean square error loss corresponding to the electricity stealing user model according to the original training sample mask sequence and the third electricity stealing reconstruction sequence; The normal user model and the power stealing user model are trained by minimizing the first and second mean square error losses, respectively.
3. The intermittent electricity stealing quantization evaluation method based on double-model reconstruction error comparison according to claim 1, wherein the inputting the test set after mask processing into the trained normal user model and the electricity stealing user model for reconstruction, respectively, to obtain a first normal reconstruction sequence and a first electricity stealing reconstruction sequence comprises: Taking the test set subjected to mask processing as an original test sample mask sequence, and respectively inputting the test set into the normal user model and the electricity stealing user model; The normal user model carries out linear projection processing on the original test sample mask sequence to obtain the first normal reconstruction sequence; and the electricity stealing user model carries out linear projection processing on the original test sample mask sequence to obtain the first electricity stealing reconstruction sequence.
4. The intermittent power-stealing quantification assessment method based on dual model reconstruction error contrast of claim 3, wherein the determining a first normal fit index and a first power-stealing fit index from the first normal reconstruction sequence and the first power-stealing reconstruction sequence comprises: calculating a first normal comprehensive fitting index according to the first normal reconstruction sequence and the original test sample mask sequence; and calculating a first electricity stealing comprehensive fitting index according to the first electricity stealing reconstruction sequence and the original test sample mask sequence.
5. The intermittent electricity theft quantification assessment method based on double-model reconstruction error comparison according to any one of claims 1-4, wherein the determining electricity theft suspicion index distribution based on user labels of users in the test set, determining an electricity theft judgment threshold range according to the electricity theft suspicion index distribution, comprises: Calculating the electricity stealing suspicion index value of each user in the test set by using the first normal comprehensive fitting index and the first electricity stealing comprehensive fitting index; determining the distribution of the electricity larceny suspicion indexes based on user labels of all users in the test set and the electricity larceny suspicion index values of all users, wherein the labels comprise normal user labels and electricity larceny user labels; The initial threshold is determined by equation (1): (1) Wherein, the Representing an initial threshold value of the value, The average value of the electricity stealing suspicion index values of normal users in the test set is represented under the threshold value parameter theta; the standard deviation of the electricity stealing suspicion index value of the normal user in the test set is represented under the threshold parameter theta; The average value of the electricity larceny suspicion index values of the centralized electricity larceny user is tested under the threshold value parameter theta; The method comprises the steps of representing standard deviation of electricity larceny suspicion index values of electricity larceny users in a test set under a threshold parameter theta, wherein J (theta) represents an optimization objective function; According to the initial threshold value, adopting a gradient descent optimization algorithm to adaptively determine an optimal threshold value; And determining the range of the electricity larceny judgment threshold according to the optimal threshold and the discrete degree of the electricity larceny suspected index distribution.
6. The intermittent theft quantization assessment method based on double model reconstruction error comparison according to claim 5, wherein adaptively determining an optimal threshold using a gradient descent optimization algorithm according to the initial threshold comprises: Calculating the gradient direction by adopting a gradient descent optimization algorithm: (2) Wherein, the The direction of the gradient is indicated, Indicating a very small amount of disturbance that is present, Is represented at a threshold value The resulting effect quantity is calculated in the following, Respectively expressed in threshold values Calculating the obtained effect quantity; updating the threshold according to the gradient direction by the formula (3): (3) Wherein, the Representing the threshold value of t +1 iterations, A threshold value representing t iterations is represented, Representing an initial learning rate; And repeatedly calculating the gradient direction and updating the threshold value until a preset iteration condition or maximum iteration times are met, so as to obtain the optimal threshold value.
7. The intermittent electricity stealing quantization evaluation method based on double-model reconstruction error comparison according to claim 1, wherein the inputting the verification set after mask processing into the trained normal user model and the electricity stealing user model for reconstruction, respectively, to obtain a second normal reconstruction sequence and a second electricity stealing reconstruction sequence comprises: Taking the verification set after mask processing as an original verification sample mask sequence, and inputting the verification set into the normal user model and the electricity stealing user model; the normal user model carries out linear projection processing on the original verification sample mask sequence to obtain the second normal reconstruction sequence; And the electricity stealing user model carries out linear projection processing on the original verification sample mask sequence to obtain the second electricity stealing reconstruction sequence.
8. The intermittent electricity larceny assessment method based on double-model reconstruction error comparison according to claim 7, wherein the determining a second normal comprehensive fitting index and a second electricity larceny comprehensive fitting index according to the second normal reconstruction sequence and the second electricity larceny reconstruction sequence, predicting the electricity larceny suspicion index value of each user to be tested in the verification set by using the second normal comprehensive fitting index and the second electricity larceny comprehensive fitting index, and assessing the electricity larceny suspicion level of each user to be tested in the verification set based on the electricity larceny judgment threshold range and the electricity larceny suspicion index value comprises: Calculating a second normal comprehensive fitting index according to the second normal reconstruction sequence and the original verification sample mask sequence, and calculating a second electricity stealing comprehensive fitting index according to the second electricity stealing reconstruction sequence and the original verification sample mask sequence; performing difference processing on the second normal comprehensive fitting index and the second electricity stealing comprehensive fitting index to obtain electricity stealing suspicion index values of all users to be tested in the verification set; And comparing the electricity larceny suspicion index value of each user to be tested in the verification set with the electricity larceny judgment threshold range to determine the electricity larceny suspicion grade of each user to be tested in the verification set.
9. The intermittent theft quantization assessment method based on double model reconstruction error comparison according to claim 1, wherein the masking process is a multiple block masking process.
10. The intermittent electricity theft quantization assessment method based on double-model reconstruction error comparison according to claim 1, wherein the constructing standard electricity use sequence data according to the electricity use sequence data of the multiple users comprises: And preprocessing the multi-user power consumption sequence data to obtain the standard power consumption sequence data.

Description

Intermittent electricity stealing quantity evaluation method based on double-model reconstruction error comparison Technical Field The disclosure relates to the technical field of electricity theft evaluation, in particular to an intermittent electricity theft quantization evaluation method based on double-model reconstruction error comparison. Background The electric power resource is taken as an indispensable basic energy source in the modern society, and the stable supply and the fair metering of the electric power resource are directly related to national life safety and economic development. The electricity stealing behavior is taken as a means for illegally acquiring the electric power, so that huge economic loss is caused to an electric power enterprise, and continuous threat is formed to safe and stable operation of the power grid. With the large-scale deployment of intelligent electric meters and electricity consumption information acquisition systems, electric power companies can acquire massive user electricity consumption time sequence data currently, and a data foundation is laid for a data-driven electricity stealing detection technology. The current electricity larceny detection technology mainly comprises two technical routes of hardware detection and software analysis. The hardware detection is mainly realized by additionally installing a special sensor or a monitoring device, and electrical parameter anomalies such as voltage, current and the like are collected in real time to directly judge. The software analysis mainly uses machine learning, deep learning and other algorithms to automatically identify abnormal modes by analyzing historical electricity consumption data of users, and the mode has become the mainstream of current researches. However, the existing technical route has the defects that due to the fact that intermittent electricity stealing behaviors have non-continuous characteristics, the existing technical route is insufficient in recognition capability of the intermittent electricity stealing behaviors and is extremely easy to generate false alarm, and high-precision separation of a normal electricity consumption mode and an electricity stealing behavior mode is difficult to realize when the intermittent electricity stealing is dealt with on the basis of a traditional characteristic engineering method or a part of deep learning model, so that the electricity stealing recognition precision is affected, and misjudgment is caused. Accordingly, there is a need to provide a new solution to ameliorate one or more of the problems presented in the above solutions. It should be noted that the information disclosed in the above background section is only for enhancing understanding of the background of the present disclosure and thus may include information that does not constitute prior art known to those of ordinary skill in the art. Disclosure of Invention The embodiment of the disclosure aims to provide an intermittent electricity stealing quantification assessment method based on double-model reconstruction error comparison, which can carry out accurate electricity stealing suspicion grade assessment on each user to be tested in verification set, and effectively avoid misjudgment on electricity stealing behaviors. According to an embodiment of the present disclosure, there is provided an intermittent electricity theft quantization evaluation method based on a double-model reconstruction error comparison, including: acquiring power consumption sequence data of multiple users, and constructing standard power consumption sequence data according to the power consumption sequence data of the multiple users, wherein the standard power consumption sequence data comprises a training set, a testing set and a verification set; Masking the standard power consumption sequence data; Constructing a normal user model and an electricity stealing user model, and respectively training the normal user model and the electricity stealing user model by using a training set after mask processing; inputting the test set subjected to mask processing into a trained normal user model and a trained electricity stealing user model respectively for reconstruction to obtain a first normal reconstruction sequence and a first electricity stealing reconstruction sequence; Determining a first normal comprehensive fitting index and a first electricity stealing comprehensive fitting index according to the first normal reconstruction sequence and the first electricity stealing reconstruction sequence, determining electricity stealing suspected index distribution based on user labels of all users in the test set, and determining an electricity stealing judging threshold range according to the electricity stealing suspected index distribution; Inputting the verification set subjected to mask processing into a trained normal user model and a trained electricity stealing user model respectively for reconstruction to obt