Research On Electricity Theft Detection Method Based On Stacking Ensemble Learning

With the acceleration of social and economic growth,the demand for electricity is constantly expanding,and the grid structure is becoming more and more complicated.In order to further regulate the order of market power supply and maintain a fair and harmonious electricity environment,it is necessary to conduct audits on electricity consumption.The difficulty is It lies in the identification of electricity theft,which will not only affect the fairness of power grid measurement,but also threaten the backbone of the power grid.Based on electricity consumption data,this article explores an efficient and accurate electricity theft detection method that is consistent with the actual characteristics of electricity theft behavior,which is of great significance to ensure the smart electricity management in power enterprises.The research content is divided into the following three points:(1)Aiming at the problem of how to extract the characteristics of electricity theft from a large amount of electricity consumption data,this paper adopts the method of data analysis to analyze the changes of related electrical parameters such as voltage and current caused by common electricity theft,and select the electrical The difference of characteristic parameters is used as the distinguishing feature of the stealing behavior,including the three-phase current unbalance rate,voltage unbalance rate,rated voltage deviation,and power factor,which provide detection indicators and characteristic basis for the following electricity theft detection scheme.(2)In view of the large sample size of electricity consumption data used in this paper,and the imbalance between normal electricity consumption samples and electricity theft samples,when a single classification algorithm is used to predict,the accuracy rate will enter a bottleneck period when the accuracy reaches a certain level.Continue to optimize the algorithm The difficulty is higher.From the perspective of integrated learning,a method of electricity theft detection based on Stacking integrated learning is introduced.First of all,three classic single classification algorithms of KNN,SVM,and DTree are introduced as the learners of the Stacking ensemble learning method according to the characteristics of electricity consumption data,and the classification models based on the Stacking ensemble learning method and the three single classification algorithms are established respectively;secondly,In order to verify the effectiveness of the Stacking integrated learning method for power theft detection,two integrated algorithms,RF and Ada Boost,are added to compare and verify the classification effect;finally,the quantum genetic algorithm(QGA)is used to search for the three classic algorithms in the Stacking integrated learning method.Compared with the method without QGA optimization,the accuracy rate is increased by 8.7%.(3)From the perspective of data balance,the FCM clustering and SMOTE oversampling method are combined for power data balance processing,and the electricity theft detection method based on F-SMOTE and Stacking ensemble learning is introduced.First,for the problem of blurring the boundary between positive and negative sample data when the SMOTE oversampling algorithm processes unbalanced data,the FCM clustering method is used to find the cluster centroids of the stealing samples;secondly,between the centroids of the stealing samples and the stealing samples SMOTE oversampling interpolation is performed to construct a relatively balanced electricity consumption sample and use it as the input of the Stacking integrated learning model.Finally,the power consumption data of a certain place in Yunnan Province is used for verification.Compared with the method that does not use the FCM clustering algorithm to preprocess the data,the correct rate is increased by 3.3%.