Abnormalities And Electricity Theft Detection Based On Electricity Consumption Behavior Analysis

With the rapid development of the economy,the consumption of electricity is showing explosive growth,and higher requirements are proposed for anomaly electro-data detection.Abnormal electricity consumption has a direct impact on the income of the power supply company,and hinders the development of smart grid.Furthermore,the electricity stealing which changes the metering device privately not only damaged the power facilities,but also easily triggered the fire,and threatened the safe and stable operation of the power grid.Therefore,it is imperative to detect abnormal power consumption for power users.In this paper,an anomaly electricity detection method based on unsupervised learning is adopted,which can be better applicable to the current situation of the few samples of the abnormal power consumption.Especially,this method detects the abnormal electricity though analyzing daily electro-data,which make it possible to respond and repair the scene quickly.Concretely,this method first analyzes the electro-data by multi-feature fusion algorithm,and detects whether there is abnormal electricity consumption.Then,the result of abnormal electricity is analyzed by electricity mode analysis,and finally the abnormal power consumption is judged,which greatly improves the accuracy.The main contents are as follows:(1)Firstly,the abnormal reliability of the related electrical parameter data,such as voltage,current and power factors,is analyzed,and the abnormal degree of these features is obtained.Then a multi-feature fusion algorithm based on D-S evidence theory is used to fuse the abnormal information of these features to get the exception reliability allocation of all the features.Finally,the abnormal degree is calculated by the abnormal index algorithm to judge the electricity consumption synthetically.(2)The abnormal results detected by multi-feature fusion algorithm are analyzed through the power mode algorithm.First,the daily load curve formed by hourly power of the user is clustered by K-means to extract the information of electricity mode and construct the parameter model,which detect the electricity consumption.Concretely,those which belong to the electricity are normal,otherwise abnormal.In the experiment stage,through the analysis of all kinds of electricity data of Guangzhou users and test the real abnormal electricity samples,this method proposes indicators such as confusion matrix,ROC curve to evaluate the test results.Furthermore,the results are compared with the LOF in the same condition.The accuracy rate reached 95% and increased by 4 percentage points.