The Research On Electricity Larceny Prevention With Data Mining In AMI

It is of great significance to ensure healthy and orderly development of the power industry since it plays an important role in promoting the national economy.However,the phenomenon of electricity larceny has become a major problem for electricity supply enterprises.In recent years,with the extensive use of electricity information collection system,the electricity company achieves marketing automation,meanwhile accumulates a large number of user data.Although these data brings management challenge,it hides great opportunities.These data contains a wealth of electricity behavior information of users,deep digging of these data can effectively reveal the behavior information hidden behind it,and accomplish the accurate analysis of the user behavior.This thesis takes this as a starting point,through the detailed analysis of abnormal behavior of electricity use,relying on a large number of user data,and using data mining technology effectively complete the detection of a variety of abnormal electricity behavior.Aiming at different detection requirements,a weighted LOF algorithm based steal electricity checking scheme and a SVM based stealing electricity judgment scheme are studied,the corresponding problems are optimized,in order to locate the abnormal electricity user efficiently.For large-scale power users,this thesis presents a weighted LOF method based on analytic hierarchy process.the technique firstly analyzes the phenomenon of stealing electricity and constructs a reasonable evaluation system of stealing.Secondly,according to different probability of the abnormal electrical indicators of data can represent the electric larceny,the weight of each electrical index is quantified by analytic hierarchy process(AHP)quantitatively.The massive data can be analyzed by weighted outlier analysis with the weighted LOF algorithm,and using a comprehensive outlier factor to characterize the theft of users.Finally the experiments through the measured data demonstrate that the proposed method can dig out more stealing users compared with the traditional LOF algorithm when the detection rate is low,and have better robustness.In order to realize accurate analysis of small scale users,a SVM based electricity larceny detection method is studied.This method identifies the abnormal users by classifying users based on the characteristics of electricity with the SVM.Aiming at the problem of sample imbalance in the detection data set,the comprehensive processing model of SMOTE + Bagging is constructed,which makes the detection effect of unbalanced samples obviously enhanced.Considering that SVM detection performance is significantly affected by parameters,and the integrated classification based on Bagging is time consuming,the differential evolution algorithm is used to optimize the parameters.The experimental results show that the differential evolution algorithm can improve the detection accuracy of SVM and greatly reduce the time required for optimization.At the same time,the proposed comprehensive processing model can effectively improve the data imbalance problem of training samples,which is of great value to enhance the abnormal electricity consumption of SVM.