Research On Identification Method Of Electric Larceny Behavior Based On Big Data Analysis And Omni-directional Anti-electric Larceny System

Stealing electricity will not only increase the abnormal line loss of power grid,damage the economic benefits of power enterprises,but also destroy the normal power supply order.In addition,the identification and evidence collection of high-tech stealing behaviors such as strong magnetic field,strong radio and high-frequency and high-voltage are also difficult problems to be solved.The traditional methods of electricity larceny identification mainly include manual regular inspection,user reporting and random sampling inspection,which have weak purpose and poor timeliness.At the same time,the accuracy of the existing method is low.Therefore,it is of great engineering significance to study an accurate,reliable and real-time method and system for identifying and evaluating the amount of stealing electricity.Based on the study of the theory of limit spectrum distribution of large-dimension random matrix and LSSVR time series prediction,this paper proposes a power stealing behavior identification algorithm based on the theory of limit spectrum distribution of large-dimension random matrix and a power stealing quantity evaluation method based on semi-supervised learning mmd-im ABC-LSSVR time series prediction algorithm.It also studies the omni-directional power stealing prevention system and realizes the omni-directional identification of traditional power stealing behavior and high-tech power stealing behavior.In this paper,the limit spectrum distribution theory of large-dimensional random matrix is studied.Based on this,the limit spectrum density and ring law of AR(1)model sample covariance matrix are improved and proposed.Mean while,MP law criterion and ring law criterion based on frechet distance and ring law criterion based on the proportion of characteristic roots outside the ring and the density of characteristic roots are also proposed.At the same time,the concept,characteristics,conditions,methods and classification of time series prediction algorithm are analyzed,and the basic steps of time series prediction algorithm based on least square support vector regression(LSSVR)are studied.It lays a theoretical foundation for the identification of stealing behavior and the estimation of stealing quantity.Based on the limit spectrum distribution theory of large-dimension random matrix,this paper improves the comprehensive criterion of power stealing behavior analysis of users based on MP law and circular law criterion,and puts forward the quantitative criteria for judging the abnormal degree of line loss,the abnormal degree of power consumption in stations and the suspected degree of power stealing by users,and takes the abnormal line loss as the screening criterion,reduces the suspected stations according to the abnormal situation of power consumption in the upper stations of abnormal lines,and finally determines the suspected power stealing user range in the suspected stations.In order to solve the problem that the time section of stealing electricity is difficult to determine,a method based on the variation of characteristic root point density with time in ring law is proposed to determine the time section of stealing electricity accurately.In addition,the method of fast identification of user's electricity using abnormality based on sliding window is studied,which has the characteristics of high accuracy and sensitivity,and realizes the fast diagnosis of user's electricity using abnormality.The validity and accuracy of the method are verified by a power grid company's actual stealing cases.In this paper,the power stealing estimation method based on LSS VR time series prediction algorithm is studied,and the LSS VR algorithm is improved by using semi-supervised learning algorithm.The multi-step prediction of time series is realized,and the rationality and accuracy of power stealing estimation are improved.At the same time,the maximum mean difference(MMD)is introduced to improve the semi-supervised learning algorithm,and the collaborative training sample composed of current data and historical data is constructed to improve the utilization rate of historical data.In addition,artificial bee colony(ABC)algorithm based on crossover operator is used to optimize the model parameters,which improves the global optimization ability of parameters and the prediction accuracy of time series.The example analysis shows that the proposed method is more accurate than mmd-LSSVR and mmd-ABC-LSSVR algorithm,and its average relative error is only 0.42%.Compared with the traditional method,the proposed method is more concise and applicable.On the basis of the above research,the omni-directional anti-theft system is developed.The system is composed of the software based on the big data analysis technology and the high-tech anti-theft device based on the strong magnetic field,strong radio interference and high-frequency and high-voltage monitoring technology.The system realizes the omni-directional identification of the traditional and high-tech means of stealing electricity,the accurate estimation of the amount of stealing electricity by users and the timely identification and evidence collection of high-tech stealing electricity.The feasibility and reliability of the system are verified by the actual stealing cases and the simulation of high-tech stealing behavior interference experiment.