Abnormal Electricity Behavior Identification With Distribution Information Data

The abnormal behavior of power users not only causes great economic losses of power companies,but also endangers the safe operation of power grids,so it is always the focus of power inspection.The traditional anomaly detection method is too dependent on manpower investigation which cause it's low efficiency,and the abnormal detection method based on metering system give too many false alarms which means it is lack of practicability,so the power companies need to develop new technology.With more and more electricity information data has flow into the central station,making using datamining to identify abnormal behavior of electricity possible.Based on the research of user behavior and anomaly identification method,this paper constructs the identification model of abnormal behavior of consumers with the information of electricity consumption,and accommplish a high efficiency of anomaly identification.The user's behavior can be described as the load data,so the abnormal behavior of electricity is the described as the abnormal power load data.Abnormal power data is caused by two reasons,technical and non-technical,the main reason is non-technical.Common abnormal behavior of electricity is stealing with the undervoltage method,stealing with undercurrent method and with phase-shifting,etc.,these various forms of stealing behavior is the primary goal of detection.The recognition algorithms of anomaly behavior include density estimation method,reconstruction based method and support domain based method.Among them,the SVDD algorithm based on support domain method is the most mainstream technology in anomaly detection field with its flexible data description capability and excellent generalization ability.In this paper,the SVDD model is proposed to be used in abnormal behavior detection.The basic idea of SVDD algorithm is to construct a hypersphere by learning to surround the normal class as much as possible.Once the hypersphere is constructed,it can be determined whether it is abnormal by detecting whether the new sample is inside or outside the hypersphere.However,because the internal classification characteristic of the userdata,which leads to the bad characteristics of the hypersphere tightness,affects the detection accuracy of classical SVDD.To solve this problem,an improved SVDD anomaly behavior identification model is proposed in this paper.The improved model pre-classifies the user data and then uses the SVDD model to detect the data,which solves the adverse effect of the SVDD algorithm.The pre-classification algorithm is an adaptive FCM algorithm,which avoids the classification error caused by subjective set of the number of classification,and obtains the optimal classification.In the example,the improved SVDD model combined with the adaptive FCM algorithm obviously increases the tightness of the hypersphere compared with the original SVDD model,and shows a better performance.Through the improved SVDD anomaly identification model,the power supply enterprises can identify abnormal customers more effectively,crack down on abnormal electricity consumption,protect the power infrastructure and corporate interests.