Integrated Data-Model-Driven Identification Method Against Power Theft In Medium And Low Voltage Distribution Networks

Under the background of power organizational reform,the operating pressure of power grid enterprises increases while power theft causes huge losses to power grid enterprises every year.Increasing the crackdown on power theft is of great and urgent economic significance to improve the operating efficiency and to reduce the cost of power grid enterprises,tapping the growth space of internal value.At the same time,with the promotion of the digital transformation of the power grid,the ways of power theft have developed from the traditional physical methods such as private wiring and shortcircuiting of electric meters to new high-tech modes based on digital storage and communication technology.The existing identification methods against power theft are difficult to cope with the challenges of various new ways of power theft.Therefore,this thesis makes an in-depth study on the integrated data-model-driven identification method against power theft in medium and low voltage distribution network,the main contents of the research are as follows:Aiming at the problem that the physical meaning of the characteristics of data-driven power theft identification method is unknown and the multi-level periodic characteristics of users’ power consumption behavior are not considered,a power theft identification method based on multi-level non-negative sparse coding of active power sequence and SVM is proposed.Firstly,the user’s monthly active power sequence is divided into weekly and daily levels,and the pattern features of weekly and daily subsequences are extracted based on multi-level non-negative sparse coding algorithm;Then,based on the comparative analysis of power consumption scenarios between normal users and powerstealing users,the empirical features of weekly and daily subsequences are extracted;Finally,taking the integrated features composed with pattern features and empirical features of weekly and daily subsequences as input,the monthly active power sequence is identified by support vector machine(SVM)algorithm.An example based on the Irish intelligent meter data set shows that the proposed integrated method of feature design can extract power consumption features more effectively than principal component analysis and independent component analysis,and then effectively improve the accuracy of power theft identification in the station area.Aiming at the problem that the data-driven identification method against power theft can be easily disturbed by the randomness of power consumption of users,which leads to the instability of the identification characteristics,a power theft identification method based on voltage-active power sequence correlation analysis is proposed.Firstly,the lowvoltage users are clustered according to the voltage amplitude,and each cluster is constructed as a virtual user of the urban station area;Then,according to the principle that there is an approximate linear negative correlation between the voltage amplitude and the active power of the normal virtual user,whether the virtual user steals power is determined by judging whether the Pearson correlation coefficient between the virtual user’s voltage and active power sequence is beyond the normal range;Finally,the lowvoltage users in the virtual users with power theft are excluded one by one,and the difference of Pearson correlation coefficient between voltage and active power series before and after is calculated,based on which the suspected rankings of power theft for each low-voltage user is given.An example of urban station area network based on Irish smart meter data set shows that the proposed method can accurately identify the powerstealing virtual users in urban station area,and the suspected ranking of several lowvoltage users within them are accurately given,at the same time,the method has good applicability in a variety of power theft modes.Aiming at the problem that the model-driven power theft identification method does not take into account the difference of measurement characteristics between the public transformer and the special transformer,and the accuracy can be easily affected by residual pollution and residual submergence,a detection and identification method against power theft in feeder based on KNN and multiple regression analysis of local variable around distribution transformerm is proposed.In the stage of power theft detection,firstly,the estimation model of feeder network loss rate is constructed based on k-nearest neighbor regression algorithm(KNN)and the measurement of voltage and power of source and load nodes of feeder,and then the power theft detection of each section is carried out according to whether the estimated residual of feeder network loss rate exceeds the limit.In the stage of power theft identification,a unified equivalent model of the local binary tree structure of the distribution transformers(including common transformers and special transformers)is constructed first,and based on this a multivariate linear regression model among local variables of distribution transformer is established;Then the parameters of the multivariate linear regression model for each distribution transformer are estimated by using the historical normal cross-section data;Finally,starting from the end of the feeder,traverse all distribution transformers from bottom to top,and substitute the measured data of the variables in the local binary tree structure of the distribution transformer into the corresponding multiple linear regression model with parameters known to estimate the voltage amplitude of the distribution transformer.We can determine whether the distribution transformer exists power theft according to whether the estimated residual error of the voltage amplitude exceeds the limit,and repair the abnormal power data of the power-stealing distribution transformer.An example of medium voltage feeder network based on IEEE 33 node distribution system and the actual load data of distribution transformer supplied by a power supply company in Chongqing shows that the proposed method can accurately identify the power-stealing distribution transformer and effectively repair its abnormal power data.At the same time,the method also has good anti-interference and sensitivity.