| With the popularity of smart meters,power companies have accumulated a large number of user electricity data,which provides a data base for data-driven detection methods of electricity theft.The intelligent detection method based on machine learning or deep learning can fully tap the value of redundant massive power consumption data of power enterprises,which is of great practical significance for power enterprises to improve the accuracy of identifying power stealing users,reduce the cost of anti power stealing and ensure the safe operation of power grid.Aiming at the problem of identification of power stealing users in high loss area,this thesis proposes a multi model fusion intelligent power stealing detection model,which makes full use of the periodic difference of power consumption data between normal users and suspected users,and constructs the intelligent power stealing detection model from four aspects of global characteristics,adjacent periodic characteristics,pre and post periodic characteristics and statistical characteristics.The main work of this thesis is as follows:(1)This thesis analyzes the background and significance of power stealing detection,which has caused huge economic losses and seriously affected the safe operation of the power grid.This thesis summarizes the research status at home and abroad,and puts forward the main chapters and working arrangements of the thesis.(2)This thesis sums up the common ways of stealing electricity in high loss area and the causes and characteristics of the abnormal data.The stealing behavior will be reflected in the abnormal data,so the detection of stealing electricity based on data-driven is feasible.Build the structure of intelligent power stealing detection model,and elaborate the basic principle of intelligent power stealing detection model.(3)Establish the method of preprocessing the original power consumption data.This thesis describes the process and main measures of data preprocessing,including data vectorization,missing value processing,abnormal value processing,label balance and value standardization.(4)An intelligent detection model of electric larceny based on multi model fusion is proposed.Based on the full connection neural network,convolution neural network and LSTM network,the multi model fusion intelligent power stealing detection model constructs the power stealing detection sub module which captures the global characteristics,the adjacent periodic characteristics and the pre and post periodic characteristics respectively,and uses the random forest as the discrimination layer of the intelligent power stealing detection model.(5)Finally,the performance of the model is tested with the real electrical data of a certain place.The test content includes data processing,model recognition accuracy,and the comparison of SVM and other classification methods.The experimental results show that the intelligent detection model proposed in this thesis has high classification accuracy,good performance in accuracy,recall rate and other indicators,and strong generalization ability.It has important reference significance and reference value for the detection of power theft. |
