Research On Intelligent Analysis Method Of Power Stealing User Location Based On Deep Learning

Power theft is an act of stealing power from the distribution system through illegal means in an attempt to reduce power expenditures.It is a substantial problem that affects the stability of the power system and the economic efficiency of the power grid.With the rapid development of social economy,industry and technology,the demand for electricity is growing and power theft will lead to greater economic losses.In addition,some ways of power theft may start fires and cause serious safety problems.Therefore,it is urgent to strengthen the investigation and punishment of power theft.And the widespread use of smart meters provides the basis for more efficient detection of power theft users.The work in this thesis proposes a power theft detection method based on principal component analysis,Gram’s corner field and deep residual neural network based on smart meter collection data.The principal component analysis method is applied to compress multiple electricity usage detection indicators,with the aim of obtaining multidimensional electricity usage data features without changing their time-series characteristics.The Gram’s corner field method is used to convert the time-series electricity data features of individual users into two-dimensional images for the purpose of maintaining user timeseries features and conducting analysis based on user units.The images generated by the Gram’s corner field method contain information about the electricity consumption features of many users,and the classification is performed by deep residual neural networks to locate the electricity theft users.The proposed algorithm is tested on a dataset consisting of normal and abnormal user data.The results show that the power theft detection method based on Gram’s corner field and deep residual neural network outperforms the traditional power theft detection method and has a high accuracy.However,there is still the problem of data imbalance,which greatly affects the accuracy of the classification model,so the anomaly data set is expanded by deep generative adversarial neural networks to achieve data balance.In addition,the model was also optimized,and a ConvNeXt neural network model was formed based on the deep residual neural network with five main aspects of optimization,which was used for classification processing.Based on the real data set,several models are designed for comparison experiments under different data sets,and it can be found that the accuracy of the models is generally improved after data enhancement,and the accuracy and generalization of the ConvNeXt model are better than the rest of the models,which can verify the effectiveness of the proposed method and can achieve better results in locating electricity thieves.