Research On Abnormal Power Consumption Detection Technology Based On Deep Learning

With the rapid economic growth and the rapid development of science and technology,the role of the power system in the national energy strategy has become increasingly prominent,and the safety of power supply has become more and more important.With the continuous expansion of the power supply network,electricity theft by power users has occurred from time to time,and the methods of stealing electricity have gradually become novel,the means have become more technological,and the process has become gradually concealed.The theft of electricity not only causes economic losses to power companies,but also threatens the safety of the power grid.Therefore,it is of great significance to study electricity theft detection technology.The traditional electricity theft monitoring mode faces problems such as difficulty in feature modeling,less data of labeled samples,severe imbalance of normal abnormal samples,and low detection efficiency.In order to solve these problems,this article combines deep learning technology to carry out the following research work:(1)First,preprocessing the original electricity consumption data.According to the different characteristics of different users,the original data set is mainly processed by reliable data screening,adding missing values,and size normalization.Through this processing,the impact of a large amount of noise in the original data set on model training can be reduced.(2)Secondly,in view of the large difference between the amount of positive and abnormal sample data,this paper first adopts the Wasserstein Distance-based Improved Generative Adversarial Network(WGAN)to generate new electricity theft samples from the existing electricity theft data,and then,The generated samples are mixed with the existing samples to obtain more sample data,so that the electricity theft sample data and the normal data reach a relatively balanced state.(3)For DBN,the random initialization weights and biases are easy to cause the model to fall into the local optimal problem.This paper uses an improved genetic algorithm(IGA)to optimize the weight and bias parameters of DBN,establishes an IGA-DBN-ELM electric stealing detection model,and compares it with multiple existing models,using accuracy,ROC curve,AUC and other parameter pairs.Its evaluation proves the effectiveness of the proposed IGA-DBN-ELM model.At the same time,in order to further ensure the efficiency of electricity theft detection,this paper also proposes a DT-SVM model optimized based on an improved Antlion algorithm.This model can accurately identify the users and their types of electricity theft,so as to realize the precise monitoring of stealing behaviors.(4)Based on the above research content,based on the PyCharm development environment,a software monitoring system for user theft detection software is designed using PyQt.