Research On Non-technical Losses Detection Method Of Distribution Network Based On Deep Learning

Non-technical Losses(NTL)are defined as the sum of unaccounted power Losses caused by power theft,instrument fault,measurement error and other reasons.Such power Losses related to various fraudulent power consumption behaviors of distribution network users can be collectively referred to as NTL,the main component of which is electricity theft.Non-technical losses not only cause economic loss to power companies,but also bring security risks to the society.Traditional non-technical losses detection methods are inefficient,low accuracy,and time-consuming and laborious.Moreover,various high-tech intelligent devices are illegally used to steal electricity,which brings new challenges to the detection of nontechnical losses.With the popularization of smart electricity meters,electricity consumption data has seen explosive growth.It is of great significance to combine power big data with data mining technology to mine effective information from electricity consumption data and develop an accurate and effective non-technical loss detection method for distribution network.In this paper,the non-technical losses detection method of distribution network is studied.Firstly,the sample of electricity consumption data obtained from users is preprocessed.Data processing is mainly aimed at the noise and missing problems existing in the data.Then,the power consumption characteristics were extracted.Based on the daily electricity consumption data collected by the smart meter,the power consumption characteristics of voltage imbalance rate,current imbalance rate,single-phase power imbalance rate and three-phase power imbalance rate were extracted.Finally,using Support Vector Machines(SVM)algorithm,Combining the Particle Swarm Optimization(PSO)with Back Propagation(BP)neural network and Long Short-Term Memory(LSTM)neural network,the SVM model is constructed,and the detection model based on PSO-BP neural network and PSO-LSTM neural network are proposed to realize the detection of non-technical losses of distribution network.The model designed in this paper is realized programmatically by using Python,and the collected data is used for example verification.According to the analysis of the experimental results,it is found that the PSO algorithm improves the detection effect of BP neural network and LSTM neural network model,and the LSTM neural network model optimized based on PSO algorithm has a better effect on the detection of nontechnical losses,and these models are able to accurately realize the detection of nontechnical losses.