Study And Application Of Electricity Theft Detection Model In Station Area

Electrical energy plays a vital role in social economic construction and people’s daily life.As the development of society,the demand for Electrical energy in all industries is also increasing.However,some users are driven by profits to use electricity without paying,which seriously damages the legitimate rights and interests of utilities companies.In recent years,as the value of virtual currency has soared,there are many users steal electricity for virtual currency mining.The extensive construction of Advanced Metering Infrastructure(AMI)has collected massive amounts of user electricity consumption data,providing reliable data sources for load analysis,load forecasting,and load management.At the same time,the rapid development of machine learning technology also provides a powerful weapon for the electricity theft detection.However,many machine learning-based models used for electricity theft detection have low accuracy due to the problems of data imbalance and missing data.Therefore,it is crucial to study the characteristics of electricity consumption data and build high-efficiency electricity theft detection models.The main research of this thesis is shown as follows:1.This thesis gives the definition and classification of electricity theft behavior.According to the content of the existing electricity theft detection research,this thesis divides the types of research into four classes: game theory-based study,power grid state analysis-based study,hardware-based solutions and machine learning-based study.This thesis elaborates the representative work,advantages,limitations and evaluation indicators of these four types of research,and summarizes the public datasets commonly used in this area.Finally,this thesis presents the prospect for the future research direction of electricity theft detection,including data preprocessing,feature engineering and industry user differentiationt.2.Aiming at the electricity theft detection in AMI,a detection method of anti-electricity theft based on gradient boosting is studied.XGBoost(e Xtreme Gradient Boosting)is chosen as the implementation model of the gradient boosting algorithm.This thesis builds an XGBoost-based electricity theft detection model,and tests its ability in detecting the samples of malicious user based the electricity consumption data information provided by the Sustainable Energy Authority of Ireland.The results show that its performance is better than traditional machine learning models,and it still has better detection accuracy in the case of data imbalance.At the same time,this thesis compares the performance of three gradient boosting models,i.e.,XGBoost,Light GBM(Light Gradient Boosted Machine)and Cat Boost,in detecting electricity consumption data of malicious users provided by the State Grid Corporation,and analyzes the impact of model parameters on its performance.The results show that the detection results of the gradient boosting model are better than those of other machine learning models,such as SVM and CNNs.3.This thesis designed a station area electricity theft detection system.The system architecture includes three parts: data access and storage,data analysis model,and human-computer interaction interface.XGBoost-based electricity theft detection model is used as the data analysis model for the detection of malicious users.The human-computer interaction interface includes the user data analysis interface,the suspected user list interface,and the on-site electricity theft verification result interface.The system has been tested in the electricity consumption data provided by the State Grid XXX Provincial Power Supply Service Management Center,and it has shown relatively good detection accuracy.