Anomaly Detection of Residential Electricity Consumption Based on Ensemble Model of PSO-AE-XGBOOST

Abstract

In modern society, the dependency on electricity is increasing, with electricity becoming a vital support for daily life, industrial production, and social development. However, incidents like electricity theft contribute to substantial energy consumption and economic losses, escalating the severity of the situation. To address this issue, a method based on the optimization of particle swarm optimization (PSO) and integration of autoencoder and extreme gradient boosting (XGBoost) was proposed in this paper for detecting abnormal electricity consumption in residential areas. Firstly, the user data was preprocessed to enhance data quality and model performance. Subsequently, the particle swarm optimization was utilized to adjust the hyperparameters of the autoencoder, thereby enhancing its performance. The particle swarm optimization searched for the optimal parameter combination in the parameter space of the autoencoder and utilized the found optimal parameters to train the autoencoder, thereby learning effective features from the data. Finally, using the features extracted by the autoencoder, XGBoost was employed as a supervised learning model for classification and anomaly detection, identifying users with abnormal electricity consumption. The results demonstrated that this model outperformed other detection methods such as support vector machines (SVM), K-nearest neighbors (KNN), and decision trees in terms of accuracy and recall. This research is significant for enhancing residential electricity security and energy utilization efficiency, and provides valuable insights for further research and practical applications in related fields.