Non-Technical Loss Detection From Smart Meter Data: A Extreme Gradient Boosting Approach

Abstract

We propose a Al-based framework for nontechnical-loss (NTL) detection from advanced metering infrastructure (AMI) in modern power systems. First, by extracting data from smart meters, four kinds of feature augmentation methods, including time-series cluster, peak demand analysis, daily consumption analysis, and maximal information coefficient, are exploited to generate 13 features to capture eccentric characteristics of NTLs. After the above feature engineering, a new tree boosting system, called the Extreme Gradient Boosting (XGBoost), will be utilized as the detector. In comparison with other existing approaches, the XGBoost method can indeed speed up the detection process and significantly improve the quality of predictions. To validate the performance of the proposed method, the data set from the Irish CER Smart Metering Project under six types of false data injection (FDI) was used for simulation studies. Experiment results demonstrate that the proposed framework can properly handle the imbalanced data from smart meter data and enhance performance on NTL detection.