A Hybrid Approach Based on Principal Component Analysis and Convolution Neural Network For Power Theft Detection

Abstract

Power theft is a persistent problem faced by electricity supply companies, leading to non-technical losses that can negatively impact the quality of electricity as well as profits. The emergence of advanced metering infrastructure (AMI) has presented a new opportunity to detect power theft using data from smart meters. In this study, we propose a hybrid approach that combines principal component analysis (PCA) and deep convolution neural network (CNN) to identify power theft and improve electricity monitoring. Our proposed technique involves three stages, namely feature selection, extraction, and classification, which are applied to smart meter data to assist energy supplier companies. The CNN is responsible for classifying the extracted features into either theft or non-theft categories, with optimized hyperparameters that enhance the accuracy of the model. The CNN-PCA method proposed in this study achieves a high accuracy rate of 94.76%, outperforming previous approaches. The models generated from this research exhibit high accuracy and low error rates in extensive simulations, making them a valuable tool for power supply companies to combat power theft.