AdaBoost-CNN: A Hybrid Method for Electricity Theft Detection

Abstract

Electricity theft causes great economic losses and potential safety hazards to society. With the development of smart meters and artificial intelligence, data-driven methods have gradually been applied to electricity theft detection, which improves efficiency and reduces costs. At present, data-driven electricity theft detection methods are dedicated to improving accuracy. To achieve higher accuracy, this paper proposes a hybrid method combining adaptive boosting algorithm (AdaBoost) and convolutional neural networks (CNN) for electricity theft detection. In the proposed method, multiple CNN-based classifiers are trained to extract different features from the electricity consumption data. Then, AdaBoost combines these classifiers into a strong one according to their performance. The experimental results based on the smart energy data from the Irish Smart Energy Trial show the hybrid classifier has better performance than other conventional data-driven methods in electricity theft detection.