A deep learning technique Alexnet to detect electricity theft in smart grids

Abstract

Electricity theft (ET), which endangers public safety, creates a problem with the regular operation of grid infrastructure and increases revenue losses. Numerous machine learning, deep learning, and mathematical-based algorithms are available to find ET. Still, these models do not produce the best results due to problems like the dimensionality curse, class imbalance, improper hyper-parameter tuning of machine learning and deep learning models, etc. We present a hybrid deep learning model for effectively detecting electricity thieves in smart grids while considering the abovementioned concerns. Pre-processing techniques are first employed to clean up the data from the smart meters. Then, the feature extraction technique, like AlexNet, addresses the curse of dimensionality. The effectiveness of the proposed method is evaluated through simulations using a real dataset of Chinese intelligent meters. To conduct a comparative analysis, various benchmark models are implemented as well. Our proposed model achieves accuracy, precision, recall, and F1, up to 86%, 89%, 86%, and 84%, respectively.