Detection of Electricity Theft in Metering Systems using Convolutional Neural Network

Financial instability, revenue loss, and operational inefficiencies are some of the factors associated with electricity theft in Nigeria and worldwide. In consequence, the problem of electricity theft will continue to undermine the growth of the power utility sector if not adequately mitigated. Traditional electricity theft detection methods, such as manual inspections, rule-based models, and statistical analytics are increasingly ineffective against emerging and advanced tampering techniques. Hence, this paper presents an intelligent energy metering system based on deep Artificial Neural Network (ANN) for real-time anomaly detection of electricity consumption patterns. Historical energy consumption data of 2,400 metered customers for 24 months in Southwestern Nigeria were used in this study. The data was first pre-processed and then partitioned into 80% training dataset and 20% testing dataset. The training dataset was used to train a Convolutional Neural Network (CNN) model, a subset of ANN based on Deep Learning (DL). The developed CNN model was tested with the testing dataset, and the performance was evaluated using precision, recall, F1-score and accuracy. Simulation results showed that the CNN model demonstrates high classification performance with an accuracy of 96.87%, precision of 96.88%, recall of 97.83% and F1-score of 96.77%. In addition, the proposed CNN model outperformed some existing methods.