Electricity Theft Detection Based on Temporal Convolutional Networks with Self-Attention

The issue of Non-Technical Losses (NTL) is a major concern for power systems, as it results in significant revenue loss for electric utility companies and has a negative impact on the stability and reliability of the power grid. These losses are primarily caused by electricity theft. Electricity theft detection and classification has been an important topic for many years and has seen great success in identifying and preventing thefts in the power grid. However, there is still a need for further research in order to improve the current techniques and address challenges related to missing values, data imbalance and classification accuracy. We introduce a system for electricity theft detection based on Temporal Convolutional Networks (TCNs) with Self-Attention. The proposed classifier is evaluated against existing classifiers from literature. Evaluation results show that this classifier exhibits high performance in terms of accuracy, F1-score, and AUC score.