Deep Neural Networks for Electric Energy Theft and Anomaly Detection in the Distribution Grid

Time series classification is a fundamental problem when applied to energy distribution issues. In this paper, the authors propose a solution for the detection of electric energy theft (as well as of electric energy anomalies) by introducing a novel time series classification. Data obtained via actual measurements in industrial sites were employed. Our approach was based on the training of a DNN to recognize whether a measurement of a single-day energy profile were subject to any anomaly. Our proposed method was tested and experimentally validated against the results of accepted benchmarks. The outcomes clearly indicate that the performance of our methodology does outperform the other architectures employed as a benchmark, considering the accuracy and its standard deviation.