Nontechnical Loss Detection using Neural Architecture Search and Outlier Detection

Abstract

Electricity supply is essential to economy growth and improvement of people’s life. For a long time, illegal electricity theft not only affects the supply of power, but also causes significant economic loss. Traditional techniques for detecting electricity theft are inefficient and time-consuming. Data-based detecting algorithms become a new solution. This article analyses the features of electricity consumption, current, voltage and opening records under various electricity theft modes and proposes a new simulation method for electricity theft users. Based on the simulation dataset, a feature extraction method based on neural architecture search (NAS) is proposed. The advantage of this feature extraction model is demonstrated in the comparison experiments with other feature extraction model. Finally, the effectiveness and accuracy of the electricity theft detection method based on NAS model and outlier detection are verified through an industrial case study.