New features for detection of nontechnical losses considering PV installed at customer side

Nowadays, nontechnical loss (NTL) has been an influential factor on the benefits of electric power utilities. At the same time, with distribute generation extensively installed, the consumption patterns having many similarities between dishonest users and normal customers with photovoltaic (PV) installed. Improving the reliability of NTL detection algorithm becomes particularly important. In this paper, consider the PV installed at customer side, the sunlight intensity and time characteristic related features have been employed to upgrade the performance of electricity theft detection algorithm. Compared with traditional feature combination, this new method gets higher accuracy and robustness to misclassified training data.