Application of importance sampling method for non-technical losses detection in electrical distribution systems using smart meters

Abstract

Electricity theft as one of the primary non-technical losses (NTL) not only causes huge revenue losses but also can result in the surging electricity, the heavy loading of electrical systems, and dangers to public safety. Smart grids help to mitigate the impacts of NTL by integrating information flows with energy flows. Although the availability of massive data generated from smart meters is beneficial in detecting electricity theft, discrimination between honest and fraudulent customers is a difficult task, because fraudulent consumers remarkably outnumber non-fraudulent ones. Therefore, the dataset has an imbalance nature. In this paper, importance sampling method is used to detect irregular consumption, since the accurate estimation of probabilities of rare events is a primary concern of this method. Rare events are almost always defined on the tails of probability density functions. They have small probabilities and occur infrequently in real applications such as the case of benign and fraud consumption. The rare events can be made to occur more often by deliberately introducing changes in the probability distributions that govern their behaviour. As the fraud samples are rare events with very low probability, so importance sampling is well exploited to reduce the variance of these samples. A comprehensive survey is presented regarding the state-of-the-art methods in NTL detection. The proposed methodology is performed by data acquisition and pre-processing to remove bad data and interpolate missing data. Validation of the results is done by using different performance metrics such as accuracy and recall.