Support vector machine based data classification for detection of electricity theft

Abstract

Most utility companies in developing countries are subjected to major financial losses because of non-technical losses (NTL). It is very difficult to detect and control potential causes of NTL in developing countries due to the poor infrastructure. Electricity theft and billing irregularities form the main portion of NTL. These losses affect quality of supply, electrical load on the generating station and tariffs imposed on electricity consumed by genuine customers. In light of these issues, this paper discusses the problems underlying detection of electricity theft, previously implemented ways for reducing theft. In addition, it presents the approximate energy consumption patterns of several customers involving theft. Energy consumption patterns of customers are compared with and without the presence of theft. A dataset of customer energy consumption pattern is developed based on the historical data. Then, support vector machines (SVMs) are trained with the data collected from smart meters, that represents all possible forms of theft and are tested on several customers. This data is classified based on rules and the suspicious energy consumption profiles are grouped. The classification results of electricity consumption data are also presented.