A Naive Bayes Classifier for Detecting Unusual Customer Consumption Profiles in Power Distribution Systems - APSPDCL

Availability of electric power has been the most essential source in acquiring industrial, social and economic developments in any state in India. Every day the Power distribution systems face new challenges to estimate the technical and commercial losses. Apart from technical losses, there are non-technical losses like electricity theft, vandalism to electrical substations, poor meter reading and improper accounting etc. In this work the non-technical losses are investigated by the end user abnormalities in power distribution system using data mining techniques, so that the transmission and distribution losses along the lines will be detected quickly and hence reduced. The model consists of two stages. In the first stage Fuzzy c-Means technique is widely used clustering technique to combine group of end users with homogeneous consumption profiles and to eliminate customers of abnormal consumption profiles. In the second stage a fine tuned classification technique, Naive Bayes is applied. The distances between clusters are measured by using the Euclidean distance, the maximum usage identifies as fraudsters. The proposed technique was tested on the real time data lead to defect detection compared record of respective electricity distribution system. Experimental results signify that the cascaded Fuzzy C-Means and Naive Bayes have enhanced the classification accuracy.