Modelling Machine Learning-based Energy Loss Detection and Monitoring System for Advanced Metering Infrastructure

Abstract

Non-technical losses (NTL) have rightly been identified as losses arising from energy generated but unaccounted for. They basically occur because of theft and other fraudulent activities surrounding illegal consumption of energy. This loss accounts for massive loss in revenue to utility companies and government and there has been concerted efforts to mitigate such abnormalities thereby saving cost. Although advanced metering infrastructure (AMI) incorporating smart meters have provided some basic organization around management of smart grids and monitoring usage information, it still fails to accurately detect NTL. In this paper therefore a solution to NTL is presented involving the deployment of support vector machines (SVM) as an underlying classifier and integrated with a real-time application interface termed Electricity Usage Classifier interface (ELUCI) for monitoring and pre-processing instantaneous electricity usage time-series data. With this configuration, a classification accuracy of 98.48% was achieved which was a 17.02% improvement over the initial classification models and with a root mean squared error (RMSE) of 0.0894 and an f-measure of 0.979. The developed system can assist governments and utilities to actively monitor and track down energy theft thereby improving revenue and avoiding economic wastages accruing from these activities.