MTopsOREDC: M Tops KNN for Online Reinforced Electric Device Classification

Abstract

Home and commercial energy management systems (HEMS and CEMS) are increasingly dependent upon fine grained analysis of device level consumption for visualization, demand side management(DSM), and long term diagnostics. Nonintrusive load monitoring (NILM) provides the means to provide this analysis without the need for intrusive and costly device level monitoring. Based on device profiles different approaches have using either high frequency voltage-current (VI) data or low frequency power data to disaggregate the loads. In this study we report the results of using a reinforcement hybrid approach using both high frequency VI and low frequency power data in a unique voting mechanism. We show that by this hybridization and reinforcement we are able to identify a wider verity of devices. Results show that through this strategy we can achieve increase the accuracy from 95 % to 98 % in standard datasets.