Non-intrusive load monitoring based on bagging decision trees and the selective features for commercial building loads

A non-intrusive load monitoring (NILM) system, also called an energy disaggregation system, allows obtaining information relating to the power absorption of individual appliances connected to a user, through the use of voltage and current transducers positioned at its connection point to the grid. In this paper, the collected data for the NILM system are obtained from four three-phase loads, namely two chillers and two induction motors with a high sampling rate. The proposed method showed all the possible features based on the transient state to be extracted and examined to select the efficient features for commercial building loads by using a bagging decision trees (BDT) classifier. Current, and harmonics features proved that can easily classify the loads rather than the other features.