Towards Light Intensity Assisted Non-Intrusive Electricity Disaggregation

Considering the contribution of building systems in the energy and electricity consumption globally and in the US, facilities management and operational strategies that help improve energy efficiency have been the subject of several studies. Electricity desegregation through non-intrusive methods is one of these strategies that provide cost-effective solutions for energy monitoring with higher spatiotemporal resolution (i.e., at appliance level). Since these techniques use machine learning for signal processing algorithms on aggregate data from one sensing point to infer contribution of each load, their application commonly calls for algorithm configuration (i.e., training) in a new environment. In this study, we have proposed a novel feature extraction approach for lighting loads through spectral analysis of the light intensity signal in order to automate the training of algorithms in the non-intrusive load monitoring systems. The proposed feature identifies the artificial light variation in presence of natural light and improves the rate of detecting lighting loads operations with minimum number of light intensity sensor.