Towards Identification of Appliances in Conventional Homes using ML and Descriptive Statistics

Providing ancillary services for future smart grids is challenging because of the rapidly growing electricity demand, while having uncertainties in renewable power generation, limited availability of conventional spinning reserves, and expensive storage systems. Thus, Home Energy Management Systems (HEMSs) have been gaining increased attention nowadays. To capitalize on the potential of HEMS, which supports customer participation and two-way power communication so as to maintain the generation-load balance, two interconnected challenges, i.e., load monitoring and identification of appliances consumption, need to be addressed. In this contribution, a comprehensive nonintrusive load monitoring (NILM) algorithm for appliance identification is proposed, which only requires a single sensing point from conventional homes, i.e., the aggregated power signal. Machine learning algorithms and both time-domain and frequency-domain based feature extraction are utilized in the development of the proposed solution. Simulation experiments are performed using the Reference Energy Disaggregation Dataset (REDD), a real household power consumption dataset. Simulation results demonstrate the effectiveness of the proposed NILM strategy with F1-score values of 97.659%, higher than those reported in the state-of-the-art.