Nonintrusive Appliance Energy and Health Monitoring Using Attention-Assisted DL Model and Dynamic Time Warping Distance-Based Algorithm

Nonintrusive load monitoring (NILM) techniques estimate the appliance-level energy consumption in a building by analyzing the aggregate energy information only. This article proposes a novel NILM technique and extends its capabilities to estimate the appliance’s health condition. This enhancement empowers building occupants to make proactive decisions, including preventive maintenance, mitigating potential disasters, and reducing energy wastage. The proposed health-aware NILM operates in two stages: the first stage employs an attention-assisted deep learning (DL) NILM technique that disaggregates the aggregate energy into appliance-specific energy profiles. Subsequently, the second stage estimates the appliance health by assessing the anomalies in the disaggregated profiles. This is done by calculating the dynamic time-warping (DTW) distance against healthy energy profile samples in a library. Test results, conducted on five appliances from three publicly available datasets, validate the technique’s feasibility, accurately detecting known anomalies. Furthermore, the study notes an enhanced anomaly detection (AD) performance aligned with the efficacy of the NILM stage.