Evaluation of Diffusion Models on Non-Intrusive Load Monitoring

Non-Intrusive Load Monitoring (NILM) refers to as a technology of disaggregating out individual electrical devices (typically appliances in a household)’ power consumption trajectories from time series smart meter readings that record the total power consumption profile. While a number of machine learning techniques have been applied to NILM tasks, a new deep learning model called the Diffusion Model (DM) shows great application potential in NILM. This paper provides a preliminary study for evaluating the effectiveness of DM on NILM. Two representative NM variants are trained tested on the UK-DALE dataset, and their performances on NILM are evaluated and are compared with another deep learning-based benchmark model. This paper is expected to provide a reference to the reader of interest on basics of DM and its application in NILM tasks.