A Time Efficient Factorial Hidden Markov Model-Based Approach for Non-Intrusive Load Monitoring

Assessment of energy consumption behaviour plays an important role in designing demand reduction programs by utility companies. Knowledge of appliance activities in a household aids in conducting the energy consumption behaviour assessment for a community load. Non-intrusive load monitoring (NILM) is a tool that can help in identifying the appliance activities. In this paper, a Modified Factorial Hidden Markov Model (MFHMM) based NILM framework is proposed, which models dependencies among appliance operating states and differential appliance operating states by considering differential in power consumption profiles over time. All the appliances are modelled as individual load models using the Hidden Markov Model (HMM). The appliance operating states are obtained with the application of an iterative k-means clustering algorithm. The aggregated power consumption profile is divided into segments using an optimization-based change-point detection (CPD) algorithm. The NILM problem is solved for each of the segments, and the obtained solution is corrected based on the voltage profile at the aggregated load point. The approach of segmentation and efficient identification of appliance operating states make the model less time complex. Simulations are carried out on publicly available datasets named AMPds, REDD, and UK-DALE. The efficacy of the proposed framework over existing frameworks is evident from the simulation results.