Residential Appliance-level Consumption Modeling and Forecasting via Conditional Hidden Semi-Markov Model

Recently, residential demand-side management has attracted great attention due to the development of the smart grid and the increasing penetration of renewable resources. Appliance-level load monitoring and forecasting are critical for load flexibility analysis and energy efficiency enhancement. This paper studies and explores the potential ways to improve the performance of the Conditional Hidden Semi-Markov Model (CHSMM) for the appliance-level demand modeling and forecasting problem. Classification and regression-based training methods are tested in CHSMM for different appliances. The case studies show that the performance of CHSMM can be enhanced by selecting a proper training method based on the characteristics of the appliance. The computational burden is discussed to ensure that training an appliance consumption forecasting model is within a trackable time. Moreover, the size of needed storage memory can be reduced by considering a portion of the training data if the consumption forecasting period is the same period as the training data.