LAUSR

Over the last few years, the development of information and communication technologies has provided a great opportunity for the residential sector to take part in demand response (DR) programs in smart grids (SGs). Optimal load scheduling via home energy management systems (HEMSs) is a typical technique used to reduce the power consumptions during the DR events. One of the major challenges faced by the HEMS manufacturers and the electric utilities is the lack of an accurate yet convenient tool for predicting the power consumptions of residential homes, particularly the air conditioners, for decision-makings. The aim of this paper is to develop an accurate self-learning grey-box room thermal model and use it to investigate DR potentials of residential air conditioners (ACs). The readily available indoor air and outdoor air temperatures in today’s HEMSs are used to train the room thermal model. The model parameters are pre-estimated and scaled to improve the optimization accuracy and computational efficiency. Three optimization techniques including trust region algorithm (TRA), genetic algorithm (GA) and particle swam optimization (PSO) are employed to identify the model parameters separately and their performances are compared. A case study shows that the room thermal model can accurately predict the indoor air temperature profile. Two types of DR strategies of residential ACs, i.e. temperature set-point reset and precooling, are then tested using the room thermal model and a simplified air conditioner energy model. Simulation results show that temperature set-point reset combined with precooling strategy can result in more than 26% power reduction during the DR hours on a typical summer day in Hong Kong, without significant change of thermal comfort.