A Transfer Reinforcement Learning Framework for Smart Home Energy Management Systems

Abstract

Transfer learning (TL) is widely used as a solution to overcome the huge amount of time required to train a deep learning model. However, a number of challenges are associated with transferring the knowledge between the same and different target domains (TDs), for instance, designing efficient mapping functions, correlating the source and target datasets, identifying necessary parameters for enabling TL, and so on. In this article, we, therefore, present a solution to enable TL for home energy management systems (HEMSs) in smart homes consisting of multiple residents and appliances. This led to eliminating the training time and data requirements associated with training home appliances. A mapping function is designed based on correlating the source and target datasets before transferring the knowledge. Furthermore, a training model is designed to iteratively train a source model, i.e., expert home (EH) and expert appliance (EA), with source dataset and feedback knowledge from the target model, i.e., learner home (LH) and leaner appliance (LA). An extensive set of simulations is carried out to evaluate the performance of the proposed scheme. Simulation results show that transferring knowledge in both the same and different domains significantly reduces the energy consumption of individual home appliances and smart homes. In the future, we will expand the research using new models to promote TL effectiveness.