Deep Learning and Reinforcement Learning for Modeling Occupants' Information in an Occupant-Centric Building Control: A Systematic Literature Review

The Occupant-Centric Control (OCC) strategy incorporates occupant information in the building facilities control to improve energy efficiency while maintaining an acceptable level of occupant comfort. Predictive control strategies are necessary to implement OCC in complex systems like HVAC, which pose a significant challenge given the stochasticity of occupant behavior in built environments. Nonetheless, the recent advancements in Machine Learning (ML) and the Internet of Things (IoT) have made data-driven strategies more feasible in OCC of building systems. In this context, Deep Learning (DL) and Reinforcement Learning (RL) techniques have gained significant attention due to their ability to handle large volumes of data and achieve high prediction accuracy. However, the current literature lacks systematic knowledge of algorithm selection in the different OCC contexts. To address this gap, this paper presents a systematic literature review of DL and RL algorithms applied to OCC and provides organized information on the choice of algorithms by classifying occupant information into four levels based on increasing personalization. Subsequently, it identifies the algorithms suitable for each level to establish a systematic foundation for selecting DL and RL algorithms based on the degree of personalization required. The paper also highlights areas for future research in this area.