A systematic review of reinforcement learning application in building energy-related occupant behavior simulation

The building and construction industry has consistently been a major contributor to energy consumption and carbon emissions. With stochastic interactions between occupants and building energy systems, the operational phase has emerged as the most energy-intensive stage throughout a building's lifecycle. In the face of challenges posed by traditional control strategies, reinforcement learning distinguishes itself with its capacity to effectively characterize dynamic occupant-building interactions and iteratively generate optimal control strategies through self-learning. However, the valuable knowledge about reinforcement learning applications scattered in the growing body of occupant behavior research has not been systematically integrated. As a result, this research innovatively combines bibliometrics and content analysis to provide a systematic review aiming to answer how reinforcement learning can be effectively adopted in occupant-centric building energy management based on existing knowledge. The state-of-the-art developments in related domains and a taxonomy of reinforcement learning are introduced based on the dissection of core concepts in occupant behavior and reinforcement learning. Subsequently, a novel framework for the complete application process of reinforcement learning in the domain of occupant energy-related behavior is originally proposed, providing potential researchers with rich existing knowledge and guidance. Moreover, future breakthroughs in framework design, data collection, agent training, and deployment are inspired through systematically concluded future directions, thus laying a solid foundation to better satisfy occupant comfort with optimal energy efficiency.