Multi objectives reinforcement learning for smart buildings: A systematic review of algorithms, applications and future perspectives

Abstract

With the rapid advancements in digital technologies, changes in regulatory and societal expectations, and increased environmental awareness and concerns among building owners and occupants, the design and effectiveness of building control systems and their energy usage are under constant scrutiny as never before. The reshaping and integration of building controls with Internet of Things (IoT) devices have led to the growing popularity of Reinforcement Learning (RL) in the built environment. Multi-objective reinforcement learning (MORL) is touted to be more effective than traditional RL in optimizing smart building operations by resolving multifaceted goals, improving policy adaptability and decision-making processes involving multiple stakeholders and criteria. Hitherto, little is known of the full potential of MORL and its application trends. In addressing this, this research aims to build a knowledge base around the application trends of MORL framework and its benefits for smart building energy design and control systems through a systematic and critical review using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. A total of 1071 studies were retrieved, of which 74 studies were included in the final assessment to present and discuss: (i) objectives RL typically in smart building context; (ii) overview of the design and control strategies of MORL in smart buildings; (iii) MORL applications and performance evaluation in smart building; and (iv) challenges and future research directions and opportunities. Overall, our findings reveal potential work done to explore the use of MORL towards controlling multiple policies and complex dynamic building environments, and that current studies tend to focus on incorporating occupancy patterns and/or occupant feedback into the MORL control loop.