Prospects and challenges of reinforcement learning- based HVAC control

Increasing worldwide energy demand and the resulting escalations in greenhouse gas emissions require a reassessment of energy usage in many sectors. The building industry, which accounts for more than 40 % of the world's energy consumption and 30 % of its greenhouse gas emissions, highlights how urgent it is to find sustainable and energy-efficient solutions. Optimization of HVAC systems has since become crucial, as these systems are part of the primary source of energy consumption in buildings. The adaptability and robustness of traditional HVAC control systems, such as model predictive control (MPC) and rule-based control, are limited, which has led to the investigation of alternative strategies. This study comprehensively explores diverse HVAC control methodologies, from traditional techniques to cutting-edge machine learning-driven models. The analysis details the strengths, weaknesses, and applications of these control methods, emphasizing the pursuit of energy efficiency and optimal occupant comfort. Notably, reinforcement learning (RL) control distinguishes itself with its adaptive nature, effectively responding to dynamic environmental changes. However, the validation of RL models predominantly focuses on residential and commercial buildings, with limited attention given to industrial structures. Reviewing performance trends reveals that value-based RL methods perform better in residential settings, while actor-critic RL approaches demonstrate superiority under commercial settings- settings based on the findings of our study, future RL-based HVAC control directions needed for further exploration were discussed.