A Community Energy Management Method Based on Explainable Reinforcement Learning

The construction industry is responsible for a substantial portion of final energy consumption (36%) and carbon dioxide emissions (39%), and this share is likely to rise as urbanization increases. This paper presents this model, which is based on HVAC (heating, ventilation, and air conditioning) systems. Utilizing reinforcement learning (RL), one can ascertain the most efficient strategy to reduce HVAC energy usage while ensuring the thermal comfort of inhabitants. Because neural networks are complex and unpredictable, which impede the broad acceptance of artificial intelligence in the power industry. Explainable Artificial Intelligence (XAI) is used to explain the decision-making process in deep reinforcement learning, HVAC systems are major energy users in buildings, and the switch to mixed ventilation modes has resulted in a more than 40% decrease in cooling device energy usage when compared to traditional methods. The energy transfer dynamics within HVAC systems can be delineated as a Markov Decision Process, with the objective of making informed choices across various states to maximize long-term benefits. The exploration of agents’ behavioral policies can be enabled through the utilization of the Multi Agent Deep Deterministic Policy Gradient algorithm (MADDPG). Experimental findings indicate that the RL-driven HVAC energy management approach is adept at adapting to diverse environmental conditions, effectively curbing energy usage, and enhancing comfort levels compared to conventional optimization control strategies.

I. Introduction

The construction industry and buildings contribute to around one-third of worldwide energy consumption, a proportion that is projected to increase as urbanization advances [1]. It is crucial to prioritize energy efficiency in the building sector, ensuring a balance between comfort and environmental sustainability through the use of advanced technology. The main energy consumers in buildings, heating, ventilation, and air conditioning systems, usually use either traditional natural ventilation or partially mechanical ventilation techniques. Natural ventilation depends on natural airflow, which may not meet specific requirements for air velocity and ventilation rates, leading to energy wastage. Traditional mechanical ventilation methods require additional energy to operate ventilation equipment, resulting in relatively low energy efficiency [2].