DuoBrain-HVAC: A Hybrid Autonomous Agent for Controlling HVAC Systems
Authors : Khalil Al Sayed (LINEACT), Abhinandana Boodi (LINEACT), Roozbeh Sadeghian Broujeny (LINEACT), Karim Beddiar (LINEACT)
Conférence : Communications avec actes dans un congrès international - 26/11/2024 - IEEE 2024 32nd Telecommunications Forum
The HVAC system consumes a significant portion of building energy, contributing to CO 2 emissions. This paper presents DuoBrain-HVAC, a hybrid deep reinforcement learning agent, designed to optimize energy use by controlling a discrete (ON/OFF) air handling unit and a continuous (temperature setpoint) actuator. Simulated in an amphitheater using EnergyPlus, the hybrid agent combines DDQN and TD3 algorithms, achieving faster convergence and reducing energy consumption by 23.68% compared to DDQN and 16.24% compared to TD3.