Abstract: The recent increase of renewable energy technology in the building industry is predicted to cut fossil fuel usage while increasing the complexity of heating, ventilation, and air conditioning (HVAC) system design and control. Due to population increase and ongoing efforts to raise living standards, research into energy saving and sustainability has grown popular. In both commercial and commercial buildings, the installation of central HVAC systems is generally on the rise. The HVAC systems in commercial buildings have been looked at as a potential demand response resource. The complexity of creating warm-powerful models and the vulnerability related with both occupant driven heat loads and climatic projections make it far from easy to improve commercial air cooling across the board. In this research, we develop a perfect control philosophy for a commercial cooling system with several zones using a sharp sans model substantial RL technique termed the profound deterministic arrangement slope (DDPG), determined to diminish energy costs without adversely influencing solace levels for building tenants. Through unsupervised, ongoing contact with a simulated building environment, the utilized deep RL-based method gathers information. When the DDPG-based HVAC control strategy is compared to the linear-based HVAC control strategy, the converge may be reduced by 56%, and when the DDPG-based HVAC control strategy is compared to the linear reinforcement model, the converge may be reduced by 15%. Also mean steps required for DDPG RL model and Linear RL model is 9.9 and 115.3 respectively.

Actor-critic learning, deep deterministic policy gradient (DDPG), deepreinforcement learning (deep RL), multi-zone commercial HVAC. Linear RL, Reinforcement Learning.