Adaptive Control Strategies for HVAC Systems: Integrating Genetic Algorithms with Deep Reinforcement Learning

HVAC systems represent roughly 40 percent of the energy consumed in a building and are vital for building energy efficiency. However, the ability of rule-based methods or PID controllers to handle the dynamic and stochastic nature of the real-world environment is limited. Despite positive results from Deep Reinforcement Learning (DRL) for Adaptive HVAC Control, DRL often exhibits slow convergence, poor exploration, and training instability. This paper introduces a novel hybrid framework based on GAs and DRL to overcome these limitations. Where GAs optimizes initial policies and hyperparameters of the DRL agent to improve exploration and achieve faster convergence. Using the integral output of MPC, the global optimization power of GAs is combined with the adaptive learning power of DRL to provide robust real-time control of HVAC systems. The hybrid approach is evaluated in a simulated, dynamic HVAC environment, with the hybrid approach reducing energy consumption by 18% while maintaining thermal comfort. Finally, this research presents a novel method to create a scalable, adaptive control system for smart buildings using GAs and DRL for energy-intensive systems.