Publication

Abstract : Multi-zone buildings need precise HVAC system control for flexible demand response, user comfort, and grid connectivity. Evolution Strategy Reinforcement Learning (RLCES), Proximal Policy Optimization (RLC-PPO), Linear Model Predictive Control (MPC-LIN), and Reduced-Order Model Predictive Control (MPC-ROM) are quantitatively compared in this paper. The five-zone thermal building model used ARX dynamics to accomplish these techniques. All controllers were created and tested in MATLAB under identical settings. Their performance was measured by rising time, settling time, overshoot, and Steady State Error (SSE). MPC-ROM performed best overall, with an average settling time of 9.44 seconds, no overshoot, and no SS inaccuracy across all zones, displaying speed and precision. MPC-LIN worked well under linearized settings with an average settling time of 11.01 seconds and 0% SSE. The RLC-PPO controller achieved a stable, slower convergence with an average settling time of 26.87 seconds, an overshoot of 3.89%, and a SSE of 0.0057, without a system model Overall, the MPCROM controller outperformed others, attaining up to 94% faster settling time, no overshoot, and no SS variation. These results prove MPC-ROM is the most effective, resilient, and energy efficient multi-zone HVAC system controller in grid-interactive smart buildings.