Thermodynamic Optimization of Building HVAC Systems Through Dynamic Modeling and Advanced Machine Learning
Samuel Moveh, Emmanuel Alejandro Merchán-Cruz, Ahmed Ibrahim, Zeinab Abdallah Mohammed Elhassan, Nada Mohamed Ramadan Abdelhai, Mona Dafalla Abdelrazig
Abstract
This study enhances thermodynamic efficiency and demand response in an office building’s HVAC system using machine learning (ML) and model predictive control (MPC). This study, conducted in a simulated EnergyPlus 8.9 environment integrated with MATLAB (R2023a, 9.14), focuses on optimizing the HVAC system of an office building in Jeddah, Kingdom of Saudi Arabia. Support vector regression (SVR) and deep reinforcement learning (DRL) were selected for their regression accuracy and adaptability in dynamic environments, with exergy destruction analysis used to assess thermodynamic efficiency. The models, integrated with MPC, aimed to reduce exergy destruction and improve demand response. Simulations evaluated room temperature prediction, HVAC energy optimization, and energy cost reduction. The DRL model showed superior prediction accuracy, reducing energy costs by 21.75% while keeping indoor temperature increase minimal at 0.12 K. This simulation-based approach demonstrates the potential of combining ML and MPC to optimize HVAC energy use and support demand response programs effectively.