Abstract
Heating, ventilation, and air conditioning (HVAC) systems are major contributors to residential energy consumption. However, most homes continue to rely on single zone thermostat control, which regulates temperature based on a single sensor and cannot account for thermal variations across multiple rooms. This often leads to uneven thermal conditions and inefficient energy use in multi-zone residential buildings. This thesis presents a deep reinforcement learning based approach for improving HVAC zoning control through dynamic airflow distribution. A physics based multi zone thermal model of a residential house was developed to simulate heat transfer processes including conduction, convection, solar and internal heat gains, infiltration, and moisture dynamics. The model was integrated with a Deep Deterministic Policy Gradient (DDPG) reinforcement learning agent implemented in MATLAB to learn optimal airflow allocation across six thermal zones using continuous control actions. The reinforcement learning controller observes the thermal state of the building, including zone temperatures, humidity ratios, external weather conditions, and time of day information, and determines damper positions and total air handling airflow to maintain thermal comfort while minimizing unnecessary airflow usage. The control framework was designed for practical deployment and integrated with motorized dampers and zone level sensing in a real residential test house. Simulation and experimental evaluation compared the reinforcement learning controller with a conventional ON/OFF thermostat baseline. Results demonstrate significant improvements in both comfort and energy performance. The DDPG algorithm reduced average temperature deviation across zones from approximately 2.14 °F (1.2 ℃) under thermostat control to 0.63 °F (0.35 ℃) while improving airflow utilization efficiency. The reinforcement learning controller also achieved median daily HVAC energy savings of approximately 61% relative to the baseline thermostat operation. These results demonstrate that reinforcement learning can effectively learn airflow control strategies that improve temperature uniformity and reduce energy consumption in multi zone residential HVAC systems. The study highlights the potential of intelligent control approaches for next-generation smart building automation and provides a framework for future research on adaptive HVAC control using reinforcement learning.
Date of publication
Spring 3-26-2026
Document Type
Thesis
Language
english
Persistent identifier
http://hdl.handle.net/10950/5041
Committee members
Mohammad Biswas, Nael Barakat, Hayder Abdul-Razzak
Degree
Master of Science in Mechanical Engineering
Recommended Citation
Alghalayeeni, Mohammad H. Z., "DEEP LEARNING BASED CONTROL SYSTEM FOR MULTI-ZONE HVAC RESIDENTIAL BUILDINGS" (2026). Mechanical Engineering Theses. Paper 44.
http://hdl.handle.net/10950/5041
Included in
Acoustics, Dynamics, and Controls Commons, Controls and Control Theory Commons, Electro-Mechanical Systems Commons, Energy Systems Commons, Heat Transfer, Combustion Commons, Other Engineering Commons, Other Mechanical Engineering Commons, Thermodynamics Commons