سرور مؤيد داوود الصالح

The Heating, Ventilation, Air Conditioning, and Refrigeration (HVAC&R) system is a complex, nonlinear behavior with a high uncertainty control system that equips the thermal comfort desired but consumes significant electrical energy and costs in different types of buildings, such as residential, commercial, and industrial. This paper introduces a new approach for online controlling of HVAC&R systems using model-based reinforcement learning (MB-RL) style to diminish energy usage and energy cost, maintain the occupants’ comfort levels by controlling the buildings’ indoor temperature, and maintain the desired carbon dioxide levels simultaneously. For this purpose, a new model based on energy and mass conservation laws is presented to model the dynamic variations of temperature and CO2 concentration levels. The HVAC&R system control trouble is defined as a specific Markov Decision Processes (MDPs) model. The reward function balances the ability to increase energy conservation while preserving the interior comfort requirements of occupants. Employing the deterministic policy algorithm (DP), the proposed methodology can manage the dimensionality curse problem due to increased state-action space. Then, it overcomes the nonlinearity and the control system uncertainty. The MB-RL algorithm, which uses a unique DP called DP-MB-RL, can select the best decisions instead of stochastic policy to reduce the calculation time. A real case, a building in Basra City, Iraq, is simulated using MATLAB software. Devoting the MB-RL and DP-MB-RL techniques to online control of an HVAC&R system, the simulation results for both methods are provided. For instance, the parameters, like electrical power, internal comfort levels, energy consumed, and energy cost at different pricing schemes, such as fixed pricing (FP), timeofuse (TOU), and real-time pricing (RTP), are assessed. The results indicated that the suggested DP-MB-RL methodology had better indoor thermal and air quality satisfaction levels, energy-saving (more than 15%), and reduced the cost of electricity by more than 15%, 13%, and 10% for FP, TOU, and RTP pricing schemes, respectively, compared to the benchmark MB-RL style controller. The DP-MB-RL controller also performed better than the Takagi-Sugeno Fuzzy (TSF) controller for the same building, saving more than 21% energy. © 2025, Tikrit University. All rights reserved.