Subject Training

Context and Problem Statement : Energy management in buildings is a major challenge in the ecological transition, requiring simultaneous assurance of occupant comfort and health. HVAC (Heating, Ventilation, and Air Conditioning) systems must reconcile three often-contradictory objectives: maintaining thermal comfort, preserving Indoor Air Quality (IAQ), and optimizing energy consumption. Traditional control methods struggle to solve this dynamic multi-objective optimization problem, thus paving the way for advanced Artificial Intelligence approaches.

Main Objective : This research aims to design, develop, and evaluate innovative control strategies based on Reinforcement Learning (RL) to manage HVAC systems in smart buildings. The central objective is to solve the dynamic optimization problem involving thermal comfort, air quality, and energy efficiency.

Methodology and Scientific Approach : The scientific approach is structured around three complementary phases. Initially, an in-depth state-of-the-art review will be conducted. This phase will establish a comparative analysis of various RL approaches applied to building control, integrating the potential for coupling with Model Predictive Control (MPC) and reviewing thermal and air-flow modeling techniques. Following this, the modeling and development phase will focus on creating a faithful simulation model of the building's thermal and air-flow behavior. This model will serve as the foundation for designing and implementing RL agents specifically adapted to CVC systems. Various architectures will be explored, including different RL variants, RL-MPC coupling, and hybrid approaches, with specific attention paid to defining robust multi-criteria reward functions to balance competing objectives. Finally, the third phase involves progressive validation. Initial validation will be performed through intensive simulation tests in MATLAB/Simulink to compare the performance of the developed algorithms. This will be followed by an experimental validation on the CRAN's ECOSUR platform, a full-scale real-world HVAC installation. The performance analysis will rely on quantitative metrics (energy consumption, comfort maintenance, air quality) and possibly with qualitative aspects (robustness, stability) to ensure the practical viability of the solution.