ARTIFICIAL INTELLIGENCE STRATEGIES APPLIED IN GENERAL AND AUTOMOTIVE AIR CONDITIONING CONTROL. A REVIEW OF THE LAST 20 YEARS{fr}STRATÉGIES D'INTELLIGENCE ARTIFICIELLE APPLIQUÉES AU CONTRÔLE GÉNÉRAL ET À LA CLIMATISATION AUTOMOBILE. UNE REVUE DES 20 DERNIÈRES ANNÉES

Abstract

This article conducts a 20-year review of studies investigating the utilization of artificial intelligence in HVAC (Heating, Ventilation, and Air Conditioning) system control, with a particular focus on automotive HVAC systems, renowned for their unique challenges due to the highly dynamic nature of automotive cabins. Artificial neural networks (ANNs) provide adaptive modeling and control capabilities to enhance energy efficiency, thermal comfort, and air quality in HVAC systems. Nevertheless, ANNs exhibit limitations, primarily their performance being constrained by training data, impeding adaptability to unforeseen scenarios. To overcome these constraints, researchers have explored hybrid approaches that integrate ANNs with PID (Proportional, Integral, and Derivative) or PI (Proportional and Integral) algorithms, resulting in systems referred to as Neuro-PID or Neuro-PI. These systems can effectively handle unanticipated situations by absorbing errors introduced by ANNs. Nonetheless, the applicability of Neuro-PID varies depending on the specific system. It becomes infeasible in scenarios with substantial delays, prompting the choice of the appropriate approach based on the controllability factor, a ratio between system delay and time constant. A controllability factor less than 2 indicates the suitability of employing ANNs for PID controller tuning.Furthermore, the intricate interaction between temperature and humidity in HVAC control systems diminishes the effectiveness of traditional PID. To proficiently regulate both variables, a viable alternative is the amalgamation of ANNs with specific control logic or the deployment of distinct ANN algorithms for varying scenarios.The heightened variability and unpredictability within automotive HVAC systems, driven by continuous vehicle movement and door/window operations, present additional challenges. In such contexts, reinforcement learning techniques have found application, allowing systems to dynamically adapt to unforeseen events.In summary, the choice of HVAC control approach hinges on the primary goal—thermal comfort or specific variable control—as well as the system's noise level, controllability factor, and system nature. While ANNs hold significant potential, combining them with other techniques, such as PID, reinforcement learning, or specific control logic, may prove essential in addressing the intricacies of HVAC systems