Hybrid modeling as a virtual sensor concept for automotive HVAC outlet temperature estimation

Efficient thermal management in vehicles is becoming increasingly critical due to stricter energy efficiency requirements and rising passenger expectations for comfort. Conventional heating, ventilation, and air conditioning (HVAC) systems rely on numerous air outlet temperature sensors, which add significant costs at scale. This study introduces a hybrid virtual sensor concept that combines physics-based mathematical modeling with data-driven methods to address this challenge. A detailed physical model predicts air outlet temperatures, while neural networks capture unmodeled heat exchanges and environmental disturbances. Two hybrid strategies are explored: one augments the physical model with a neural network to learn residual heat flows, and the other employs an adiabatic physical model enhanced by a neural network that maps the total environmental heat exchange, supported by a disturbance observer for accurate heat flow estimation. Evaluation using measurement data from a real test vehicle in a simplified two-vent scenario demonstrates that hybrid virtual sensors can provide reliable temperature estimates, reduce reliance on hardware sensors, and maintain high control performance, highlighting the potential of integrating physics-based and machine learning methods in vehicle climate control systems.