Application of machine learning-based thermal comfort model to optimize vehicle air-conditioning control strategies

Optimizing control strategies is crucial for improving the performance of vehicle air-conditioning (AC). To address limitations of thermal comfort variations accuracy across different regions in the mathematical parameter model system, the study introduced a validated 3D model of the heat load in the cabin and coupled it with vehicle AC to generate accurate driver predicted mean vote (PMV) values. Comprehensive AC parametric analysis establishes design rules for a multi-stage collaborative control strategy of compressor speed and evaporator blower airflow (MCS-EBA). The strategy parameters were then optimized based on a genetic algorithm (GA). In this process, a high-precision PMV prediction model was used to circumvent lengthy CFD simulations in the coupling model. The predictive model employed a recurrent neural network (RNN) to learn 60 datasets, which were derived from the mapping relationship between the thermal environment parameters of the cabin and the PMV of the driver under different cooling scenarios in the coupled model. Results demonstrate that the optimized MCS-EBA saves energy consumption by 43.7%, 5.8%, and 5.0% and reduces RMSE (vs. PMV=0) by 84.7%, 21.4%, and 73.1%, compared to the ON/OFF, PID, and MCCS, respectively. The technical approach of this strategy holds significant engineering application value. • We pioneered the integration of a 1D vehicle AC model with a 3D cabin thermal model, overcoming the limitations of 1D models in accurately simulating interior temperatures. • An innovative strategy aims to achieve thermal comfort and personalized preferences for passengers in different seats while reducing vehicle AC energy consumption. • An RNN-based PMV prediction model was developed to circumvent the complex CFD calculations required for coupled models during optimization strategies.