Lifecycle-adaptive AI-driven neural thermal modeling and optimal control for vehicle air conditioning

Traditional thermodynamic vehicle air conditioning models rely on iterative computation and suffer from poor real-time prediction performance on vehicle onboard controllers. Conventional data-driven models have weak generalization capacity and fail to adapt to seasonal variations, extreme high ambient temperature and slow component performance degradation over the full service life of new energy vehicles. To solve this issue, this paper puts forward an AI-enabled adaptive digital twin system tailored for full-lifecycle thermal management. A high-precision neural network substitutes complex thermodynamic iteration modules to rapidly predict evaporator outlet air temperature and total cooling energy consumption. An online twin updating framework fusing factory baseline calibration data and accumulated real vehicle historical data is established, and the model performance is continuously optimized through periodic iterative retraining. Test results show that compared with the static baseline BP neural network, the prediction accuracy of temperature and energy consumption rose by 67 % and 87 %, respectively. This system effectively identifies external environmental disturbances and internal equipment performance drift and maintains stable full-lifecycle adaptive capacity. A lifecycle-oriented dynamic programming (LC-DP) strategy considering solar radiation thermal load is further proposed to cut energy consumption under standardized PMV thermal comfort constraints. Validations under near-real driving cycles demonstrate that LC-DP reduces air conditioning energy use by 4.4 % compared with PID control and 41.3 % compared with traditional on–off control. This research provides targeted technical references for the practical engineering deployment of intelligent thermal management systems for new energy passenger vehicles.