Effective dynamic prediction of air conditions within car cabin via bilateral analyses of theoretical models and artificial neural networks

Abstract

Dynamic prediction of air conditions within a car cabin is significant to technology design and usage for thermal comfort, air quality, and energy consumption of vehicles. In this paper, theoretical models are completely derived from mass and energy balances of moist air and carbon dioxide from an outdoor environment via vehicle envelope to car cabin. However, it is demanding to determine genuine modeling hypothesis and/or physical properties of car cabins for accurate prediction. Regardless of those requirements, artificial neural networks can be applied as universal models, which are derived from numbers of input/output data from real dynamic systems. Without any experimental efforts, the input/output data is generated from the theoretical models under various conditions. A few input/output data from experiments are combined for making prediction close to actual behaviors. With bilateral analyses, the artificial neural networks are trained effectively to simulate dynamic behaviors of air, moisture, and carbon dioxide within a car cabin. This viability of the proposed methodology is confirmed by that the averaged coefficients of determination to the perfect prediction under parking and driving conditions of a sedan car are R²=0.9394, and R²=0.9314, respectively. The numbers of experimental data for training are 3.5% of total numbers of training data.