关键词: 自动驾驶汽车；乘坐舒适性；BP 神经网络；舒适性评价系统

Abstract: Aiming at the problems of prominent subjectivity in the ride comfort evaluation of autonomous vehicles on urban roads and insufficient engineering adaptability of quantitative models, this paper constructs a human-vehicle-road coupled evaluation index system, and conducts research on real-vehicle tests, machine learning modeling and the development of an automated evaluation system. Taking the second ring viaduct in Kunming as the test section, a multi-source data synchronous acquisition platform is built to obtain 3 672 sets of valid samples. The validity of the indexes is verified through reliability and correlation analysis, the ridge regression method is adopted to solve the multicollinearity of parameters and determine the index weights, and an optimized back propagation (BP) neural network evaluation model is constructed. An evaluation system is developed based on the hybrid programming of LabVIEW and MATLAB, with its functions and accuracy verified comprehensively. The results show that lateral acceleration is the core index affecting ride comfort, and the model achieves a classification accuracy of 82.45% for the five ride comfort grades. The research realizes the automation, standardization and visualization of ride comfort evaluation, and provides a core model and engineering tools for the optimal design, test and certification of the ride comfort of autonomous vehicles.

Key words: autonomous vehicles; ride comfort; BP neural network; comfort evaluation system