Abstract: To identify road roughness, four typical neural networks and their application selection, input scheme optimization and evaluation indicators are studied, and a solution scheme is proposed for the input selection and input combination optimization of four typical neural networks. A four DOF plane model for vehicle system vibration is built, with the input and output of neural networks obtained by simulation. 32 input schemes of each neural network are determined by orthogonal experimental design, the evaluation indicators for the input scheme of each neural network under common road grade B and 60 km/h driving speed are obtained, and the optimal input scheme of each neural network and the optimal neural network among four typical ones are selected through variance analysis. The results show that among four typical neural networks, NARX neural network is the optimal one in identifying road roughness with its correlation coefficient and root mean square error of optimal input scheme being 96.75% and 0.003 3 respectively

Key words: road roughness identification, typical neural network, optimal neural network, input scheme optimization