Abstract: For applying NARX neural network to identify road roughness, NARX neural network and its training process and structure design are analyzed. Correlation coefficient and root mean square error are used as evaluation indicator of identification effect of NARX neural network. The white noise model of road roughness and four DOF planar model for vehicle ride comfort are established, on which a simulation is conducted to get the road roughness and vehicle responses. With measurable vehicle responses as the input of NARX neural network, a method for determining the input scheme of NARX neural network is proposed by using orthogonal experimental design, and the road roughness for the front wheel of a car running on a common-grade road with common speed is identified. The results show that with measurable vehicle responses as the input of NARX neural network,combined with orthogonal experimental design can solve the problem of determining the optimal input scheme of NARX neural network

Key words: road roughness identification, NARX neural network, orthogonal design of experiment, ride comfort