The damping characteristics of the shock absorbers in high speed trains changes greatly in varying temperature environment. Establishing a mechanical model which can represent temperature dependent damping characteristics of shock absorbers is of great significance for the analysis of vehicle dynamics. This paper establishes a LM-BP neural network model for hydraulic shock absorbers based on the digital-physical fusion approach. Experimental investigations on the damping characteristics of a shock absorber are conducted under varying temperature conditions. A back propagation neural network with optimized Levenberg Marquardt algorithm was established to extract the nonlinear mathematical representations between temperature, exciting frequency, and damping characteristics, and then a mechanical model was achieved that can characterize the nonlinear damping characteristics of shock absorbers with varying temperature and exciting frequency. The results show that the established model can accurately map the damping characteristics of the hydraulic shock absorber and the vehicle dynamics simulation in changing temperature environments was achieved through the SIMAT (SIMPACK and Matlab) joint interface.