High-Precision Dynamics Modeling of Electric Vehicle Based on Hybrid Physical and Data-Driven Methods

Vehicle modeling is a key technology in the field of vehicle dynamics control, and high-precision vehicle modeling not only accelerates the iteration of dynamics control algorithms during the simulation development phase, but also improves the accuracy of model-based vehicle dynamics control methods. In recent years, the introduction of data-driven methods in vehicle dynamics has provided new ideas for high-precision dynamics modeling. In this paper, we first build vehicle 7DOF model based on physical method, then construct longitudinal and lateral tire model based on the Burckhardt formulation and brush model. In order to improve the modeling precision under extreme operating conditions, an error compensation black-box neural network is introduced for the compensation of lateral response during large lateral slip operating conditions. Finally a hybrid model authentication architecture is proposed to combine the physical model and black-box neural network. The hybrid model is validated on both Carsim dataset and real vehicle test dataset, and the accuracy of the hybrid model is improved by 20%-50% compared with the physical model under different extreme operating conditions.