基于PINN的路面附着系数估计方法

Road adhesion coefficient, as one of the indicators for assessing the amount of friction between road surface and tires, influences the decision control strategies for autonomous driving systems. For the problem of road adhesion coefficient estimation, this paper proposes a road adhesion coefficient estimation method based on physical information neural network (PINN). Firstly, the vehicle dynamics analysis is carried out to construct a seven-degree-of-freedom vehicle dynamics model. Based on the Dugoff tire model, the accuracy of the nonlinear region of the model is improved to obtain an improved two-degree-of-freedom dugoff tire model. Then, the relationship between the kinetic parameters and the road adhesion coefficient is analyzed to determine the input to the neural network, and the network model for road adhesion coefficient estimation is constructed by combining the attention mechanism and the long and short-term time series model. Finally, based on vehicle model and tire model, the vehicle dynamics equations are introduced to construct the physical constraint loss function, and the PINN network is constructed. The experiments show that the introduction of the loss function with physical model increases the convergence speed of the model compared to the model without physical constraints, and reduces the mean absolute error by 44.68％ and the root mean square error by 39.87％.