The Generalization Ability of the Tire Model Based on Bayesian Regularized Artificial Neural Network

This paper aims to develop a practical tire model that keeps the balance between generalization ability and accuracy for Formula Student application. Up to now, the advantages of Artificial Neural Networks for tire modelling have been investigated by some studies, which are briefly introduced in this paper. However, tire models based on Artificial Neural Networks were likely to over-fit the given data, or were sensitive to the noise. And far too title attention has been paid for the generalization ability, which is essential for a tire model. In this paper, a Bayesian regularization method based on the Bayes’ theorem is proposed to solve the major problems described above by improving the generalization ability. And a large number of measured data were used for testing the trained models with different configurations. The results show that the tire models based on the Bayesian regularized artificial neural networks can achieve better generalization ability, and are practical for racing applications such as Formula Student.