Abstract:

In order to improve the accuracy and reliability of drivability evaluation method, a subjective and objective comprehensive evaluation method combining limit gradient lifting algorithm and Shapley interpretation algorithm is proposed. In this paper, the starting condition of the vehicle is taken as the research goal, and 9 objective evaluation indexes are defined under the starting condition. In this paper, the limit gradient enhancement algorithm is used to evaluate the objective index value and subjective score of bidirectional mapping, and Shapley interpretation algorithm is used to attribute the features of the mapping model, and a comprehensive evaluation model of driving performance is constructed with both prediction accuracy and interpretability. This method is applied to many road tests, and several comprehensive driving evaluation methods are compared and analyzed. The results show that: Compared with mainstream driving evaluation algorithms such as BP neural network, random forest and extreme learning machine, the mapping accuracy of the proposed method is significantly improved. Moreover, the proposed comprehensive evaluation method is interpretable to a certain extent, and has reference significance for subjective and objective comprehensive evaluation of driving evaluation.

Key words: drivability, starting condition, evaluation method, limit gradient lifting algorithm, Shapley explains the algorithm