Objective Evaluation for the Driving Comfort of Vehicles Based on BP Neural Network

Abstract

Driving comfort, which is mainly influenced by vibration and shock, is an essential factor to evaluate the performance of intelligent vehicles. The evaluation methods of driving comfort mainly contain subjective and objective evaluation. Subjective evaluation is time-consuming, expensive and sensitive to personal feelings. And objective evaluation is difficult to exactly define the relationship between objective parameters and driving comfort. In order to combine the advantages of subjective and objective evaluation, a neural network that adopt objective indicators as input and subjective ratings as output was established for evaluating driving comfort. First, a road test with about 9000 km was conducted and key parameters of vehicle status were recorded, as well as subjective ratings. Secondly, 25,165 segments were extracted from the naturalistic driving data. Then, total weighted root-mean-square accelerations of all segments were computed according to ISO 2631–1997 Standard. And the result shows that the comfort levels calculated by weighted root-mean-square accelerations cannot match the subjective ratings given by professional evaluators very well. Finally, a 20-128-256-256-128-6 BP neural network was established and trained. And the accuracy of evaluation based on neural network is better than evaluation based on weighted root-mean-square value. The result reveals that it is feasible to establish a neural network model based on collected naturalistic driving data to evaluate the driving comfort of vehicles.