A Normalized Approach for Evaluating Driving Styles Based on Personalized Driver Modeling

摘要

Driving style may affect fuel economy and driving ability, and therefore, its evaluation becomes an important research topic for vehicle calibration and control. In this work, we propose to evaluate the driving style by normalizing driving operations in a standard test procedure. Firstly, a personalized driver model is established for each driver by learning his/her driving operations during real-world driving. This is accomplished by using the locally designed neural network, i.e., CMAC in this work, and the real-world vehicle test data (VTD). Secondly, the established driver model is applied to speed control as required by standard test procedure, i.e., FTP-75, thus the driving operations may be normalized. Finally, the energy spectral density (ESD) is computed on normalized driving data to obtain a quantitative index for evaluating the driving style of each driver. Simulations are conducted to verify the effectiveness of the proposed scheme.