基于自然驾驶数据的高速公路跟驰模型参数标定

为了研究中国驾驶人在高速公路上的跟驰行为特征，从上海自然驾驶研究试验数据库中提取48位驾驶人在高速公路上的跟驰事件并进行特征分析。利用自动化筛选准则及人工验证方式提取1 548个有效事件，选取后车车速与车头间距为性能指标，其均方根百分比误差之和为目标函数，利用遗传算法对Gazis-Herman-Rothery模型、GIPPS模型、智能驾驶人模型、全速度差模型和Wiedemann模型进行参数标定及效果验证。基于误差、碰撞及后退等异常情况出现次数等比较其表现性。研究结果表明：不同模型对中国驾驶人的适应性不同，智能驾驶人模型具有最小的误差和误差标准差，更加适合仿真中国驾驶人在高速公路上的跟驰行为。研究结果对于开发适合于中国驾驶人与道路环境特征的跟驰模型具有重要价值。

Calibrating Car-following Models on Freeway Based on Naturalistic Driving Data

To study the characteristics of car-following behavior of Chinese drivers on freeways, this study extracted car-following events of 48 drivers from the experimental database of the Shanghai Naturalistic Driving Study. A total of 1 548 valid events were extracted based on automated filter guidelines and an artificial validation method. In this study, both the following car speed and space were selected as the performance indexes, and the sum of the mean square percentage errors of these indexes was selected as the objective function. Then, a genetic algorithm was used to calibrate and validate the Gazis-Herman-Rothery, GIPPS, intelligent-driver, full-speed difference, and Wiedemann models. The performances of the five models were compared based on the errors and the number of abnormal situations such as collisions and retreats. The results show that different models have different adaptability to Chinese drivers, and the intelligent-driver model has the lowest error and error standard deviation, which is the most suitable for simulating the following behavior of Chinese drivers on freeways. The results are of great value in developing car-following models suitable for Chinese drivers and road conditions.