基于自然驾驶数据的中国驾驶人城市快速路跟驰模型标定与验证

为了评估既有跟驰模型在仿真中国驾驶人跟驰行为方面的表现,对5种代表性跟驰模型进行参数标定与效果验证。基于"上海自然驾驶研究项目"采集的60位驾驶人、累计超过16万km的实际驾驶行为数据,根据雷达、车辆总线数据自动提取2 100个城市快速路稳定跟驰行为片段;采取5-折交叉验证法划分标定与验证数据集,即将每位驾驶人的50个跟车片段随机划分成5个不相交的子集（每个子集包含10个跟车片段）,其中4个子集作为标定数据集,剩下的1个作为验证数据集,依次轮换标定数据集与验证数据集5次,展开5次模型标定与验证。基于标定数据集,采用遗传算法对Gazis-Herman-Rothery、Gipps、智能驾驶人、全速度差（FVD）以及Wiedemann模型进行参数标定;基于验证数据集,评估5种模型在预测两车间距方面的精度。结果表明：FVD模型在5种模型中表现最佳,具有最小的误差（21%）和误差标准差;相对于微观交通仿真软件VISSIM中所采用的Wiedeman模型,FVD模型具有精度高、易于标定、对不同驾驶人鲁棒性强3个优势,更加适应于仿真中国驾驶人的跟驰行为。研究结果对于开发适合于中国驾驶人与道路交通环境特征的跟驰模型及微观交通仿真系统具有重要价值。

Calibrating and Validating Car-following Models on Urban Expressways for Chinese Drivers Using Naturalistic Driving Data

To evaluate the performance of existing car-following models when applied to Chinese drivers, five representative car-following models were calibrated and validated. Based on 60 Chinese drivers' driving data with a total mileage of 16×10⁴ km collected in the Shanghai Naturalistic Driving Study, 2 100 stable urban-expressway car-following periods were automatically extracted using radar and vehicle controller area network (CAN) data. A 5-fold cross validation technique was used to generate calibration and validation datasets. Specifically, the 50 car-following periods for each driver were randomly divided into five equal subsets; therefore, the model calibration and validation processes were repeated five times. In each iteration, four subsets were used to calibrate the car-following model and the remaining subset was used to conduct intra-driver validation. Based on the calibration dataset, values of parameters were calibrated using genetic algorithms for the Gazis-Herman-Rothery, Gipps, intelligent driver, full velocity difference, and Wiedemann models. The performance of these models on predicting inter-vehicle spacing were then validated with validation datasets. The results showed that the full velocity difference (FVD) model had the lowest error term on the validation dataset (21%), and the smallest standard deviation of errors. Compared to the Wiedemann model used by VISSIM, the FVD model is more easily calibrated and demonstrates a higher and more robust performance, justifying its suitability to be applied for microscopic traffic simulations in China. These results would be valuable for developing intelligent vehicles and microscopic traffic simulation tools tailored to the characteristics of Chinese drivers as well as road and traffic environments in China.