考虑分布不确定性和相关性的跟驰行为建模及仿真验证

Car-following Behavior Modeling and Simulation Verification Considering Distribution Uncertainty and Correlation

分析跟驰模型的参数特征有助于提高微观交通仿真的准确性和可靠性。为同时考虑跟驰模型参数的分布不确定性和相关性,提出一种基于贝叶斯模型平均Copula(Bayesian Model Averaging Copula,BMAC)的跟驰模型参数建模方法,以刻画跟驰行为特性。基于HighD自然驾驶数据集提取跟驰事件轨迹,并对Gipps模型、智能驾驶人模型(Intelligent Driver Model,IDM)、全速度差模型(Full Velocity Difference Model,FVDM)和纵向控制模型(Longitudinal Control Model,LCM)的模型参数进行标定。采用BMAC方法对参数标定结果进行统计建模,该方法通过集成了不同分布模型的优势,能够克服分布不确定性的问题并捕捉参数相关性;并进行了一系列的重复性数值仿真来验证BMAC方法的有效性。试验结果表明:跟驰模型的参数中存在分布不确定性和相关性,而BMAC方法能够捕捉参数相关性并解决参数分布不确定性的问题,从而更准确地描述跟驰行为特性。基于BMAC的参数采样方法(记为BCCO)取得了最佳的仿真效果,在4种跟驰模型下的碰撞时间、速度和间距的Kolmogorov-Smirnov(K-S)统计量均值皆最小。以LCM模型为例,碰撞时间、速度和间距的K-S统计量均值分别为0.231、0.310和0.294。因此,所提出的基于BMAC参数采样方法能够有效提高微观交通仿真的真实性,为交通管理与控制策略提供准确的仿真评价结果。

The accurate analysis of the parameter characteristics of car-following models improves the accuracy and reliability of microscopic traffic simulations.To simultaneously consider the distribution uncertainty and correlation of car-following model parameters,this study proposes a modeling method based on the Bayesian model averaging Copula(BMAC)to describe car-following behavior characteristics.Car-following event trajectories were extracted from the HighD naturalistic driving dataset and the parameters of the Gipps model,intelligent driver model(IDM),full velocity difference model(FVDM),and longitudinal control model(LCM)were calibrated.The BMAC approach was employed to model the parameter calibration results statistically,and the advantages of various distributions were leveraged to overcome the issue of distribution uncertainty and capture parameter correlations.In this study,a series of repeated numerical simulations were conducted to validate the effectiveness of the BMAC approach.The experimental results demonstrate that distribution uncertainty and correlation exist among the parameters of the car-following models.However,the BMAC approach can capture parameter correlations and address the issue of parameter distribution uncertainty to provide more accurate descriptions of car-following behavior characteristics.The parameter sampling method based on BMAC(denoted as BCCO)achieved the best simulation performance,achieving the lowest average Kolmogorov-Smirnov(K-S)statistics for the time to collision,speed,and spacing among the four car-following models.Taking the LCM as an example,the average K-S statistics for the time to collision,speed,and spacing were 0.231,0.310,and 0.294,respectively.Therefore,the proposed BMAC-based parameter sampling method effectively enhances the realism of microscopic traffic simulations and provides accurate simulation evaluation results for traffic management and control strategies.