基于交叉熵算法的跟驰模型标定

Calibration of Car-Following Model Based on Cross-Entropy Method

摘要: 跟驰模型的标定是为了更好地重现真实驾驶情况从而增强交通安全和分析如停-走间断流等复杂的交通流情况。然而，跟驰模型的标定并不是一件容易的事。这是因为某些参数是不能从交通流数据中直接观测得到。此外，传统的确定性标定方法会导致大量局部最优值的出现。在此基础上，本文提出了基于交叉熵算法的跟驰模型标定的框架，基于蒙地卡罗与重要样本策略逐步逼近参数的最优概率密度函数。实例分别采用合成数据与实测数据标定智能驾驶模型，验证了交叉熵算法搜索全局最优值的能力，并体现了交叉熵算法适用于实测交通流数据标定的潜能。

Abstract: Calibration of car following models seeks for a more realistic representation of car following behavior in complex driving situations to improve traffic safety and to better understand several puzzling traffic flow phenomena, such as stop-and-go oscillations. However, calibrating these models is never a trivial task. This is caused by the fact that some parameters are generally not directly observable from traffic data. Moreover, conventional deterministic calibration methods always result in a large number of local optima. This contribution puts forward a framework of calibration based on Cross-Entropy Method (CEM), which approaches the optimal probability density function with monte carlo and important sampling strategy. Empirical cases calibrate the intelligent driving model with synthetic data and NGSIM data. The results not only verify the ability of CEM to search global optima, but also confirm the great potential of CEM to adopt into actual traffic measurements.