A multi-objective calibration framework for capturing the behavioral patterns of autonomously-driven vehicles
A multi-objective calibration framework for capturing the behavioral patterns of autonomously-driven vehicles
复制标题
DOI：
10.1016/j.trc.2023.104151
复制
发表时间：
2023-07
期刊：
影响因子：
--
通讯作者：
Shi-Teng Zheng;M. Makridis;Anastasios Kouvelas;Rui Jiang;B. Jia
中科院分区：
文献类型：
--
作者：
Shi-Teng Zheng;M. Makridis;Anastasios Kouvelas;Rui Jiang;B. Jia
跟驰（CF）模型的校准被认为是再现个体车辆行为和集体交通现象的一项非常重要的任务。在大多数工作中，对前车和后车的轨迹进行速度或间距量的校准，根据文献，两者的结合加上加速度量被认为是最合适的。随着自适应巡航控制 (ACC) 技术的出现，人类驾驶的车辆和 ACC 驱动的车辆之间存在内在的行为差异，这在实验观察中显而易见。一些例子包括启用 ACC 的车辆的恒定行驶速度、绳索不稳定、迟滞行为、类人响应时间等。由于校准仅针对基本车辆动力学（即间距、速度和加速度）进行，因此即使经过适当的参数化，我们也无法确定 CF 模型能够重现部分或全部上述现象以及重现程度。这项工作的目的是提出一个多目标校准框架，并根据汽车排的经验观察进行验证。此外，它还研究了 ACC 驱动车辆的交通动态和行为模式是否可以重现以及在多大程度上可以重现。使用经验数据集测试了两个最先进的 CF 模型。结果表明，所提出的框架带来了可接受的误差和可比较的性能，有助于模型使用基本车辆动力学校准以迄今为止不可能的方式捕获现象。
Calibration of car-following (CF) models is considered a very important task towards reproduction of individual vehicle behaviors and collective traffic phenomena. In most works, calibration is performed on trajectories of leading and following vehicles either on speed or spacing quantities, with the combination of the two plus the acceleration quantity to be considered as the most appropriate according to the literature. With the advent of adaptive cruise control (ACC) technology, there are intrinsic behavioral differences between human- and ACC-driven vehicles that are visible in experimental observations. Some examples include constant headway for ACC-enabled vehicles, string instability, hysteretic behavior, human-like response time and others. Since calibration is performed only on basic vehicle dynamics, i.e., the spacing, speed and acceleration, even after proper parametrization we cannot be certain that CF models will be able to reproduce some or all of the above phenomena and to what extend. The aim of this work is to propose a multi-objective calibration framework validated on empirical observations of car platoons. Furthermore, it investigates if, and to what extent, the traffic dynamics and behavioral patterns of ACC-driven vehicles can be reproduced. Two state-of-the-art CF models are tested with the empirical dataset. The results indicate that the proposed framework leads to acceptable errors and comparable performance, facilitating the models to capture phenomena in a way that is not possible until now, using the calibration on basic vehicle dynamics.