基于自然驾驶数据的分心状态特征分析与跟驰行为建模

Distracted Car-Following Behavior Analysis and Modeling Based on Large-Scale Naturalistic Driving Experiment

摘要:

基于行为特征对分心跟驰行为进行分类和建模，根据模型标定的方法探究不同状态下跟驰模型的适应性。首先从持续开展3年有余的上海自然驾驶数据中提取了大量分心跟驰样本，基于驾驶行为刺激反应框架对分心状态特征进行了初步分类，得到了5类分心跟驰行为；其次分析了现有4类经典跟驰模型（GHR、GIPPS、IDM和Wiedemann）对分心跟驰行为的适应性，同时根据五折交叉验证适应性结果对分心跟驰行为分类进一步优化，最终得到3类分心跟驰行为（麻木反应、过激反应和延迟反应）；最后探讨了分心状态下的2种跟驰行为建模策略（AIDM和TDIDM）。结果表明，对IDM模型进行合理标定即可较准确地描述不同类型的分心跟驰行为。

Abstract:

Driving distraction is a common abnormal behavior which seriously disturbs the safety and stability of traffic flow. Because the vehicle has a large degree of freedom in the longitudinal car-following（CF） driving context， most of the distracted driving behaviors exist in the CF driving context. Therefore， it is of great significance to deeply understand the characteristics of the distracted CF， and accurately describe the distracted CF behavior. This paper classified and modeled distracted the CF behavior based on behavioral characteristics and explored the adaptability of CF models in different states using model calibration methods. First， a large number of distracted CF samples were extracted based on the naturalistic driving data of Shanghai which spaned more than 3 years， and the distracted CF events were preliminarily classified into five kinds based on the stimulus-response framework of driving behavior. Then， the adaptabilities of four existing classical CF models （GHR， Gipps， IDM， and Wiedemann） on distracted CF behavior were compared， and the classification of distracted CF was optimized according to the results of the model adaptability using five-fold cross-validation. As a result， three types of distracted CF behavior （numbness reaction， overreaction， and delay reaction） were determined. Finally， two modeling strategies for the distracted CF behavior were discussed （AIDM and TDIDM）. The results show that rational calibration of the IDM model can accurately describe different types of distracted CF behavior.