罗雪

       随着汽车保有量的持续增长，出现了交通事故趋于频发、通行效率逐渐低下和交通拥堵日益加剧等严重的交通问题。运用人工智能、移动互联、大数据等新一代信息技术，基于车辆巡航系统和辅助驾驶操作，实现智能网联道路环境，能够为缓解当前的交通问题提供技术支撑。论文聚焦智能网联环境下车辆跟驰行为的研究，试图为智能网联环境下人、路、车一体化系统的构建提供理论基础。

       在智能网联环境中，车辆依托环境感知技术、智能决策技术和控制执行等技术成为智能网联车辆，其跟驰行为与普通车辆相比存在较大差异，因此传统的跟驰模型不能直接应用于智能网联车辆。论文根据现有政策规划标准，结合智能网联车辆技术逻辑，分别研究具有信息感知的网联车辆和具有决策控制的智能网联车辆跟驰行为，主要的研究内容和成果如下：
       首先，结合“上海自然驾驶研究”项目的中国驾驶人自然驾驶数据，从超过十六万公里的驾驶数据中提取跟驰片段，进行数据特征分析，获得同一驾驶人在不同的状态下跟驰片段车头时距分布特性，通过数据拟合构建了具有三类驾驶风格的跟驰行为模型。
       其次，结合三类驾驶风格，分别对无信息感知的普通车辆和具有信息感知的网联车辆的跟驰行为进行研究。针对无信息感知车辆在跟驰过程中，前方车辆的行驶状态变化刺激与后车驾驶决策的非线性关系，运用模糊控制理论，建立考虑驾驶风格的跟驰模型，通过数据验证表明，该模型能够准确地描述不同驾驶风格的驾驶人跟驰过程。针对具有信息感知车辆，考虑网联环境中前方车辆信息和路侧设备信息对于跟驰模型的影响，根据驾驶人对于网联信息的采纳程度区分为不同的驾驶风格，采用IDM 模型进行优化，改进后的模型能够更准确地描述网
联车辆跟驰过程。
      然后，针对具有决策控制的智能网联车辆的跟驰行为进行研究。通过分析智能网联车辆的运动控制方式，运用安全距离模型的理论，对Gipps 模型进行加速度优化和速度约束条件改进。选择与智能网联车辆特性接近的自然驾驶数据，用Theil’s U 函数作为目标函数，采用K 折交叉验证方法，通过遗传算法对模型进行标定与评价。结果表明，改进后模型的速度、加速度和位移的评价指标（平均误差、平均绝对误差、平均相对误差和均方根误差）均有所降低，模型更接近于真实的驾驶过程。
      最后，运用仿真分析，验证了所建立的跟驰模型，结果表明：

      1）本文所建立的具有信息感知的车辆跟驰模型，运用稳定性理论分析显示，模型的稳定性较好，且在接收前方5 辆车辆信息反馈时稳定性最佳。通过数值仿真分析起步、加速、拥堵和交通事故场景，显示出模型能够较好地描述在网联环境下，不同驾驶风格的驾驶人在不同场景下的跟驰过程。用跟驰过程中最小碰撞时间小于5s 的累积分布进行安全性分析，结果显示传统IDM 模型占比为19%，而采用本文改进后的模型占比则为5%，减少了14%的安全风险，安全性更优。对舒适性指标进行分析的结果显示，本文改进的模型整体的驾驶舒适性有所提升，
舒适度低于不舒适等级的占比约为19.5%，相较于传统IDM 模型减少了14%。
      2）本文所建立的具有决策控制的车辆跟驰模型，通过单车道环形数据仿真法分析其稳定性，结果显示：在小扰动状态下，100 辆车队头尾跟驰行驶时，受到小扰动影响的车辆为10 辆左右；持续扰动状态下，受到影响的车辆仅为20 辆左右，影响范围较小且无滞后现象，稳定性较好。模型最小碰撞时间的累积分布中，小于5s 的跟驰事件为4%，相较于传统Gipps 模型减少了11%，安全性更优。本文改进的模型舒适度低于不舒适等级的占比约为35%，相较于传统Gipps模型减少了23%；相较于自然驾驶降低了50%，不舒适的驾驶情况缓解明显。

关键词：驾驶风格；智能网联车辆；跟驰行为；跟驰模型；跟驰模型评价；仿真分析

ABSTRACT

With the continuous growth of the number of vehicles, there are serious traffic problems such as frequent traffic accidents, low traffic efficiency and increasing traffic congestion. With the development of the new generation of information technology, such as artificial intelligence, mobile Internet, big data, etc., the vehicle cruise system and auxiliary driving operation are gradually mature, and the intelligent transportation system will be realized in the future. It can provide technical support to alleviate the current traffic problems. This paper focuses on the research of car-following behavior in the intelligent transportation system, and tries to provide a theoretical basis for the construction of the integration system of human, road and vehicle in the intelligent transportation system.

In the intelligent transportation system, intelligent connected vehicles rely on environment sensing technology, intelligent decision-making technology and control execution technology. Compared with the ordinary vehicle, the car-following behaviors of the intelligent connected vehicle are quite different. Therefore, the traditional car following model can’t be directly applied to the intelligent connected vehicle. According to the policies and standards and the technical logic of the intelligent connected vehicle, this paper studies the car-following behaviors of the connected vehicle with information perception technology and the intelligent connected vehicle with decision-making control technology respectively. The main research contents and achievements are as follows:

First of all, the data of this paper is based on the “Shanghai Natural Driving Research”. The car-following segment is extracted from more than 160,000 kilometers of Chinese drivers’ driving data for analysis. The analysis of data characteristics is carried out to obtain the time headway distribution characteristics of the same driver’s following segments in different states. Three driving styles are fitted according to the data characteristics.

Secondly, combined with the three driving styles, the car-following behaviors of the general vehicles without information perception technology and the connected vehicles with information perception technology are studied respectively. Aiming at the non-linear relationship between the driving state change stimulus of the vehicle ahead and the driving decision of the vehicle behind in the car-following process of the vehicle without information perception technology, the car-following model considering the different driving style is established by using the fuzzy control theory. The data verification shows that the model can accurately describe the car-following process of the driver with different driving styles. For vehicles with information perception technology, considering the impact of vehicle information ahead and roadside equipment information on the car-following model in the intelligent transportation system, the car-following model is improved based on intelligent driver model (IDM model). According to the degree of adoption of information in intelligent transportation system by drivers, the model is divided into different driving styles and optimized. The simulation results show that the improved model can describe the following process of connected vehicles more accurately.

Then, the car-following behavior of intelligent connected vehicles with decisionmaking control technology is studied. By analyzing the motion control mode of the intelligent connected vehicle, and using the theory of safe distance, the acceleration optimization and speed constraint conditions are improved based on the Gipps model. By selecting the Chinese drivers’ driving data which close to the characteristics of the intelligent connected vehicle, the paper takes the Theil’s U function as the objective function, uses the K-Fold cross-validation method, and uses the genetic algorithm to calibrate and evaluate the model. The results show that the evaluation indexes including mean error, mean absolute error, mean relative error and root mean square error of speed, acceleration and displacement of the improved model are all reduced. The improved model is closer to the real driving process.

Finally, the simulation analysis is used to verify the improved car-following model. The results show that:

1) Based on the analysis of stability theory, the improved car-following model with information perception technology has a better stability comparing with IDM model. When receiving the information feedback of 5 vehicles ahead, its stability is the best in the tests. Through the numerical simulation analysis of starting, acceleration, congestion and traffic accident scenarios, it shows that the model can better describe the car-following process of drivers with different driving styles in different scenarios under intelligent transportation system. The cumulative distribution of minimum time to collision less than 5 seconds is used for safety analysis. The results show that IDM model accounts for 19%, while the improved model accounts for 5%, which reduces 14% of the security risks and improves the security. The results of the analysis of the comfort index show that the driving comfort of the improved model is improved. The proportion of the comfort level lower than the discomfort level is about 19.5%, which is 14% less than IDM model.

2) The car-following model with decision-making control technology established in this paper is simulated by single-road method to analyze its stability. The results show that: in the small disturbance state, when 100 vehicles are following each other, the number of vehicles affected by the small disturbance is about 10. In the continuous disturbance state, the number of vehicles affected is only about 20. The disturbance has little influence on the model, and there is no lag phenomenon. The improved model has better stability. The cumulative distribution of minimum time to collision less than 5 seconds of improved model is 4%, which is 11% less than that of Gipps model. As a result, the security of improved model is better. The proportion of the comfort level lower than the discomfort level is about 35%, which is 23% lower than Gipps model, and 50% lower than Chinese drivers’ driving data, so the uncomfortable driving situation is alleviated obviously.

Key words: driving style; intelligent connected vehicle; car-following behavior; carfollowing model; car-following model evaluation; simulation analysis