自动驾驶车辆的出现为安全高效的出行提供可行的解决方案。在自动驾驶车辆的控制任务中，换道决策和轨迹规划是重要的组成部分。作为交通流中频繁出现的行为之一，换道不仅会对交通流的稳定性造成影响，还可能引发交通事故。在此背景下，本文研究自动驾驶与人类驾驶混行以及全自动驾驶两种交通场景下的自动驾驶车辆换道行为。首先，为了给混行场景下的自动驾驶车辆换道控制提供数据支撑，本文探索了人类驾驶车辆的换道特性。在决策层面，用特征集成选择算法提取人类驾驶车辆的代表决策变量集合，使用更少的决策变量将预测准确率最高提升10.68个百分点。在轨迹层面，通过轨迹聚类算法提取了人类驾驶车辆的保守型、比较保守型、比较激进型和激进型四种典型的横向换道轨迹模式。其次，本文提出了一种不同渗透率的混行场景下自动驾驶车辆的换道决策模型，弥补了现有研究较少考虑渗透率的不足。本文使用神经进化策略实现了个体优先、相对个体优先、均等优先和群体优先四种换道决策方案的训练，分析了不同渗透率和不同决策方案下的车流速度、排放和能耗，得出了不同渗透率下的最佳决策方案的匹配方式。进一步，本文提出了一种混行场景下普适性较强的轨迹规划模型。本文不对其他车辆的行为做理想化静态约束，而是通过定义安全、可行和高效三个子目标，建立动态约束条件，以求解优化问题的思路求解原问题。结果表明，当周边人类驾驶车辆使用前述四种换道模式时，自动驾驶车辆均能准确调整轨迹并完成换道操作，应对复杂多变的交通情况。最后，本文提出了一种全自动驾驶场景下多车协同换道轨迹规划模型。与现有研究相比，该模型兼顾了换道完成率、换道时间和车辆平均速度。本文将进入协同区域内的车辆动态分组，将多车协同换道轨迹规划问题转化为组内协同与组间协同两个子问题，允许多辆车同时换道以充分利用时间和空间资源。仿真实验的结果表明，所提出的模型在换道完成率、换道时间和换道期间的车辆平均速度上的综合表现优于对比模型。本文为自动驾驶车辆的换道控制研究提供了新的思路，为自动驾驶车辆在真实交通场景中的应用提供了理论和技术支持。
The appearance of autonomous vehicles provides a feasible solution to safe and efficient trips. Among all control tasks of autonomous vehicles, tasks of decision making and trajectory planning of lane change are important parts. Frequent lane change may not only have an impact on the safety and stability of traffic flows, but also lead to a traffic accident. In the present situation, this paper studies lane change behaviors for autonomous vehicles in mixed traffic scenes with autonomous vehicles and human-driven vehicles and fully autonomous traffic scenes.Firstly, this paper explores lane change features of human-driven vehicles. In terms of lane change decision making, this paper uses the Feature Ensemble Selection Algorithm to extract representative variables of human-driven vehicles, which increases prediction accuracies up to 10.68 points with fewer features. In terms of trajectories, this paper extracts the conservative type, relatively conservative type, relatively radical type, and radical type of lane changes through the Trajectory Cluster Algorithm, which provides support for lane change control in mixed traffic flows. Secondly, this paper proposes the lane change decision-making model for autonomous vehicles in mixed traffic flows with different penetration rates of autonomous vehicles, which covers the shortage of the latest researches. This paper realizes the training of four decision-making maneuvers of lane change based on the evolutionary neuron network, which contains maneuver of individual priority, relatively individual priority, equal priority, and group priority. Then average velocities, emissions and fuel consumption are analyzed, which supports the matching of penetration rates and decision-making maneuvers. Thirdly, this paper proposes a universal trajectory planning model in mixed traffic flows. Without relying on idealized static constraints to non-objected vehicles, this paper defines sub-targets of safety, feasibility and efficiency and builds up dynamic constraints, which solves the trajectory planning problem based on optimization problems. Results show that when human-driven vehicles are among four different lane change modes, autonomous vehicles are able to adjust trajectory and conduct lane changes quickly to adapt to complex and changeable traffic situations.Finally, this paper proposes a trajectory model for vehicles in the fully autonomous traffic scene. Compared with the latest researches, this model considers completion rates of lane change, time of lane change and average velocities of vehicles. This paper divides autonomous vehicles into several groups in the cooperative zone, and transfers the cooperative task into intra-group tasks and inter-group tasks, which makes full use of time and space resources. Results show that this model is better than other models in the above three criteria of vehicles in simulation experiments.This paper provides new ideas and methods for researches of lane-change control for autonomous vehicles, provides theoretical and technical supports for the application of autonomous vehicles.
