混行模式下考虑加速车道长度与汇入风险时空演化特性的智能车辆自主汇入策略研究

With small batches of road test of intelligent vehicles, the era of coexistence of traditional vehicles and intelligent vehicles on the existing roads is coming. In the coexistent mode, the autonomous merging of intelligent vehicles from the acceleration lane to the host lane on the freeway is a pervasive scenario that requires human intervention. Limited by the remaining length of the acceleration lane and the random traffic flow of the host lane, the intelligent vehicles must complete the merging behavior within a restricted time, which would inevitably make the intelligent vehicles conflict with the traditional vehicles in the host lane. In view of the above issue, this project intends to implement a research on the subjective identification of the critical state of lane change behavior and merging behavior safety for drivers in practical traffic environment, and establish the boundary constraints of driver's security risk perception on merging behavior safety by analyzing the decision-making mechanism of lane change behavior and merging behavior safety for different drivers. Through applying the established risk boundary constraints to the merging strategy of intelligent vehicle, the safety constraint framework of intelligent vehicle's merging behavior risk is depicted. Then a dynamic spatiotemporal safety mechanism during the merging process is generated by introducing the remaining length of the acceleration lane and combining the traffic environment data. Under the restriction of risk constraint framework of merging behavior, a dynamic merging model related to the remaining length of the acceleration lane is formed by learning the established safety mechanism. Based on the comprehensive balance mechanism between the merging demand and the merging risk of the intelligent vehicles, the existing acceleration lane design scheme is optimized to realize the autonomous merging of intelligent vehicles from the acceleration lane to the host lane on the freeway.

随着智能车辆的小批量路测，传统车辆与智能车辆在现有道路的混行模式已经开始。混行模式下，智能车辆在高速公路加速车道的自主汇入是目前普遍需要人工干预的场景。受加速车道剩余长度和主线随机交通流限制，智能车辆必须在有限时间内完成汇入，此过程中不可避免的与主线传统车辆发生一定冲突。针对上述情况，本项目拟开展真实交通环境下驾驶人对换道、汇入安全临界状态的主观辨识研究，分析驾驶人的换道、汇入安全决策机理，建立驾驶人对汇入安全的安全风险认知边界约束。将此风险边界约束反过来填充于智能车辆的汇入策略，刻画智能车辆汇入风险安全约束框架。通过引入加速车道剩余长度，结合交通环境数据，生成汇入进程中的动态时空安全机制。在汇入风险约束框架限制下，通过学习生成与加速车道剩余长度相关联的动态汇入模型，基于智能车辆汇入需求和汇入风险的综合平衡机制对现有加速车道设计方案进行优化，实现智能车辆在高速公路加速车道的自主汇入。

智能车辆在加速车道的汇入过程是目前需要人为接管的代表性场景之一，如何提升智能车辆的类人化熟练汇入是目前存在的基础技术难题。针对上述问题，本项目设计开展了2类汇入安全决策数据采集实验。第1类实验中，共计40名被试驾驶实验车完成了自然驾驶实验。第2类实验中，在高速公路汇入区路侧安装隐蔽式数据采集平台，采集了超过2000次以上的自然汇入数据。在此基础上，主要研究成果如下：.（1）车辆在汇入过程中平均加速度小于1m/s^2，超过90%的车辆选择在加速车道初、中段进行汇入，仅有1%的车辆汇入时加速车道剩余长度小于50m；转向灯开启率为45.12%，驾驶人平均在执行汇入时刻之前2.68s开启转向灯，方向盘转角多为小幅转角；除前方区域外，驾驶人对左侧区域的注视频次最高。.（2）参数考虑剩余加速车道长度参数的组合模型决策效率均优于不考虑剩余加速车道长度参数的组合模型，且考虑剩余加速车道长度参数的组合在更短的时间窗口下决策效率达到最大，这也证明了剩余加速车道长度参数在汇入决策过程中的重要性。.（3）以剩余加速车道长度为参数，建立并标定了基于安全距离和自车在决策时刻所需加速度的安全汇入理论模型，确定了在安全汇入前提下自车在汇入起点所需最小加速度。采用信号检测论确定了汇入风险决策阈值，大小为0.18m/s^2。该阈值对安全样本的决策召回率为92.77%，对危险样本的决策召回率为95.15%，对汇入风险的整体决策准确率为93.88%。进行结果表明，基于安全汇入理论模型的汇入风险决策阈值更符合驾驶人对汇入风险的感知。.（4）针对智能车辆对汇入成功率的需求，并以加速车道的“加速效应”为依据，尝试验证了不同加速度决策阈值对汇入风险决策的影响效果。考虑常规乘用车加速能力，在阈值设置为1m/s^2的情况下对模型进行仿真，结果表明所提出的面向汇入风险决策的智能车拟人化汇入策略，在保证汇入安全的前提能实现97.5%的汇入成功率，可应用于智能车辆的汇入决策过程。