基于智能汽车环境感知的交通流全车微观轨迹重构研究

Whole traffic state information, including micro trajectories of all vehicles in traffic flow, plays a key role in enabling intelligent transportation systems to make fine traffic management and control strategies. However, most of existing studies concentrate on coarse-grained link level traffic state estimation, rarely investigate the reconstruction of vehicle trajectories in traffic flow at the microscopic level. This research project aims to develop model of integrating macro-meso-micro level to reconstruct micro trajectories of all vehicles in traffic flow based on micro trajectory data from environment perception of intelligent vehicles. Firstly, at link level, lane change probability distribution with respect to lane traffic state is modeled using multinomial logit discrete choice model; and by extending the Eulerian traffic flow model, Lagrangian multi-lane traffic flow model is formulated using partial differential equation and graphic derivation. Then, at lane level, Lagrangian lane traffic state estimation model is developed by integrating extended Kalman filter with Lagrangian multi-lane traffic flow model. Next, at vehicle level, applying dynamic programming for determining lane change and a constrained continuous optimal control model for determining vehicle trajectory at certain lane, micro trajectories of all vehicles in traffic flow are reconstructed by embedding continuous optimal control problems in dynamic programming framework. The outcomes of this research project will advance the theory system of traffic state estimation and provide technical support for connected autonomous vehicles-based intelligent transportation systems.

完整交通状态信息，囊括全体车辆运动所形成的微观轨迹，对智能交通系统实现精细交通管控至关重要。以往研究多专注于粗粒度的路段级交通状态，鲜见从微观层面重构交通流全车时空轨迹的研究。本项目拟基于智能汽车环境感知采集而得部分微观轨迹，研究针对交通流全车微观轨迹重构的宏观-中观-微观一体化模型。首先对路段，基于多项Logit离散选择模型，建立依车道状态的换道概率模型；并基于对欧拉交通流模型的扩展，结合偏微分方程与图形推导，构建拉格朗日多车道交通流模型。然后对车道，基于拉格朗日多车道交通流模型和扩展卡尔曼滤波的融合创新，建立拉格朗日车道交通状态估计模型。进而对车辆，利用动态规划来确定车辆换道，用带约束连续最优控制模型对车道内车辆纵向运动轨迹建模，将连续的最优控制问题嵌入到动态规划的框架里实现交通流全车微观轨迹重构。项目成果将拓展交通状态估计理论体系，为自动驾驶与车联网智能交通系统提供技术支撑。

装配有高精度传感设备的智能汽车正在成一种新型交通流数据采集技术。由于智能车渗透率、感知范围和车辆遮挡的限制，基于智能车采集的交通流轨迹存在缺失的情况。为了获取交通流全车全样本微观轨迹，本项目围绕着智能车数据采集模、车道交通状态估计、全样本微观轨迹重构和混合交通流仿真等四方面开展研究。项目具体开展了七方面研究并取得相关成果：（1）从智能车环境感知原理入手，提出基于全样本轨迹数据集的智能车数据采集仿真方法，并开发了相应的软件系统；（2）定义数据采集技术效率评价指标，考虑车辆随机动态性和车辆间遮挡关系，基于随机几何理论构建智能车感知效率理论模型。进一步构建了定点检测和浮动货车采集效率理论模型；（3）通过时空轨迹图，分析智能车数据和车道排队之间的关系，构建基于智能车的车道排队检测以及估计方法；（4）分析定点检测、浮动车和智能车数据的特性，提出了融合多源数据基于扩展卡尔曼滤波的车道级交通状态估计方法；（5）针对完全未被探测到的车辆轨迹，提出基于交通流时空热力图的车辆轨迹重构方法；（6）为了重构部分缺失车辆轨迹，考虑智能车轨迹片段化、匿名性等特性，提出基于车辆跟驰模型的全样本轨迹迭代重构方法；（7）开发面向智能网联协同感知的智能车混合交通流仿真平台，该平台通过对交通流仿真平台SUMO，机器人仿真平台Gazebo和机器人操作系统ROS的集成开发，实现了智能车与普通车辆混合交通流在三维虚拟空间的微观仿真运行。项目成果可为数字高速、数字孪生和车路协同自动驾驶等提供技术支撑。