基于大规模出租车GPS轨迹数据的交通信号灯动态计时信息感知

In arterial network, the unreasonable settings of traffic signal timing could result in continuous “stop-and-go” movements of vehicles, and thereby increasing traffic delay and greenhouse gas emissions, and lowering fuel usage. Therefore, sensing and prediction of traffic signal timing is a valuable research topic to be resolved. The goal of the research project is to develop efficient methods and systems to discover the timing parameters of traffic signals and to further predict their future states. Theoretical contributions of the project include: (1) modeling the problem of finely reconstruction of moving trajectories from low-sampling rate trajectories as an multi-constraints optimization problem, and developing an efficient method to solve it; (2) introducing an movement model to characterize the relationship between the restart delay of vehicles waiting for greens and their positions in the queue, for accurately estimating the restart delay time; (3) representing the estimation of timing parameters of traffic lights from noisy and incomplete measurements as the periodicity analysis problem, and proposing robust and reliable methods to solve it. Research works of the project are also of great practical values, which are helpful for lowering emissions, reducing idling time at red signals and congestion, saving travel time, improving fuel efficiency and driving safety.

在主干道路网中，因交通信号灯计时设置不当引起的车辆的反复“停-走”也会导致大量的交通延时、增加尾气排放以及燃油浪费。所以，交通信号灯动态计时信息的感知和预测是一个非常值得研究的课题。本项目的研究目的是设计能从低采样出租车GPS轨迹数据的统计模式中挖掘信号灯的配时参数并对其状态信息进行预测的方法和系统。本项目研究的理论意义在于：(1) 将基于低采样轨迹的轨迹细粒度重构问题描述成多约束优化问题，并提出有效的方法解决；(2) 提出有效模型用于描述路口车辆的启动延时与其在车队中位置关系，并用于准确估计其启动延时；(3) 将基于含噪、不完备观测数据的信号灯配时参数学习问题描述为周期分析问题，并提出快速算法求解。本项目研究的应用价值在于：它有助于提高燃油利用率，降低有害气体排放；减少不必要的停等和路口拥堵，以节省出行时间，加强行车的安全性。

本课题组按照项目计划书所设定的目标及工作计划开展研究。本项目总体目标是研究如何利用大量低采样率的出租车历史GPS轨迹数据挖掘和推断主干道路网中交通信号灯的配时参数信息的方法。为了实现该目标，我们解决的关键问题有：低采样轨迹的细粒度重构；路口停等红灯车辆启动延时模型研究和估计；基于含噪且不完备观测数据的定时信号灯配时参数估计。在量化考核指标方面，课题设定的考核指标为：拟在国内外高质量的期刊或会议上发表学术论文5-8篇，发明专利1-2 项。在本项目的资助下，课题组已发表相关论文6篇，其中2篇SCI期刊论文，1篇CCF推荐B类中文期刊论文，3篇国际会议论文；已申请发明专利5项，其中1项已授权，4项进入公开阶段，完成了上述预期研究目标。主要的理论研究成果包括：(1) 设计和提出了两种高效快速的地图匹配算法用以解决低采样率轨迹的细粒度重构问题。(2) 提出基于低采样率GPS轨迹的有效信号灯状态信息的提取方法用以解决信号灯状态感知问题。(3) 提出两种基于公因子逼近的系列周期估计算法：MFAGCD(Most Frequent Approximate Greatest Common Divisor) 算法和rMFAGCD(robust MFAGCD) 算法，很好地解决了基于稀疏、含噪且含离群观测数据的信号灯计时周期估算问题，能够比较准确地估计路口的计时周期进而分析路口计时周期的变化规律。