基于手机GPS定位数据的交通方式换乘点识别方法研究  

Method for Traffic Mode Transfer Behavior Identification by Using Smartphone GPS Data

作　　者：姚振兴 许心越[2] 邵海鹏[1] 杨飞[3] YAO Zhen-xing;XU Xin-yue;SHAO Hai-peng;YANG Fei(College of Transportation Engineering,Chang'an University,Xi'an 710061,Shaanxi,China;School of Traffic and Transportation,Beijing Jiaotong University,Beijing 100044,China;School of Transportation and Logistics,Southwest Jiaotong University,Chengdu 611756,Sichuan,China)

机构地区：[1]长安大学运输工程学院,陕西西安710061 [2]北京交通大学交通运输学院,北京100044 [3]西南交通大学交通运输与物流学院,四川成都611756

出　　处：《中国公路学报》2021年第12期276-287,共12页China Journal of Highway and Transport

基　　金：国家自然科学基金项目(52002030,52072313,52072044);教育部人文社会科学基金项目(20XJCZH011);陕西省自然科学基金项目(2021JQ-256);陕西省社会科学基金项目(2020R035);中央高校基本科研业务费专项资金项目(300102341676);重庆市规划和自然资源局项目(KJ-2021007)。

摘　　要：交通方式换乘点识别长期以来是手机大数据交通调查领域的一大技术难点,既有研究大多通过设置出行时间、距离阈值进行识别,算法经验性强,普适性不佳,且易将起讫点、信号控制、交通拥堵等停留误识别为换乘停留。为此,提出了一种基于手机GPS定位数据的交通方式换乘点识别新方法:首先,构建模糊时空聚类算法识别个体运动-静止状态,算法同步实现了定位点时空密度双重聚类约束与聚类边界弹性需求,对个体运动状态识别效果更佳;其次,建立支持向量机模型进行交通方式换乘点识别,有效解决了起讫点、信号控制、交通拥堵等停留对换乘停留造成的干扰;最后,从出行链视角出发,提出了基于序列相似度算法的误差回溯自检与优化模型,能够有效修复换乘点漏识别与错误识别问题。此外,在成都市开展了大范围实测试验,由150名志愿者采集了近2 160 h得到的777.6万条数据被用于技术实证评估。试验结果表明:所述方法对交通方式换乘点平均识别准确率达89.3%,换乘时间平均识别误差控制在20 s以内;与既有空间聚类、小波分析算法相比,换乘点识别精度提升近10%,换乘时间误差最大可降低20 s以上,算法适用性与效果更佳。研究成果可为基于活动的交通需求模型演进提供数据支撑,为交通规划与管理部门决策提供技术支持。Mode transfer point identification is a thorny problem in smart phone based travel survey. Existing studies mostly use rule-based methods with specific dwelling time and distance thresholds for mode transfer point identification, the thresholds are highly dependent on experts’ experience and lack universality across travel environments. These methods also have disadvantages of misidentifying trip ends, signal stops and traffic congestions as mode transfer points. Therefore, this paper tries to propose an innovative method for mode transfer point identification by using smartphone GPS data. First, a fuzzy spatial-temporal clustering algorithm was proposed for move-or-stay identification. The method considered both spatial-temporal travel characteristics and soft clustering demand at the same time, and performed much better for travel state identification. Second, a support vector machine based method was further proposed for mode transfer behavior identification. The method had outstanding capabilities in distinguishing mode transfer points from other kinds of travel stays caused by trip ends, signal stops and traffic congestions etc. Based on this, a sequence similarity algorithm based back-stepping optimization model was proposed to improve mode transfer point identification result. In this study, field experiments were carried out in Chengdu to verify the proposed methods. Nearly 2 160 hours of 7.776 million data collected by 150 volunteers were used for technical evaluation. Results show that the average mode transfer point detection rate reaches 89.3% and the average mode transfer time detection errors are within 20 s. Compared with existing point density clustering algorithm and wavelet analysis algorithm, the mode transfer point detection rate can be improved by 10%, and the mode transfer time detection error can be reduced by 20 s. The proposed method has much better performance in both accuracy and consistency. This study can provide detail data support for the evolution of activity-based model and the

关 键 词：交通工程 换乘点识别 模糊时空聚类 手机GPS定位数据 支持向量机 出行链 

分 类 号：U491.1[交通运输工程—交通运输规划与管理]