基于手机GPS数据的出行端点识别方法研究

发布时间：2018-04-11 12:19

本文选题：手机GPS技术 + 聚类算法　；参考：《西南交通大学》2017年硕士论文

【摘要】：近年来,随着我国城市化进程的不断加快和城市人口的集聚,交通拥堵问题日益突显,交通基础数据采集是解决城市交通问题的重要手段。居民出行调查是把握城市居民交通需求的基本途径,是交通相关部门制定科学的交通管理政策以及编制交通规划方案的重要依据。手机GPS定位技术作为一种新型居民出行调查技术,相对于传统的居民出行调查,其采集的个体出行轨迹数据具有实时性、动态性等特点,也极大提高了调查效率和调查数据质量,因此基于手机GPS定位技术的调查具有明显的优势,逐渐得到国内外相关学者的重视。本文利用手机GPS定位技术采集不同出行目的类型下的个体出行轨迹数据,然后结合出行日志对试验数据进行采集效果评估。首先,本文设计了通勤出行、上下学出行、非通勤出行(包含日常生活、文化娱乐休闲、接送人、就医、业务等)多种出行目的条件下的个体出行试验,出行个体随身携带安装有专业GPS数据采集软件的手机,同时记录真实的出行日志,将采集到的手机GPS数据与出行日志进行对比,挑选出质量满足要求的原始数据,为后续的出行信息识别效果评估奠定坚实的基础。其次,总结了现有出行端点识别算法的不足之处,针对重复路径停留误识别和同一位置多个出行端点少识别这一问题,提出了一种三步骤识别出行端点的方法:第一步,数据清洗与部分信号缺失轨迹补充;第二步,利用基于密度的时空聚类分析算法(ST-DBSCAN)来提取潜在的出行端点信息;第三步,结合公交站点位置数据来剔除换乘停留导致的误识别。其中ST-DBSCAN在经典的空间聚类算法DBSCAN核心点的定义上有所改进,考虑了时间距离这一因素,更符合真实出行轨迹点的聚集特性。最后,本文以成都市的居民出行手机GPS数据为研究对象,运用基于规则定义的识别算法、基于密度的空间聚类算法DBSCAN、基于密度的时空聚类算法ST-DBSCAN三种算法来进行出行端点识别评估,并通过各种参数水平值组合情景下的识别效果来确定参数选择组合。试验结果表明:ST-DBSCAN算法结合GIS地图匹配算法的识别效果最好,可以达到82.86%的查全率以及10.38%的误差率。
[Abstract]:In recent years, with the rapid development of urbanization and the agglomeration of urban population in China, traffic congestion is becoming more and more serious. Traffic basic data collection is an important means to solve urban traffic problems.The resident travel survey is the basic way to grasp the urban residents' traffic demand, and is the important basis for the traffic related departments to formulate the scientific traffic management policy and the traffic planning plan.Mobile phone GPS location technology as a new type of resident travel survey technology, compared with the traditional resident travel survey, its collection of individual travel trajectory data has real-time, dynamic and other characteristics.It also greatly improves the efficiency of investigation and the quality of survey data, so the investigation based on mobile phone GPS positioning technology has obvious advantages, and has been gradually paid attention to by domestic and foreign scholars.In this paper, the mobile phone GPS positioning technology is used to collect individual travel trajectory data under different travel destination types, and then the acquisition effect of the test data is evaluated by combining the trip log.First of all, this paper designs individual travel experiments under various travel purposes, such as commuting, going up and down, commuting (including daily life, culture, entertainment, leisure, escort, medical treatment, business, etc.).Travel individuals carry mobile phones equipped with professional GPS data acquisition software, record real travel logs, compare the collected mobile phone GPS data with travel logs, and select the original data that meet the requirements of quality.It lays a solid foundation for the subsequent evaluation of the effect of travel information recognition.Secondly, this paper summarizes the shortcomings of the existing trip endpoint identification algorithms, aiming at the problem of repeated path stopping error identification and less identification of multiple trip endpoints in the same location, a three-step method is proposed to identify the trip endpoint: the first step.Data cleaning and partial signal missing track supplement; step two, using density based spatio-temporal clustering algorithm (ST-DBSCAN) to extract potential travel endpoint information; third, combining bus station location data to eliminate misidentification caused by transfer stoppage.ST-DBSCAN improves the definition of the core point of DBSCAN, which is a classical spatial clustering algorithm, and considers the factor of time distance, which is more consistent with the clustering characteristics of the real travel locus.Finally, this paper takes the GPS data of mobile phone in Chengdu as the research object, and uses the recognition algorithm based on the definition of rules.Density-based spatial clustering algorithm DBSCAN-based spatio-temporal clustering algorithm based on density ST-DBSCAN algorithm is used to evaluate travel endpoint identification.The experimental results show that the recognition efficiency of the GIS map matching algorithm is the best, with 82.86% recall and 10.38% error rate.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491

【相似文献】