基于GPS轨迹数据的交通出行方式识别研究

发布时间：2018-06-28 14:57

本文选题：GPS + 转换点识别　；参考：《北京交通大学》2017年硕士论文

【摘要】：随着我国社会经济发展和城市化进程速度的加快,交通拥堵、交通事故、交通环境等问题已成为我国城市常见的"城市病"。科学的交通规划和管理被大多数学者认为是解决城市交通拥堵等问题的有效手段,而居民出行信息的掌握是科学的交通规划和管理的信息支撑。居民出行信息获取最广泛的调查方法是传统的居民出行OD调查。但该方式在实际中受被调查者主观意识影响,出现漏报、错报的现象,影响调查数据的质量,同时存在成本高、工作量大、回收率低和处理周期长等问题,影响后续的交通规划和管理工作。现今,全球定位系统(GPS)的广泛应用,以及GPS出行数据不受被调查者主观意识的影响,能够体现更准确、完整的个人出行行为信息,同时还具备调查效率高、数据精度高、获取信息量大等优点,使得这个调查方式成为居民出行行为信息获取的有效的新途径。挖掘分析GPS数据获得更为精确、完整的出行行为信息,进行交通出行方式识别是获取个人出行信息的重要组成部分,尤其转换点的识别作为交通出行方式识别的重要组成部分,是当前研究的难点所在。本文基于GPS轨迹数据的交通出行方式识别研究。首先提出两种基于相似性度量和窗口的转换点识别方法,即多段方法和移动窗口两种方法,然后比较两种转换点识别方法,获取最优的方法,最后把该方法运用到交通出行方式识别中获取出行方式段,提取所获出行方式段的特征参数,采用BP神经网络、决策树、KNN和支持向量机(SVM)四种模式识别的方法识别交通出行方式。采用Geolife工程的GPS轨迹数据对本文所提出算法进行验证。在基于相似性度量和窗口方法的转换点识别中,比较多段和移动窗口两种算法结果,得到采用多段的方法使得识别结果最优,F-score取值接近80%,其中召回率接近90%。在交通出行方式的识别中,四种模式识别出的结果与实际交通出行方式进行比较验证,得到采用SVM识别交通出行方式能够取得最优结果,训练集和测试集分别为92.75%和88.77%。最后,采用时间指标,进行交通出行识别结果的综合评价,得到识别的准确率由88.77%提高到91.84%。验证了本文所提出的算法具有较高的识别精度。本文提出基于GPS轨迹数据的交通出行方式识别算法,通过实验验证了该方法的有效性,其研究结论可应用于分析居民出行行为特征,该研究结果所获得的居民出行信息特征是后期交通规划和管理的重要数据支撑。
[Abstract]:With the rapid development of social economy and urbanization in China, traffic congestion, traffic accidents, traffic environment and other problems have become a common "urban disease" in Chinese cities. Scientific traffic planning and management is considered by most scholars to be an effective means to solve problems such as urban traffic congestion, and the information of resident travel information is the information support of scientific traffic planning and management. The traditional OD survey is the most widely used method to obtain residents' travel information. However, this method is affected by the subjective consciousness of the respondents in practice, and the phenomenon of underreporting and misreporting affects the quality of the investigation data. At the same time, there are some problems, such as high cost, large workload, low recovery rate and long processing period, etc. Impact on subsequent traffic planning and management. Nowadays, with the wide application of GPS and the fact that GPS travel data are not affected by the subjective consciousness of the respondents, they can reflect more accurate and complete personal travel behavior information, at the same time, they also have high investigation efficiency and high data accuracy. Because of the large amount of information obtained, this investigation method is an effective new way to obtain the information of residents' travel behavior. Mining and analyzing GPS data to obtain more accurate and complete travel behavior information, the identification of traffic travel mode is an important part of obtaining personal travel information, especially the identification of transition points as an important part of the identification of traffic travel mode. It is the difficulty of current research. In this paper, the identification of traffic travel mode based on GPS track data is studied. At first, two methods based on similarity measurement and window are proposed, that is, multi-segment method and moving window method. Then, the two methods are compared to obtain the best method. Finally, the method is applied to obtain the travel mode segment in the identification of traffic travel mode, and the characteristic parameters of the obtained travel mode segment are extracted, and the BP neural network is used. Decision tree (KNN) and support vector machine (SVM) are used to identify traffic travel modes. The proposed algorithm is verified by using the GPS trajectory data of Geolife project. In the conversion point recognition based on similarity measurement and window method, the results of multi-segment and moving window are compared, and the optimal F-score of recognition results is obtained by using multi-segment method, in which the recall rate is close to 90. In the recognition of traffic travel mode, the results of the four patterns are compared with the actual traffic travel mode. The results show that SVM can obtain the best result, the training set and the test set are 92.75% and 88.77% respectively. Finally, the time index is used to evaluate the result of traffic trip identification, and the accuracy of identification is improved from 88.77% to 91.84%. It is verified that the proposed algorithm has high recognition accuracy. In this paper, an algorithm based on GPS trajectory data is proposed to identify traffic travel patterns. The validity of the method is verified by experiments. The research results can be used to analyze the characteristics of residents' travel behavior. The characteristics of resident travel information obtained from the results of this study are important data support for later traffic planning and management.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P228.4;U491.1

【参考文献】