基于无监督学习技术的位置大数据分析

发布时间：2018-01-27 03:11

本文关键词： 出租车轨迹特征出行方式识别　出处：《浙江理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着社会的发展和科学技术的不断进步,移动通信和传感设备等位置感知技术形成了大量的位置数据,对这些位置数据进行挖掘和分析,从而发现其潜在的有用信息,合理进行城市建设规划和出行路线规划,将会大大促进现代社会的智能化、信息化。同时,从人们出行的轨迹数据中得到其出行的交通方式,也将有助于研究人员从中推测人们对交通工具拥有情况、收入水平和职业情况。本文围绕位置大数据进行了一些相关研究。首先,对出租车GPS轨迹数据进行分析,分析了可能产生数据误差的影响因素,并有针对性地对原始数据做预处理,包括对数据进行误差分析,数据处理依据和方法,之后对居民出行的起止点进行识别,并利用ArcGIS进行地图的可视化呈现。基于处理后的GPS数据,分析了居民出行的时空分布特征,包括工作日和休息日的出行量,日出行的高峰时段,并得到了相关的结论。将聚类分析方法应用到出租车GPS数据的应用研究中,选取合理的聚类方法对研究区域进行交通小区的划分,并建立出行OD矩阵。其次,采用基于时间空间聚类的K-Means方法,得到乘客上下车的活跃中心点,并结合ArcGIS空间分析工具,构建缓冲区找到乘客最容易搭乘出租车的路段,解决了活跃中心点偏离道路的问题。此外,利用密度聚类的方法,对城市居民出行热点区域的分布规律进行了研究,并与实际情况相结合,证明了本文方法的有效性。然后,为提高不同出行方式的识别率,提出了一种基于深度学习的出行方式识别模型,以微软亚洲研究院收集的用户GPS轨迹数据集为基础,利用时间特性提取不同出行方式对应的GPS轨迹段,分析了不同出行方式的特征,考虑了迭代次数对均方根误差和网络训练运行时间的影响,选择合适的迭代次数,以尽可能短的时间,使网络达到最优的训练效果。最后,将本文方法与BP神经网络以及SVM方法进行实验对比,结果表明,本文提出的基于深度学习的模型能对不同的出行方式进行有效识别,其识别准确率较传统方法有明显的提高,证明了该模型在出行方式识别问题上的可行性。
[Abstract]:With the development of society and the continuous progress of science and technology, location sensing technology such as mobile communication and sensor equipment has formed a large number of location data, mining and analysis of these location data. In order to find its potential useful information, reasonable planning of urban construction and travel route planning, will greatly promote the intelligence of modern society, information. At the same time. Getting the way people travel from track data will also help researchers speculate on how people own vehicles. Income level and career situation. This paper carried out some related research around location big data. Firstly, the paper analyzed the GPS track data of taxi, and analyzed the factors that may cause the data error. And the original data preprocessing, including data error analysis, data processing basis and methods, and then the residents travel start and stop point identification. Based on the processed GPS data, this paper analyzes the spatial and temporal distribution characteristics of residents' travel, including the travel volume on weekdays and rest days, and the peak time of daily travel. And get the relevant conclusion. Apply the cluster analysis method to the taxi GPS data application research, select the reasonable clustering method to the study area to divide the traffic area. Secondly, the K-Means method based on time-space clustering is used to get the active center points of passengers, and combined with ArcGIS spatial analysis tools. Constructing buffer zone to find the section where passengers can easily take a taxi solves the problem of active center point deviating from the road. In addition, the density clustering method is used. This paper studies the distribution law of urban residents' travel hot spots, and combines with the actual situation to prove the effectiveness of this method. Then, in order to improve the recognition rate of different travel modes. Based on the user GPS trajectory data collected by Microsoft Asia Research Institute, a travel pattern recognition model based on deep learning is proposed. The GPS trajectory segments corresponding to different trip modes are extracted by using time characteristics. The characteristics of different travel modes are analyzed and the effects of iteration times on root mean square error and network training running time are considered. The appropriate number of iterations is selected for the shortest possible time. Finally, the experiment results show that the proposed method is compared with BP neural network and SVM method. The proposed model based on depth learning can effectively identify different travel modes, and its recognition accuracy is obviously higher than that of the traditional method, which proves the feasibility of the model in the problem of trip pattern identification.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13

【相似文献】