位置感知下的移动行为特征建模

发布时间：2018-01-19 05:01

本文关键词： 移动轨迹空间划分特征模型模式匹配终点预测逡逑　出处：《西安科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着科技的迅速发展,GPS定位设备以及无线通信技术的普及应用程度大大提高,基于位置的移动终端设备愈来愈趋向智能化,我们可以方便的收集移动对象的轨迹数据,这些数据蕴含着大量的移动行为特征规律,可以为城市交通规划、广告推送以及人群特征行为的研究提供决策支撑。因此,对移动行为特征建模具有重要的研究意义。本文以移动对象的轨迹数据为研究对象,以移动对象的行为特征建模和移动轨迹终点预测为主要目标。本文的研究工作主要包括以下四个方面:(1)在数据预处理阶段,本文以真实的移动轨迹数据为基础,提出了两种对于移动轨迹所在空间区域的动态结构化表示方法——Q网格划分法和KD-树网格划分法,避免了现有技术中常用到的等尺度网格划分法存在的划分粗糙、轨迹语义化程度差的问题;(2)在特征建模阶段,针对在基于贝叶斯公式的预测算法中易遇到的“数据稀疏”问题,本文通过轨迹划分与重构的方法,在贝叶斯网络理论和马尔科夫模型的基础上,提出了基于网格序列编号、基于轨迹序列编号的两种步长统计方法以及建立混合移动行为特征建模的方法;(3)在轨迹预测阶段,本文在OD模式匹配预测算法的基础上加入了轨迹的局部特征点,提出OMD特征点混合模式匹配预测方法,改善了O 特征点模式匹配预测算法只考虑移动轨迹全局模式而造成的轨迹匹配度低的问题;(4)在预测结果修正阶段,本文在粗粒度空间划分模型的预测结果基础上,融合了细粒度路网模型下的轨迹预测结果对原始预测结果进行修正,提高了轨迹的预测精度。本文以Matlab2014作为实验平台,以深圳市真实的出租车移动轨迹数据作为原始轨迹数据库进行实验验证和分析。通过数据预处理、空间结构划分、建立混合特征模型以及模式匹配预测等过程对轨迹的终点进行预测并与实际终点进行比较。实验结果表明,本算法的终点预测率可达到94.6%,实现了轨迹终点的预测功能。
[Abstract]:With the rapid development of science and technology, GPS positioning equipment and the popularity of wireless communication technology have been greatly improved, the mobile terminal equipment based on location is becoming more and more intelligent. We can conveniently collect the trajectory data of moving objects, which contains a large number of characteristics of mobile behavior, and can be used for urban traffic planning. The research of advertising push and crowd characteristic behavior provides decision support. Therefore, it is of great significance to model mobile behavior features. This paper takes the trajectory data of moving object as the research object. The main goal of this paper is to model the behavior characteristics of moving objects and predict the end point of moving trajectory. The research work in this paper mainly includes the following four aspects: 1) in the stage of data preprocessing. Based on the real moving trajectory data, this paper presents two dynamic structured representation methods for the space region of the moving trajectory, I. E. Q mesh division method and KD-tree grid partition method. It avoids the problems of rough partition and poor semantic degree of locus, which are often used in the prior art. 2) in the stage of feature modeling, aiming at the problem of "data sparsity" which is easily encountered in the prediction algorithm based on Bayesian formula, this paper adopts the method of trajectory partition and reconstruction. On the basis of Bayesian network theory and Markov model, two step size statistics methods based on grid sequence numbering and trajectory sequence numbering are proposed, as well as the modeling method of hybrid mobile behavior features. In the phase of trajectory prediction, based on OD pattern matching prediction algorithm, the local feature points of trajectory are added, and a hybrid pattern matching prediction method of OMD feature points is proposed. The O feature point pattern matching prediction algorithm only considers the problem of low trajectory matching degree caused by moving trajectory global pattern. 4) at the stage of forecasting result correction, this paper combines the track forecast result of fine grained road network model to revise the original forecast result based on the prediction result of coarse-grained space partition model. The accuracy of trajectory prediction is improved. In this paper, Matlab2014 is used as the experimental platform. The real taxi track data of Shenzhen city is used as the original track database for experimental verification and analysis. The spatial structure is divided by data preprocessing. The hybrid feature model and pattern matching prediction are established to predict the end point of the trajectory and compared with the actual end point. The experimental results show that the endpoint prediction rate of this algorithm can reach 94.6%. The prediction function of trajectory end point is realized.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495

【参考文献】