SIA OpenIR  > 信息服务与智能控制技术研究室
基于时空轨迹数据的移动行为模式挖掘研究
其他题名Mining Moving Behavior Patterns from Spatiotemporal Trajectory Database
王亮1,2
导师胡琨元 ; 库涛
分类号TP311.13
关键词时空轨迹数据 数据挖掘 移动行为模式 群体移动行为动力
索取号TP311.13/W34/2013
页数118页
学位专业机械电子工程
学位名称博士
2013-11-28
学位授予单位中国科学院沈阳自动化研究所
学位授予地点沈阳
作者部门信息服务与智能控制技术研究室
摘要随着智能传感技术的不断发展以及广泛应用,使得我们获取数据的广度与深度有了进一步的提升。基于移动时空轨迹数据以研究社会系统中个体及群体的移动行为模式,挖掘其中所蕴含的潜在规律及知识,对于诸如社会管理、交通系统规划与监控、基于位置的服务(Location Based Service, LBS) 等领域都具有非常重要的作用,同时这也是当前数据挖掘与社会计算领域的一个研究热点,引起了学术界与产业界的广泛关注,目前在这一领域已经涌现出了一些有价值的研究成果。 然而,通过对现有研究工作的分析,发现现有的基于时空轨迹数据的移动模式挖掘方面的研究缺乏对于连续时间维度、空间维度及其相关衍生属性特征的系统性研究,缺乏对于不确定时空轨迹数据的移动行为模式分析研究,同时在基于移动交通轨迹的城市环境下群体移动模式等方面尚存在较大的研究空间。 因此本论文的研究目的一方面是基于空间无约束时空轨迹深入研究时空轨迹数据模式挖掘领域的基础性、普遍性问题,为面向行为模式应用的社会计算领域积累理论储备;另一方面是基于交通路网时空轨迹以研究城市背景下,群体移动行为模式分析与挖掘的新工具、新方法,进一步扩展基于时空轨迹数据知识发现的应用领域。 论文的主要研究内容包括:轨迹数据时空维度下的特征抽取及移动模式挖掘方法、基于生物信息素集成浓度的移动停留模式挖掘、不确定时空轨迹数据的表示及模式挖掘、城市环境下群体移动行为动力学分析方法以及基于交通轨迹的市民移动出行模式挖掘。具体的研究内容和创新性成果概括如下: (1) 轨迹数据时空维度下的特征抽取及移动模式挖掘方法 分别针对时空轨迹数据在时间维度上的连续性分布特征以及空间维度上的空间语义连续近似分布特征展开研究,在时间维度上就等时间间隔采样轨迹内在的时间维度离散化性质,提出了对应瞬时采集时间上的时空热点区域的概念与发现方法;针对随机时间间隔采样轨迹在时间维度上分布疏密不均的特点,提出了密集时间窗体自动检测方法以发现一维时间投影数据中的密集区间;进而将原始时空轨迹数据转换为以时空热点区域表示的序列数据,设计并实现了适应于等时间间隔轨迹的闭频繁移动模式挖掘算法MTCloSpan,随机时间间隔轨迹的移动模式算法FMTPM。在空间维度上针对传统网格划分方法存在的Sharp Boundary问题易产生部分移动模式丢失的现象,提出了空间模糊划分方法,以基于距离的空间网格隶属度对轨迹位置点进行多近邻网格匹配计算,进而基于深度优先与广度优先搜索设计了相应的带严格时间约束的移动模式挖掘算法VTPM-PrefixSpan与VTPM-GSP,通过实验仿真测试验证了所提出方法对于解决对应问题的有效性。 (2) 基于生物信息素集成浓度的移动停留模式挖掘 基于生物信息素集成浓度的思想以解决时间空间组合维度下的移动停留模式挖掘问题,借鉴生物界中的昆虫觅食、通信释放信息素的原理,通过计算生物信息素集成浓度空间分布以分析移动对象在不同空间上的停留时间分布特征,进而以空间停留时长为参数计算空间区域偏好程度。进而通过数据语义转换处理得到停留时间移动序列数据,通过对时间注释频繁模式挖掘算法进行改进,提出了适应于时空组合维度下移动停留模式挖掘问题的TPM算法。通过对相关实验结果的分析,显示本文所提出的方法可以有效挖掘时空轨迹数据的移动停留模式。 (3) 不确定时空轨迹数据的表示及模式挖掘 在实际应用中,所得到的时空轨迹数据往往带有不确定的因素,其或是表现于时间维度,或是表现于空间维度。本文分别从时间不确定与空间不确定角度研究了不确定时空轨迹的移动模式挖掘问题。在时间不确定时空轨迹部分,针对以区间值表示的不确定时间信息,通过模糊集合理论求取过渡时间区间值对于模糊谓词逻辑的隶属度值,进而分别以深度优先搜索与广度优先搜索设计并实现了相应的模式挖掘算法FTP-PrefixSpan与FTP-Apriori。在空间不确定时空轨迹部分,针对不确定时空轨迹 -邻域的空间分布特征,构建了基于相邻网格分割面积网格隶属值计算方法,进而设计并实现了改进的序列模式挖掘算法UTFP-PrefixSpan,以进行严格时间间隔约束条件下的概率序列数据模式挖掘。通过测试数据的仿真实验结果表明,本文所提出的方法在针对不确定时空轨迹的移动行为模式挖掘方面,其效率与参数扩展性均具有较好的性能。 (4) 城市环境下群体移动行为动力学分析方法 以城市交通GPS时空轨迹数据为基础,提出了一种多尺度空间弹性离散化划分方法,该方法基于移动轨迹O-D点的空间分布特征实现了与移动时空轨迹语义相匹配的多尺度空间划分集合,避免了等尺度网格硬划分对密集区域边缘的破坏,同时增强了密集区域与稀疏区域的区分能力;通过多尺度空间弹性离散化划分方法,实现了对城市环境下群体移动行为动力学的计算及分析,主要包括不同时间区间内的群体移动活跃度空间分布、移动活跃度规则性测算以及群体移动区域连通关系统计分析。 (5) 基于交通轨迹的群体移动出行模式挖掘 针对城市环境下,群体移动出行行为模式挖掘问题进行了深入探讨与研究,以多尺度空间弹性离散化划分方法为基础构建移动出行全局模式,以城市路网拓扑关系模型为基础构建移动出行过程模式,其中城市路网拓扑关系模型通过移动交通GPS历史轨迹数据进行关键位置节点的探测与抽取;在此基础之上,提出了移动全局模式与移动过程模式相结合的群体移动出行模式的构建与挖掘方法,即通过移动轨迹O-D点位置对与移动过程序列数据进行移动全局模式与过程模式的综合发现。通过深圳市出租车GPS数据的实验结果表明,本文所提出的方法与现有方法相比在区域划分、数据转换等方面均具有更好的性能,同时挖掘结果语义更为丰富,可解释性更强。
其他摘要With the continuous development of smart sensor technology and extensive application allows us to obtain the breadth and depth of the data has been further improved. Temporal trajectory data based on mobile social systems to study the movement of individuals and groups of lines and patterns, which are inherent in mining law and knowledge of the potential for such social management, transportation system planning and monitoring, location-based services (Location Based Service, LBS ) and other fields have a very important role, and this was also the current data mining and social computing a research hotspot, academia and industry attracted widespread attention, both in this area also emerged a number of valuable research results. However, through the analysis of existing research work and found that the existing data on the movement of temporal trajectory pattern mining research for the lack of continuous-time dimensions, characteristics of spatial dimensions, the lack of time for the transition, the residence time in the mobile behavior pattern the analysis, based on mobile traffic in both urban environments trajectory group movement patterns and other aspects of surviving in a larger research space. Therefore, the purpose of this thesis is based on the one hand, freedom of movement trajectory in-depth study temporal trajectory data pattern mining areas of basic, universal, patterns of behavior -oriented social computing applications has accumulated theoretical reserves ; the other hand, is based on the traffic movement trajectory to study the urban context, the research group movement patterns of behavior analysis and mining tools and methods , extended temporal trajectory data based on knowledge discovery applications. Contents of the study include: spatial and temporal dimensions of the track data feature extraction and moving pattern mining method, based on the concentration of bio- integrated mobile pheromone stays pattern mining and uncertain data representation and temporal trajectory pattern mining, urban environment group movement dynamics methods of Analysis and mobile -based public transport travel trajectory pattern mining. Specific research content and innovative achievements are summarized below: (1) The feature extraction and moving pattern mining from trajectory dataset in spatiotemporal dimension. Respectively, for temporal trajectory data continuity in the time dimension and spatial distribution of the spatial dimension of the semantic distribution of successive approximation conduct research in the time dimension on the other track sampling interval time dimension inherent discrete nature of the proposed acquisition of different time of the hot space to represent the concept of regional time-varying hot zone ; on random sampling intervals in the time dimension trace uneven density distribution characteristics , time -intensive form proposed automatic testing to detect one-dimensional projection data of the time the intensive interval ; thus temporal trajectory in the original data into a hot zone represents the basis of sequence data , designed and implemented to adapt to such frequent intervals closed trajectory moving pattern mining algorithms MTCloSpan, random intervals trajectory moves mode algorithm FMTPM. In the spatial dimension of traditional meshing methods exist Sharp Boundary problem is easy to produce the phenomenon of missing some mobile models proposed fuzzy partition space approach to distance-based membership degree of spatial grid points for multi- track position on the grid matching neighbors calculations, and then based on depth-first and breadth -first search ideas designed with strict time constraints corresponding movement pattern mining algorithm VTPM-PrefixSpan with VTPM-GSP, testing and validation through simulation experiments of the proposed method for solving the problem of the validity of the corresponding . (2) Moving stay pattern mining based on pheromone concentration. Integrated pheromone concentration based on biological ideas to solve the combination of time and space dimension for mobile residence pattern mining problem, drawing biosphere foraging, communication pheromone principle, through integration of computational biology pheromone concentration in order to analyze the spatial distribution of moving objects in space, the residence time distribution, and thus the spatial residence time length parameter calculation regional preference, based on the semantics of data conversion processing to obtain residence time moving sequence data, through the time Notes frequent pattern mining algorithm is proposed to improve the adaptation in spacetime portfolio dimension of mobile travel pattern mining problem TPM algorithms. Through an analysis of the relevant experimental results reveal that the proposed method can effectively track the movement of data mining spatiotemporal stay mode. (3) The representation and moving behavior pattern mining from uncertain spatiotemporal trajectory. In practical applications, the data are often obtained with a moving trace of uncertainty, or expressed in the time dimension; or a spatial dimension. This paper foucus the parobelm of moving behavior patterns from uncertain trajectory data, with uncertain temporal information and uncertain spatial information. At time uncertain temporal trajectory data moving parts , for an interval value indicates uncertain timing information through the strike interval-valued fuzzy set theory for fuzzy membership degree of predicate logic , and thus were thought to depth-first search and breadth -first search ideas design and implement the corresponding pattern mining algorithm FTP-PrefixSpan with FTP-Apriori. Uncertainty in the spatial temporal trajectory moving parts , based on uncertain trajectory - the spatial distribution of the neighborhood by dividing ratio of the area adjacent grid points to build a grid trajectory membership function , using the improved sequential pattern mining algorithm UTFP- PrefixSpan strict constraint condition interval probability sequence data pattern mining . Test data through simulation results show that the proposed method in pattern mining efficiency and scalability parameters has better performance. (4) Analysis of collective moving behavior dynamics in urban environment. In city traffic GPS trajectory data, proposes a multi-scale spatial discretization flexibility division DSPG algorithm, which is based on trajectory OD spatial distribution of points achieved with the movement trajectory semantic matching set of multi-scale spatial division, avoid hard and so dense regions scales meshing right edge of the damage, while enhancing the dense regions with sparse regions distinguish capacity; flexibility through multi-scale spatial discretization division method, the realization of the urban environment group movement dynamics analysis , including groups within different time intervals spatial distribution of mobile activity, mobile activity measurements, and group movement orderly regional connectivity statistical analysis. (5) Mining moving trip behavior pattern of citizens from GPS traffic data. For the urban environment, the group movement travel behavior pattern mining issues in depth discussion and research to multi-scale space elasticity discrete classification method based on building mobile travel global schema is proposed based trajectory model road network topology construction method, by way of probing the gateway node to extract key topological relationship model, in order to achieve the process of moving pattern space conversion; on this basis, proposed moving the global model and the process of moving pattern mining method combining that position by moving the locus OD on with the process of moving sequences respectively move a global model and process models discovery. Shenzhen taxi GPS data through experimental results show that the proposed method is compared with the existing methods in zoning, data conversion, and so has a better performance while mining results semantically richer, stronger interpretability.
语种中文
产权排序1
文献类型学位论文
条目标识符http://ir.sia.cn/handle/173321/14806
专题信息服务与智能控制技术研究室
作者单位1.中国科学院沈阳自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
王亮. 基于时空轨迹数据的移动行为模式挖掘研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2013.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
基于时空轨迹数据的移动行为模式挖掘研究.(4041KB) 开放获取CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[王亮]的文章
百度学术
百度学术中相似的文章
[王亮]的文章
必应学术
必应学术中相似的文章
[王亮]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。