【作者】 袁冠；

【导师】 夏士雄；

【作者基本信息】 中国矿业大学 ， 计算机应用技术， 2012， 博士

【摘要】 近年来，随着GPS设备、RFID传感器、卫星和无线通信等技术的快速发展，全球范围内的各种大小的移动对象都可以得到有效跟踪，由此产生了越来越多的移动对象轨迹数据被收集并存储在移动对象数据库中。这些数据蕴含着大量的信息，迫切需要研究人员对其进行有效地分析。本课题以移动对象轨迹数据为研究对象，以移动对象的活动模式发现为研究的主要目标。本课题研究的工作主要包括如下几个方面：1．以移动对象轨迹数据挖掘的目标和任务为驱动，在分析现有移动对象数据挖掘系统的目标和特点的基础上，深入研究现有移动对象数据挖掘相关理论、方法，提出了一种新的移动对象周期行为活动挖掘系统框架，能够从不同层次对移动对象轨迹数据展开分析和挖掘，并发现移动对象的活动特点。2．针对现有移动对象数据挖掘方法过分追求效率而忽略轨迹运动特征问题，提出了一种基于结构特征的轨迹分析方法，该方法从微观角度对移动对象的运动模式和轨迹特征进行分析。通过抽取轨迹的结构特征对移动对象的运动轨迹进行比较，能够从更全面的角度分析对象的运动特点，此外，通过设置轨迹的结构权重，可以灵活地调整轨迹结构的敏感程度，从而快速、高效、灵活地对移动对象的运动轨迹进行分析。3．为深入分析移动对象的活动特点，提出了一种基于协同过滤的移动对象兴趣活动发现算法。该方法从宏观角度对移动对象的兴趣活动以及兴趣路径进行发现。该方法首先对移动对象的轨迹数据进行建模，给出了移动对象活动的热点区域发现算法，解决了对象活动发现和表示的问题。通过借助协同过滤算法，发现在兴趣活动上较为相似的对象，并以近邻的历史活动为基础推荐对象潜在的兴趣活动。在对象兴趣活动的基础上，引入时间标记的最大公共子模式方法，发现近邻对象之间的兴趣活动路径。4．通过对移动对象活动的时空特点进行分析，研究移动对象活动的周期特性，提出一种基于多粒度的移动对象周期活动发现方法，用于在多种时空粒度下对移动对象活动进行周期活动发现。该方法对对象的活动进行时空建模，并给出多粒度活动发现算法，对移动对象的活动进行多粒度表示。该方法不仅能在能够在周期未知的情况下发现单对象的活动周期，还能够发现关联活动的周期。5．通过构建移动对象轨迹数据挖掘原型系统，实现了移动对象轨迹数据的一系列分析，并结合煤矿领域关于人员定位相关的需求，给出了移动对象数据挖掘在煤矿领域的启发式应用，有效地将移动对象数据挖掘方法与实践进行了结合，印证了移动对象轨迹数据挖掘相关方法的可行性和有效性，为多领域移动对象数据挖掘理论和方法提出了新的思路和新的探索。更多还原

【Abstract】 In recent years, with the rapid growth of GPS devices, sensor network, satellitesand wireless communication technologies, various kinds of moving objects can betraced all over the world. At the same time, more and more moving objectstrajectories are collected and stored in database. These data often contain a great dealof knowledge, which need an urgent analysis. This dissertation takes moving objectdata mining as research object and considers the discovery of moving objects’periodic activities as main goal. The main research works are listed as follows:1. Driven by the goals and tasks of trajectory data mining for moving object,and under analyzing the characteristics of current existing moving object miningmethodologies, a novel framework of moving object periodic activity mining ispresented. With the framework, moving object trajectory data can be furtheranalyzed and mined from different aspects, and full moving object activities can befound.2. A trajectory analysis method based on structure features is presented toovercome the shortages existing in current algorithms. This method analyzes movingobjects’ movement patterns and trajectory features from microcosmic viewpoint. Bycomparing the extracted structure features from motion trajectory, this method cananalyze objects’ motion features from different angles. Moreover, setting trajectorystructure weights makes the sensitive degree of trajectory structure more easilyadjusted, and motion trajectory of moving objects also can be analyzed faster, highefficient, comprehensive and more flexible.3. In order to analyze object’s activity in deep view, an interesting activity ofmoving objects discovery algorithm based on collaborative filtering is put forward.This method discovers moving objects’ interesting activities and interesting routesfrom macroscopic viewpoint. Firstly, hot regions discovery algorithm is given totransform sporadic and redundant trajectory data into activities sequence. Thenobjects’ potential interesting activities are recommended on a basis of neighbors. Themethod also makes use of largest common sub-patterns to discover interestingactivity routes among neighbor objects, which lays a solid foundation for furtherresearching objects’ activities.4. A periodic activities discovery method based on multiple granularities.Moving objects’ activities sequence is multi-granularity modeled on a basis of objects’ interesting activities discovery. By space priority algorithm of multiplegranularity activities discovery and time priority algorithm, the activities arerepresented using multiple granularities. A new periodic pattern discovery algorithmof single activity is proposed to find objects’ activity period with unknown periods.In addition, Max Sub-pattern Tree is introduced to discover periodic pattern ofobjects’ linked activities more flexible and high efficient.5. Finally, this dissertation designs and develops a trajectory data miningprototype system. Combining with the requirements of mine personnel position, themethods and theories are heuristic applied and verified in mine personnel positionsystem, which verifies the feasibility and effectiveness of correlation mining methodabout moving objects’ activities. The proposed methods provide new ideas and waysto explore the theories and techniques in moving object data mining.更多还原