时空序列数据挖掘中若干关键技术研究

发布时间：2018-01-15 23:02

本文关键词：时空序列数据挖掘中若干关键技术研究　出处：《中南大学》2013年硕士论文　论文类型：学位论文

【摘要】：时空序列数据挖掘作为时空数据挖掘的一个重要分支,是专门针对时空数据中时空序列类型的数据进行研究。时空序列数据不仅描述了地理对象或现象存在的空间特征,而且有效地记录了地理对象或现象随时间的演变状态,因此对其研究具有重要的意义。本文回顾了国内外相关研究成果,结合现有的空间数据挖掘与时间序列数据挖掘理论体系,提出了对时空序列数据进行挖掘,探讨了时空序列数据挖掘的主要内容与技术手段,就时空序列数据挖掘的技术中存在的特定问题,提出了相应的解决策略。本文主要工作包括： (1)在时空序列聚类分析研究方向,针对“时序相似,空间邻接”的聚类要求,提出了种子点扩散的时空序列聚类算法,首先选取与空间近邻时间序列相似性最高的对象作为种子,对种子进行标记并且将标记扩散到其空间近邻,然后选取下一个种子点,进行标记、扩散操作,直到所有的时空序列依附的实体都被标记,该方法计算简单、效率高并且无需进行参数的设定,避免了参数选取的主观性。 (2)在时空序列关联规则研究方向,针对“后件已知,前件未知”的关联条件,提出了一种约束条件下事件关联规则算法,首先在后件目标事件已知的条件下,通过一个有效时间窗口来顾及前件事件间及前件事件与后件目标事件在时间上的滞后因素,然后在计算前件事件集中,只考虑对后件目标事件有效时间窗口中的候选前件事件集,而不需要对整个事件序列中的频繁事件集进行搜索,避免对整个事件序列中的频繁集计算,从而降低了算法的复杂度。 (3)在时空序列预测建模研究方向,针对GM(1,1)能够对“小样本,贫信息”进行建模预测,而缺乏对空间自相关性的考虑,提出了STGM(1,1)建模方法,此方法是结合空间自相关特性与灰色理论预测模型,空间自相关性是对空间对象或现象在空间上的依赖性描述,因而STGM(1,1)能够处理具有小样本的时空序列数据。最后,总结了本文的研究成果,并展望了时空序列数据挖掘进一步需要研究的工作。
[Abstract]:As an important branch of spatio-temporal data mining, spatio-temporal sequence data mining is an important branch. Spatio-temporal data not only describe the spatial characteristics of geographical objects or phenomena, but also focus on the types of spatio-temporal data. And the evolution of geographical objects or phenomena with time is recorded effectively, so it is of great significance to study them. This paper reviews the relevant research results at home and abroad. Combined with the existing theory system of spatial data mining and time series data mining, this paper puts forward the mining of space-time series data, and discusses the main contents and technical means of spatio-temporal series data mining. In view of the specific problems existing in the technology of spatio-temporal series data mining, the corresponding solutions are put forward. The main work of this paper is as follows: 1) in the research direction of spatio-temporal sequence clustering analysis, aiming at the clustering requirements of "temporal similarity, spatial adjacency", a spatio-temporal sequence clustering algorithm based on seed point diffusion is proposed. Firstly, the object with the highest similarity to the spatial nearest neighbor time series is selected as seed, the seed is labeled and the marker is diffused to its spatial neighbor, and then the next seed point is selected for marking and diffusion operation. Until all the objects attached to the spatio-temporal sequence are marked, the method is simple and efficient, and does not need to set parameters, thus avoiding the subjectivity of parameter selection. 2) in the research direction of temporal and spatial sequence association rules, an event association rule algorithm based on constraint condition is proposed for the association condition of "the posterior part is known, the previous part is unknown". First, under the condition that the target event is known, the lag factor between the former event and the latter object event is taken into account by an effective time window, and then the time lag factor between the former event and the latter object event is considered, and then the time lag factor is considered in the calculation of the previous event set. Only the candidate pre-event set in the effective time window of the latter target event is considered, and the frequent event set in the whole event sequence is not searched to avoid the calculation of the frequent set in the whole event sequence. Thus, the complexity of the algorithm is reduced. 3) in the research direction of prediction modeling of time-space series, aiming at "small sample, poor information" can be modeled and forecasted, STGM(1 is put forward because of the lack of consideration of spatial autocorrelation. 1) Modeling method, which combines spatial autocorrelation characteristics with grey theory prediction model, spatial autocorrelation is the spatial dependent description of spatial objects or phenomena, so STGM(1. 1) capable of processing temporal and spatial sequence data with small samples. Finally, the research results of this paper are summarized, and the further research work of spatiotemporal series data mining is prospected.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P208

【参考文献】