基于点过程模拟的时空级联模式统计挖掘方法

摘要

从时空统计的角度,将时空级联模式的频繁度评价建模为多元独立分布零假设下的显著性判别问题,提出一种基于点过程模拟的时空级联模式统计挖掘方法.首先,采用时空点过程模拟每类地理要素的观测数据集,构建显著性判别的零模型;其次,通过蒙特卡洛模拟获取零假设下每种候选时空级联模式频繁度的实验分布;最后,对候选模式的观测频繁度进行显著性检验,识别显著的时空级联模式.研究结果表明:本文方法能够用于有效识别地理要素间的时空级联模式,且避免了挖掘结果对频繁度阈值设置的依赖.%The discovery of spatio-temporal cascading patterns was modeled as a significance test for prevalence of candidate patterns under the null model of independence, and a statistical approach based on point process simulation was proposed. Firstly, null model was constructed by simulating the point process of different features. Then, the empirical distribution of prevalence of each candidate pattern was estimated by using Monte Carlo simulation. Lastly, significant spatio-temporal cascading patterns were identified by performing the significant test for the observed prevalence. The results show that the approach can effectively detect the meaningful spatio-temporal cascading patterns without threshold of prevalence measure.