Period Table Based Spatio-temporal Association Rules Mining
Zhou Cheng-hu · Geo-information Science · 2011
As periodical geographical phenomena cover lots of rules,geographic data mining provides a way to find out such rules.In this paper,an algorithm called PRules-Miner is designed based on period table to mine spatio-temporal association rules.Using this mining model,spatio-temporal data were reorganized from sequential dataset to period table set.And spatio-temporal association rules,which describe the tele-connected movement model of two or more objects,can be dug out through three steps: 1) Filtering disorder data in period table: we extract spatio-temporal frequent status in each row and store such status into spatio-temporal frequent item set;2) Matching objects in the item set based on downward closure lemma and spatio-temporal topology: we match the objects in order to create the spatio-temporal association candidate set;3) Verifying the candidate set under spatio-temporal topology to find the rules which have to satisfy the spatio-temporal support and spatio-temporal confidence.And the final rules are the spatio-temporal association rules.To check the validation of the algorithm,we use 20 years' AVHRR Product 016,which is sea surface inversion temperature data provided by PO.DAAC and the same period records of Nanjing's daily precipitation provided by National Academy of Meteorological Sciences to mine the tele-connection rules between Eastern Indo Ocean and Western Pacific Ocean Warm Pool and Nanjing's precipitation.The results show,this mining model has the following characteristics: 1) this algorithm is object-orientated and can describe geographical status independently.Thus,the final spatio-temporal association rules are not correlated with spatial scale or temporal scale.2) The candidate item set is created by Cartesian product,and it can represent complicated spatio-temporal topology between objects.And the spatio-temporal topology can be set manually so as to find the association of none adjacent objects in spatio-temporal dimensions.After setting spatio-temporal topology,spatio-temporal association rules can be mined and validated from candidate set.In the final rules,one object's frequent status is combined with another object's frequent status with given spatio-temporal topology.Thus,the association of objects with uncertain time lag can be extracted.