Frequent Spatiotemporal Association Patterns Mining Based on Granular Computing

In order to discover multi-dimensional spatiotemporal association patterns, and improve the efficiency of traditional mining algorithms for spatiotemporal association patterns, this paper firstly constructs a star association model based on event, which can show more spatiotemporal information on the basis of the present star association model. Besides traditional attributes association with point, line and plane, the model also can fully and flexibly express temporal association, orientation association, and topology association, namely, it can quickly and simply form multi-dimensional spatiotemporal association patterns. And then for the star association model based on event, an algorithm of discovering frequent spatiotemporal association patterns based on granular computing is proposed, which is different from traditional association patterns mining algorithms. One is that the algorithm breaks traditional thinking of generating candidate frequent itemsets, namely, it generates candidate frequent itemsets by updating the mixed radix numeral. The method is quick and simple to avoid redundant complicated calculations for adopting complex FP-tree data structure or generating candidate by joining frequent itemsets. The other is that the algorithm for discovering frequent spatiotemporal association patterns only needs to read database once via granular computing, in other words, it discovers each frequent spatiotemporal association pattern via constructing a spatiotemporal information granule, where the intension can be mapped to the mixed radix numeral from the mixed radix notation system based on spatiotemporal information system. Finally, this paper further discusses the characteristics and the optimal application environments of the algorithm. Experimental results indicate that the algorithm is simpler and faster than these traditional frequent patterns mining algorithms on the optimal application environments.