The Research Of Several Key Technologies On Spatial Time Series Data Mining

Spatial Time Series Data Mining, as an important branch of spatial-temporal data mining techniques, is specifically for the study of spatial-temporal data in temporal sequence categories. Not only can it describe the spatial characteristics of geographic objects or phenomena, but also can effectively record the time evolution of those. Consequently, the research in this field is more meaningful. After reviewing relevant research progress and pointing out the existing problems and deficiencies, this paper systematically studies the measurement of spatial time series data mining in theories and methods. To solve the specific problems that exist in the data mining technology, some corresponding strategies are given. Main works are summarized as follows:(1) In the analysis of the spatial sequence clustering. This paper proposed the seed point counted expansion clustering algorithm to match the clustering requirements of timing similar and space adjacent. Firstly, the algorithm selected an object which have the highest sequence similarity among space neighbors as seed, marked it and expanded to the nearside. Then chose the next seed point, repeat the above operation, until all the entities which those spatial and temporal sequence depended have completed the mark. The method is more simple, efficiency and without the class parameter setting, thus avoid the subjectivity of the parameter selection.(2) In the spatial-temporal sequence association rules. This paper proposed an algorithm of constraints to meet the rule of the latter is known, preceding events unknown conditions. A valid time window was firstly set to account the time lag factor both exist between the front target incidents as well as the preceding and latter target events under the conditions of the rear are known. And then concentrated in the calculation of the first event, considering only the candidate of the preceding events that among the latter events’effective time window. Without the need to search the entire sequence of events in the frequent event set, this algorithm avoid the calculation of those frequent set in the entire sequence, thereby effective reducing the complexity of it.(3) For the spatial time series prediction modeling, this paper proposed the STGM(1,1) method to solve the defection of GM(1,1) model for it can model and forecast the small size and poor information sample, but lacking consideration of the spatial autocorrelation. The STGM(1,1) method has combined spatial autocorrelation characteristics with Grey Prediction model. Because spatial autocorrelation was able to describe the space-dependency in spatial objects or phenomena. Thus, this method was useful to handle the small sample points in the spatial-temporal sequence data.Finally, this paper summarizes main findings, and highlights further research directions in the near future.