Research On Feature Analysis And Mining Technology Based On Spatiotemporal Data

With the continuous improvement of information technology and the widespread use of mobile devices and networks,a large amount of geoscience data has been obtained.In recent years,some mature data mining techniques,such as classification prediction,cluster analysis,and association rules,have been gradually used for mining and analyzing time series and spatial structure data.However,compared with traditional data mining technology,spatio-temporal data mining research is still far from mature.How to find suitable data mining algorithms and how to analyze these models requires more effective and practical data management and data analysis technology.In the traditional data mining method,this thesis models the spatiotemporal data and uses clustering technology to analyze the spatiotemporal characteristics.At the same time,based on the results of feature analysis,the method of predicting events is studied.The main research contents are as follows:(1)Research on the organizational modeling method of spatiotemporal data.This thesis extends the traditional space-time cube model and uses space,time and attribute dimensions to express and organize spatio-temporal data,making the data model applicable to multi-level spatio-temporal scales and different aggregation rules.;(2)Study the method of cluster analysis of spatiotemporal hotspots.In different time and space environments,the law of occurrence of events is very different,so nondiscriminatory treatment will weaken the regularity shown by data in different regions.Based on the traffic data of highways,this thesis selects the density-based DBASCAN algorithm for space-time clustering analysis based on the spatial and temporal distribution characteristics of data,so as to realize the reasonable division of spatial hotspots and time hotspots,thus solving the problem of temporal and spatial heterogeneity;(3)Study the hotspot event prediction model based on Bayesian network.In the construction process of Bayesian network,mutual information calculation is added and the traditional mountain climbing method is simplified,so that the operation efficiency is improved under the condition that the structure is guaranteed to be correct.Moreover,the model is trained by high-speed traffic data,and the effectiveness of the prediction results is proved.(4)Designed and implemented a highway road accident analysis and early warning system.The established forecasting model is used to complete the early warning operation of the high-incidence road section,and the multi-dimensional visual analysis is combined with the accident-related data.