Study On Visualization Data Mining Technology Applied In Urban Underground Space GIS System

With the development of urban underground space engineering, a large amount of spatialand non-spatial data is acquired and stored. How to make more effective use of these data, andto find the useful information which is hidden and complied with certain rules, so as to servicegeological prediction in advance and analysis of urban underground space engineering, is animportant research area for data analysis and comprehensive utilization of urban undergroundspace. For this reason, this paper combines data mining and visualization technology to increasethe flexibility, efficiency and interactivity in entire data mining process.There are partial researches on GIS visualization and evaluation methods of spatial datamining technology, but the visualization technology used in data mining is only as anexpression tool for data objects and is no effective visualization in analytic methods and processitself, The relationship between visualization and data mining technology in currentvisualization data mining system is loose. In addition, in the prediction of urban geology inadvance, there are partial researches and some software systems to apply GIS technology in thefield of geological engineering, geotechnical engineering, but some of these systems are onlyused in3D simulation of surface topography, and not very well meet the requirements of usersin geometric modeling, analysis function and interactive function.This paper systematically discusses the key technologies for GIS system of urbanunderground space and construction methods based on visualization data mining technologies,improves machine learning algorithms, clustering algorithm of space and non-space, studies therelative visualization technologies combined with data mining algorithms, and develops a set ofGIS prototype system of urban underground space to support visualization data mining.Themain works are summarized as follows: (1)Study on the technology of visualization spatial data mining. Through comprehensiveanalysis from the angle of the technical characteristics of data mining, mass data characteristicsand data integration with multi-dimension and multi-source, the integration application betweenvisualization data mining and GIS technology is adopted in geological prediction in advance ofurban underground space. In the spatial data mining technology, the spatial data miningmethods based on association rules, support vector machine (SVM) and clustering analysis aremainly used. In the visualization of spatial data mining technology, a visualization method ofmultidimensional multi-temporal spatial data based on parallel coordinates theory is presented,which can well deal with the visualization of mass spatial data, realize the modeling display ofcomplex geologic body by use of Java3D technology, as well as3D display of spatialinterpolation results.(2)Study on support vector machine algorithm. To combine spatial association rule andcase-based reasoning (CBR), the spatial data mining method based on support vector machineis analyzed in-depth, as a starting point by use of GIS technology and spatial data models, twoimproved algorithms are presented to further improve classification accuracy and reducetraining time, such as the SVM with CBR to initial select training subset and the SVM withspatial region partition. Comparison with conventional methods, experimental results show thatthe two algorithms can shorten training time and improve efficiency of spatial data mining inthe case of large amount of data. The SVM algorithm based on space partition can also shortentraining time in a certain extent. In addition, the space classification method based on distancemeasurement is improved for spatial data mining, that is, statistical distance instead ofEuclidean distance, so as to eliminate the misclassification effect from data self correlation.(3) Analysis on classification technology of urban underground space GIS and data qualitycontrol. For the data of point, line and surface from urban underground space, the spatialclustering analysis method based on distance, or mathematical morphology, or topology andspatial association rule is used to classification; and the text classification is realized by theprocess of text pre-processing, feature selection, determining feature-weight and specificclassification. In addition, for the problem of sampling distribution in spatial analysis process,the spatial sampling method based on Sandwich space sampling model is used to simulate and improve that of urban underground space in the data acquisition process, and the purpose ofreducing the cost of geological data acquisition is achieved under the premise of no loss inreliability and accuracy.(4) Development of visualization spatial data mining system. On the basis of design ofdatabase, integration method and data process, the detailed function design on urbanunderground space GIS is completed, and a GIS spatial data mining prototype system based onplug-in form is developed to apply in geological prediction of urban underground space inTianjin. Among them, the use of plug-in software architecture design patterns not only canachieve good performance in software loose coupling and achieve the goal of elastic systems,but also can reduce the cost of systems development remarkably and increase developmentefficiency.