Clustering Analysis And Visualization Of Meteorological Data

In recent years,along with the advancement of meteorological industry modernization and informatization,how to store and use vast amounts of meteorological resources,to dig out the useful and valuable information is particularly important,it is also improving weather forecast accuracy,the key to disaster early warning and timeliness and so on various questions.Clustering analysis is a set of physical or abstract objects into multiple classes of similar objects cluster process.K-Means algorithm is a kind of similarity based on Euclidean distance space partition clustering algorithm,think that the smaller the Euclidean distance between two samples of the object of the higher similarity degree.But classical K-Means algorithm in dealing with huge amounts of data,the existence of initial points randomly selected,noise impact and algorithm efficiency and scalability.In this paper,we first use "maximize the minimum distance" thought optimization selection of K initial center;Influence of second noise points against the source data set in this problem,proposed the concept of two sample center of mass and the sample density,to eliminate noise in low density area source data set point;Finally the optimized K Means clustering algorithm and graphs of Hadoop platform distributed programming model,the combination of classical K-Means,in order to solve clustering algorithm in dealing with huge amounts of data algorithm efficiency and scalability problems.Data were first introduced in the experimental stage,speed ratio(Speedup),expansion rate(Sizeup)and the expansion rate(Scaleup)three types of index evaluation HKM clustering algorithm performance,and then use the Grid Analysis and Display System System will clustering results in the form of chart of visual Display.