Research On The Method Of Weather Forecast Based On Support Vector Machines

Since the variation of short-term climate is nonlinear, and the mechanisms and factors influencing it are very complex, in this paper, we research the performance of Support Vector Machines (SVM) on short-term climate prediction which is based on NN-SRM Analysis and statistical learning theory based VC dimension, and make experimental research on its application in meteorology. The following is the main research progress:(1) Take the rich and useful forecasting information including in the one-dimensional time-series observations as the research object, take the average monthly temperature for examples, present the result of the method combing the Empirical Mode Decomposition (EMD) compare with (Least Squares Support Vector machines) LS-SVM, the LS-SVM without decomposition of the EMD method and the RBF neural network method. The experimental results show that the method combing EMD and LS-SVM is better than the others, and has higher prediction accuracy and good generalization ability.(2) In short-term climate forecast research, take northern Zhejiang Province (May to September) monthly rainfall for examples, we compare a new prediction method which combined the Mean Generating Function (MGF) and SVM, the SVM without Mean Generating Function method and the subsection regression method. The experimental results show that the method combing Mean Generating Function and SVM applying in weather forecast is a new study method and direction for the short-term climate forecast.(3) Research on the weather forecast for disaster research by means of SVM classification algorithm. The disaster weather forecast, in winter, is divided into two categories in pattern recognition field (yes or no). We select the weather forecast factors as the samples and establish two class SVM model, and compare it with RBF networks method. The results show that applying SVM to small probability events prediction such as disaster weather has good performance, and it can be further used in practice disaster weather forecast. Meanwhile, the prediction of summer heat or cool summer is divided into three categories:heat, cool summer and normal. We also select the weather forecast factors as the sample, and build multi-classification model, the results show that the SVM multi-classification used in weather forecasting has good prospects.