A Short-term Climatic Prediction Method Based On Orthogonal Function Decomposition And SVM

The climate system has a complex nonlinear characteristic,and it is a typical non-stationary system.At present,the short-term climate prediction of the monthly and seasonal scale is mainly based on statistical methods.First,it is assumed that the time series is stationary,and then processed the data,Such as,the prediction of temperature and precipitation,which result in the forecast accuracy is generally low.In order to improve the prediction accuracy,this experiment mainly improved the classic support vector machine(SVM)prediction method,and selected the monthly precipitation and temperature data of the 160 National Meteorological stations,and then calculate the seasonal average precipitation and seasonal average temperature data as the test data,which from January 1951 to in December 2015.The main experimental contents and conclusions are as follows:In this paper,first,with seasonal average precipitation and average temperature data,the empirical orthogonal function(EOF)method was applied to separate the time series,which is to be used for prediction.The precipitation data accumulative variance contribution of the first seven principal components is more than 85% and the temperature data accumulative variance contribution of the first five principal components is more than 99%.Than with the time series,the empirical mode decomposition was applied.In the process of decomposition for EMD boundary problem,this paper put forward the extreme extension method to reduce the influence of the non-stationary time series on the prediction.During the prediction,for the support vector machine,cross validation algorithm is added in the machine learning process,then compared the result with the traditional method.The results show that,compared with the traditional support vector machine method,improved method for the change trend and scope of the short-term climate has better forecasting ability and higher accuracy,and it is practical.