Research on Short-range Climatic Forecast Method Based on EMD and SVM

• Main Authors: Shuo-Ben Bi, Yin Xu, Xuan Chen, Bi-Qiang Wang
• Format: Conference Proceeding
• Language: English
• Subjects: Artificial neural networks; Computer science; Empirical Mode Decomposition (EMD); Information science; Nonlinear systems; Predictive models; short-range climatic forecast; Signal analysis; Signal processing; Signal processing algorithms; Support Vector Machine (SVM); Support vector machines; Technology forecasting; time series
• ISSN: 2160-7443
• DOI: 10.1109/ICIC.2010.300

Climate is a nonlinear system, and the BP neural network algorithm or the Support Vector Machine (SVM) algorithm which is superior in dealing with nonlinear problems is usually used in the climate forecast. Meanwhile, the climatic time series also include nonstationary feature, so this paper introduces a new method of signal processing-the Empirical Mode Decomposition (EMD) algorithm for making climatic time series placidly, and combines with the SVM algorithm for short-range climate forecast. At first, the nonstationary time series are decomposed into a series of IMFs with features of stationarity and multiple time scale, then for each IMF component, constructing different models of SVM to forecast, and finally would be straight line fit to final forecast result. This paper uses the anomaly percentage of accumulated precipitation in summer in Guangxi Zhuang Autonomous Region for reality testing, and the result shows that comparing to the direct forecast methods, method of EMD with SVM algorithm has the higher precision and better generalization ability.