摘要
介绍一种新的非线性回归分析方法——SVM回归。利用EOF能分解数据场和SVM回归分析可建立因子与预报量非线性关系的优势,设计预报方案:(1)将因子场和预报场分别用方差标准化、EOF场展开,提取两场时间系数;(2)用SVM回归分析实现因子场时间系数对预报场时间系数非线性预测;(3)由预测的预报场时间系数与对应空间函数反演原场。用交叉检验的方法,对1960～2003年1月热带海表温度场预报汛期(6～8月)华中区域降水场进行试验。SVM回归44年独立预报平均技巧评分10.4%,较随机预报具有明显的技巧水平,优于经典回归。
A new nonlinear regression, SVM regression, was introduced. With the superiority of both of EOF （Empirical orthogonal functions） separating fields and nonlinear SVM regression forecasting a program is projected： （1） factor fields and predicted fields are standardized, then EOF, and the time coefficients of two fields are extracted respectively; （2） with SVM regression the time coefficients of predicted fields are estimated by those of factor fields; （3） the original predicted fields are recovered by linear combination of the time coefficients and the eigenvectors. Summer rainfall over central China was predicted with January tropical sea surface temperature, and the cross-validations over 44 years were tested. The score is 10.4%, and this program is obviously superior in forecasting skills to both of random and classical regression.
出处
《热带气象学报》
CSCD
北大核心
2006年第3期278-282,共5页
Journal of Tropical Meteorology
基金
国家自然科学基金项目(60072006)
关键词
支持向量机
回归分析
场估计
support vector machines
regression
estimate of fields
