基于最小二乘支持向量机集成的降水预报模型  

RAINFALL FORECASTING MODEL BASED ON LEAST SQUARE SUPPORT VECTOR MACHINE REGRESSION ENSEMBLE

作　　者：罗芳琼[1] 吴建生[1] 金龙[2] 

机构地区：[1]柳州师范高等专科学校数学与计算机科学系,广西柳州545004 [2]广西壮族自治区气象减灾研究所,广西南宁530022

出　　处：《热带气象学报》2011年第4期577-584,共8页Journal of Tropical Meteorology

基　　金：广西科技厅青年科学基金(NO.0832092);广西教育厅面上项目(NO.200807MS098)共同资助;柳州师范高等专科学校科研基金(No.LSZ2009B005)共同资助

摘　　要：准确的降水天气预报是一个十分重要的研究课题。以中国气象局的T213和日本的细网格数据资料为基础,首先利用粒子群——投影寻踪对众多气象物理因子降维,其次在低维子空间利用四种线性回归方法提取降水系统的线性特征,四种神经网络模型提取降水系统的非线性特征;最后利用最小二乘支持向量机对其集成,对广西6月的逐日降水量进行试验结果表明,该模型预报稳定性好,预报准确率较高,具有较好的业务应用前景。Accurate forecasting of rainfall has been one of the most important issues in hydrological research. In this paper, a novel nonlinear regression ensemble rainfall forecasting model is proposed for rainfall forecasting based on the Least Square Support Vector Machine regression. First of all, rainfall factors of the model are obtained by the Projection Pursuit and Particle Swarm Optimization algorithm in which the Particle Swarm Optimization algorithm optimizes the projection index from high dimensionality to a lower dimensional subspace. Secondly, for rainfall systems, different linear regressions are used to extract their linear characteristics, and different ANN algorithms and different network architecture are used to extract their nonlinear characteristics. Finally, the Least Square Support Vector Machine Regression is used in a nonlinear ensemble model. Empirical results obtained reveal that in terms of the same evaluation measurements, the prediction with the ensemble model presented in this study is generally better than those obtained using other models. Our findings reveal that the ensemble model proposed here can be used as an alternative forecasting tool for meteorological application in achieving greater forecasting accuracy.

关 键 词：投影寻踪 粒子群算法 线性回归 神经网络 最小二乘支持向量机 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]