摘要
为了有效提高预报精度,将一种基于神经网络并能综合有效利用全场信息的非线性释用技术应用到海洋SST预测上。通过CCA-BP法建立的典型因子,可以代表气象因子场与SST之间的大部分协方差关系,使气象因子与站点要素相关性大为提高,进而通过神经网络技术(BPNN)建立非线性预报模型。利用该模型尝试对热带太平洋表层海温形势(ENSO)进行预报,并建立了该区域逐点海温的预报方案。试报结果表明,该方法对预测春季海温形势有较好的效果,有效预报时效可达1 a以上;对6 a的3月份热带太平洋表层海温预报,平均绝对误差为0.22°C。该方法为海洋SST统计预报提供了一个值得参考的途径。
To improve the accuracy of short-term numerical SST prediction, a nonlinear interpretation model using full-scale information of the numerical products was proved to be effective in raising the accu racy of short-term SST prediction. Canonical variables were constituted using CCA-BP method which mainly represented covariance between factors and the predictive variables. The relativity between the newformed composed factors and the predictive variables increased significantly, hence a nonlinear regression model was developed by BPNN method. The nonlinear interpretation scheme was used to establish the predictive equation for the SST prediction in the tropical Pacific Ocean in the test, and then a scheme for forecasting SST of each point in this region was proposed. The forecast experiment results indicate that the interpretation behaves perfectly well in ENSO forecasting, and the effective period of validity can reach more than one year. For the 6-year March SST prediction of each point in the tropical Pacific Ocean, the average absolute error is 0. 22℃. This method offers a reference for interpretation prediction of SST.
出处
《解放军理工大学学报（自然科学版）》
EI
北大核心
2009年第4期391-396,共6页
Journal of PLA University of Science and Technology(Natural Science Edition)
基金
国家重点基础研究发展规划资助项目(2006CB400505)
国家自然科学基金资助项目(40675040)
关键词
解释应用
海温预测ENSO预测
典型相关
神经网络误差后传算法
interpretation
SST（sea surface temperature） prediction
ENSO（EI Nino suthern oscillation）prediction
canonical correlation
BPNN （back-propagation neural network）
