用Matlab中的Neural Network Toolbox仿真赤道东太平洋SST的预报模型

基于ＮＣＥＰ／ＮＣＡＲ再分析资料和ＣＯＡＤＳ海洋资料中的全球月平均海平面气压场、８５０hPa纬向风场及海洋温度场，利用Ｍatlab中的Ｎeural Network Toolbox仿真环境和ＢＰ模型改进算法比较准确地仿真和反演出了南方涛动指数、赤道纬向风指数和滞后的赤道东太平洋海温之间的动力结构和预报模型。该模型具有很好的拟合精度和可行的预报效果。可在一定时效内预测赤道东太平洋月平均海温的变化趋势。由于所建系统是具有直接因果关系的预报模型，因此不仅可直接用于预测，而且可有效避免类拟非线性微分方程组在积分过程中由于对初值敏感性而可能产生的对预报结果的不确定性。

IMITATED SST PREDICTION MODEL OF EQUATORIAL EAST-PACIFIC USING NEURAL NETWORK TOOLBOX OF MATLAB

Based on NCEP/NCAR and COADS data set,a dynamic frame and prediction model between equator East-Pacific lag SST and SOI,ELWI is better retrieved and imitated by using BP model and its improved calculation method of Matlab neural network toolbox. The established network model has better coupled accuracy and feasible forecast effect, which can effectively predict the basic development and variation trend of equator East Pacific SST in suitable time range. For the established dynamic system is a prediction model with direct cause/effect function relation,which not only can be used to predict weather system directly, but also can effectively avoid such instability of predicted results caused by initial condition sensitivity of non linear differential equations.