主成分分析在线性模型与非线性模型的应用研究

为了解主成分分析在线性模型与非线性模型预报中的应用效果，在2001—2011年热带气旋历史观测资料基础上，采用主成分分析方法，结合线性回归模型和神经网络模型，开展西北太平洋热带气旋的强度预报技术研究试验．根据提取的主要影响因子构造线性回归模型与BP神经网络的输入样本进行不同样本的台风强度预测．计算结果表明，主成分分析通过降低线性回归模型和BP神经网络模型的维数，减少自变量之间的复共线性，减小模型的预报平均绝对误差．

Linear Model and Nonlinear Model based on Principal Components Analysis and Its Application

In order to evaluate the potential forecast efficiency of principal components analysis （PCA） in linear model and nonlinear model, based on 2001-2011 historical tropical cyclone observation data, the PCA efficiency are investigated through multiple linear regression model and neural network model focusing on the northwestern Pacific Ocean tropical cyclone intensity prediction technology. According to these main factors, the input samples of linear regress model and BP neutral network are definite, and the models could be trained to predict tropical cyclone intensity. Result shows that PCA reduces the models dimension of line- ar regression and BP neural network, and weaken multi-collinearity among the independent variables, and the method based on PCA lessens average absolute error （MAE） of tropical cyclone intensity.