CAJ | 학술논문

为了解主分量因子分析在热带气旋强度客观预报中的应用效能,在NCEP再分析资料、T106L19模式产品和热带气旋历史观测资料基础上,采用主分量因子分析技术,结合多元线性回归和BP型人工神经网络,开展了西北太平洋热带气旋的强度客观预报技术研究试验.试验包含完全预报法原理下的预报因子有无主分量分析、线性与非线性预报方法建模等方面的比较.结果表明,预报因子的主分量分析通过降低线性回归和BP人工神经网络模型的维数,提高因子间独立性,可减小模型强度预报平均绝对误差,提高模型实际预报能力.
위료해주분량인자분석재열대기선강도객관예보중적응용효능,재NCEP재분석자료、T106L19모식산품화열대기선역사관측자료기출상,채용주분량인자분석기술,결합다원선성회귀화BP형인공신경망락,개전료서북태평양열대기선적강도객관예보기술연구시험.시험포함완전예보법원리하적예보인자유무주분량분석、선성여비선성예보방법건모등방면적비교.결과표명,예보인자적주분량분석통과강저선성회귀화BP인공신경망락모형적유수,제고인자간독립성,가감소모형강도예보평균절대오차,제고모형실제예보능력.
In order to evaluate the potential efficiency of principal components analysis (PCA) in tropical cyclone (TC) intensity objective forecasting, based on NCEP reanalysis data, T106L19 model forecasting data and CMA historical tropical cyclone observation data, the PCA efficiency are investigated through multiple linear regression and back propagation (BP) neural network focusing on the northwestern Pacific Ocean TC intensity objective prediction technology. This research includes perfect prognostic theory guided factor analysis with and without PCA consideration dimension reduction, linear and nonlinear modeling. Result shows that factor analysis with PCA consideration reduces the models independency. Thus the method lessens average absolute error of TC intensity, and increases the capability of TC intensity objective prediction.