计算机工程与应用 ›› 2007, Vol. 43 ›› Issue (31): 192-195.

• 工程与应用 • 上一篇    下一篇

模型输入对模糊神经网络预报模型的影响研究

金 龙,史旭明   

  1. 广西气象减灾研究所,南宁 530022
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2007-11-01 发布日期:2007-11-01
  • 通讯作者: 金 龙

Study on impact of model inputs on performance of fuzzy neural network prediction model

JIN Long,SHI Xu-ming   

  1. Guangxi Research Institute of Meteorological Disasters Mitigation,Nanning 530022,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-11-01 Published:2007-11-01
  • Contact: JIN Long

摘要: 为了探索模型输入对模糊神经网络预报模型预测性能的影响,研究了通过减少预报模型自变量组合的复共线性影响,并结合相似系数计算分析方法建立了一种新的模糊神经网络预报模型。以气象学科的逐日降水预报作为研究对象,利用这种新的模糊神经网络预报模型进行了实际预报试验,并与常规的模糊神经网络预报方法,中国气象局T213数值预报模式以及逐步回归预报方法的预报结果进行了对比分析。结果表明,这种基于条件数和相似系数计算的模糊神经网络预报新方法对49天降水的独立样本预报平均绝对误差为7.33 mm,预报误差比模糊神经网络预报模型下降了5.9%,比传统的逐步回归方法下降了14.9%,比中国气象局T213数值预报模式的预报结果下降了13.4%。显示了很好的应用前景。

Abstract: Owing to the accuracy of traditional Fuzzy Neural Network(FNN) forecasting model is limited by the existence of multicollinearity among matrix samples,a novel FNN model is proposed where the conditional number algorithm is used,and the training samples are chosen by means of similar coefficient computation. The case computational results of regional mean rainfall show that the mean absolute error of independent samples of 49 days for the novel FNN prediction model is 7.33 mm,and under the condition of same predictors and prediction samples its prediction error is 5.9%,14.9% and 13.4% lower than those for FNN forecast model,stepwise regression forecast model,and T213 numerical weather forecast products from China Meteorological Administration,respectively,thus showing a good prospect of operational application.