摘要
根据旱涝灾害具有非线性和突变性的演变特征,提出了利用门限值和门限变量构造神经网络的旱涝灾害预报新方法。由于该方法可以根据预报量与预报因子间的不同相关关系构成不同的神经网络学习矩阵,因此实际的预报计算结果表明,该方法对历史样本的拟合和预报精度比一般的逐步回归方法有明显提高。
Based on the characteristics of nonlinear and sudden evolution of drought and waterlogging disasters, a new method for drought and waterlogging prediction is developed by constructing ANN (artificial neural network) with threshold value and threshold variables Because, with this method, different ANN learning matrixes can be constituted on the basis of the correlation between predictand and predictors, the practical prediction results indicate that the fitness to historical cases and prediction accuracy of this method are better than that of traditional stepwise regression means
出处
《灾害学》
CSCD
2003年第2期1-6,共6页
Journal of Catastrophology
基金
国家自然科学基金项目资助(40075021)
关键词
门限回归
突变
人工神经网络
旱涝灾害
threshold regression
sudden change
artificial neural network
drought and waterlogging
