摘要
本文在用人工神经网络BP模型对流域年均含沙量进行多因素建模过程中 ,对BP算法进行了改进。在学习速率 η的选取上引进了一维搜索法 ,解决了人工输入 η时 ,若 η值过小 ,收敛速度太慢 ,η值过大 ,又会使误差函数值振荡 ,导致算法不收敛的问题。建模实践表明 ,改进后的BP算法可能使网络误差函数达到局部极小点 。
In this paper we have improved the BP algorithm in the processes of applying BP algorithm of artificial neural networks for the annual average sediment concentration in a watershed by multi-factors. In order to overcome the disadvantage of fixed learning rate η that the BP algorithm does not converge when learning rate η is too big or the rate of convergence is very slow when learning rate η is too small, we introduce one-dimensional optimization method for selecting learning rate. Applying improved BP algorithm, it shows the improved BP algorithm can easily converge into the local minimum point and it can improve the accuracy of model.
出处
《泥沙研究》
CSCD
北大核心
2000年第4期51-54,共4页
Journal of Sediment Research
基金
国家自然科学基金委及水利部联合资助! ( 59890 2 0 0 )
关键词
BP算法
学习速率
年均含沙量
一维搜索法
拟合精度
BP algorithm
learning rate
annual average sediment concentration
one dimensional optimization method
