摘要
引进BP神经网络优化Bayesian方法中似然函数的计算,得到一种新的BP-Bayesian方法,用来反演河网中各河段糙率.通过一个9河段组成的河网算例,使用本方法得到各河段糙率的后验分布和估计值,最大误差不超过3%;在测量值出现校准误差时,也能有效给出合理的估计值.BP-Bayesian方法能得到糙率估计值的概率密度分布,并从中得到有效的估计值,避免了传统优化方法容易陷入局部最优的缺点;同时,与传统Bayesian方法相比能节省大量计算时间.
BP NN is introduced to simplify the calculation of likehood function in Bayesian arithmetic. A new BP-Bayesian method is proposed to inverse the channel＇s roughness parameters of river network. In the computing example of river network composed of nine channels, estimation and posterior distribution of roughness parameters of the channels are computed based on this new BP-Bayesian method. The maximal error is less than 3 percent. A rational estimation can also be obtained in the existance of measured errors. In conclusion, the probability density distribution of the roughness parameter can be obtained based on the new method. A valid estimation can also be computed to avoid plunging the local optima by traditional optimization. The new arithmetic has great advantage in terms of speediness compared to traditional Bayesian method.
出处
《江苏大学学报（自然科学版）》
EI
CAS
北大核心
2008年第1期47-51,共5页
Journal of Jiangsu University：Natural Science Edition
基金
国家自然科学基金资助项目(50609024)
关键词
河网糙率
BP-Bayesian方法
反演
后验分布
roughness parameter of river network
BP-Bayesian method
inversion
posterior distribution
