基于改进BP神经网络的库区渗漏量敏感性分析  

Sensitivity Analysis of Reservoir's Seepage Discharge Based on Improved BP Network

作　　者：钱文江[1] 李同春[1] 丁林[1] 

机构地区：[1]河海大学水利水电学院,南京210098

出　　处：《三峡大学学报（自然科学版）》2012年第6期23-27,共5页Journal of China Three Gorges University：Natural Sciences

基　　金：国家"十二五"科技支撑计划课题(2012BAK10B04);水利部公益性行业科研专项经费项目(201301058)

摘　　要：进行库区帷幕防渗时,为确定合理帷幕范围、渗透系数、厚度和深度,确保帷幕防渗质量和提高经济效益,需要进行渗流量敏感性分析.基于传统BP神经网络,并采用批处理、动量滤波、可变学习速率和遗传算法对之改进,建立网格权重和标准化重要性的关系,确定4种防渗方案中各因素对渗漏量的敏感性大小.对比分析可知,方案4更为经济合理,此方案中影响渗漏量的主要因素为渗透系数,且增加帷幕深度比增加帷幕厚度更为经济有效.While conducting curtain grouting in a reservior bank＇s anti-seepage project, the sensitivity analysis of seepage is required in order to determine a resonable range, permeability coefficient, thickness and depth of the curtain, thus ensuring the quality of the anti-seepage curtain and improving the economic benefits. The paper uses and improves the traditional BP neural network by using batch processing, momentum filter, vari- able learning rate and genetic algorithm to establish the relationships between mesh weights and the standard- ized importance, and then identifies the sensitivity values of various factors related to seepage discharge of the four anti-seepage schemes. By contrasting and analysing it is show that the scheme 4 seems more economic and reasonable. The most important factor influcing the seepage discharge of this scheme is the permeability coefficient; and increasing the curtain＇s depth seems more economic and effective than that increasing curtain＇s thickness.

关 键 词：敏感性分析 渗流量 防渗帷幕 神经网络 遗传算法 

分 类 号：TV223.43[水利工程—水工结构工程]