基于均匀设计及神经网络的堆石材料参数反分析  

Back analysis to material parameters of rockfill dams based on uniform design and BP neural network

作　　者：燕乔[1,2] 吴长彬[2] 张岩[2] 

机构地区：[1]中国水利水电科学研究院,北京100038 [2]三峡大学土木水电学院,湖北宜昌443002

出　　处：《人民长江》2010年第15期43-46,共4页Yangtze River

基　　金：中国水利水电科学研究院开放研究基金项目(300m级高面板堆石坝面板开裂机理与控制措施研究)

摘　　要：鉴于堆石坝材料参数反演分析问题的复杂性,在BP神经网络反分析的基础上,利用均匀设计理论构造参数样本,结合有限元分析,对水布垭面板堆石坝主、次堆石料的邓肯E-B模型参数进行反演分析。为了验证反演结果的可靠性,再利用反演分析结果,重新进行有限元计算,将计算得到的坝体沉降位移值与实际监测值相比较,结果显示,有限元计算位移值符合大坝变形的基本情况,反分析结果满足精度要求。说明利用均匀设计结合BP神经网络的反分析方法,可以减少堆石坝材料参数反演分析中网络学习的样本数量,提高反分析效率及准确性。In view of the complexity of back analysis for rockfill material parameters,based on the BP neural network back analysis,we construct parameter samples by uniform design theory,and conduct back analysis on the parameters in Duncan E-B model for filled materials of the Shuibuya′s main and secondary dams in the combination of finite element analysis.In order to verify the reliability of back analysis results,we re-use back analysis results to carry out finite element analysis and compare the calculated and measured dam settlements.The results show that： the calculated settlements accord with the practical settlements and the back analysis results meet accuracy arequirement;the uniform design in the combination of BP neural network can reduce the samples of network learning in the back analysis of rockfill materials′ parameters,and improve efficiency and accuracy of the back analysis.

关 键 词：均匀设计 BP神经网络 反演分析 邓肯E-B模型 

分 类 号：TV641.4[水利工程—水利水电工程]