基于SIR维数压缩的边坡稳定性神经网络评价  

Evaluation of the Slope Stability Using ANN Based SIR for Dimension Reduction

作　　者：夏元友[1] 蒋超[1] 

机构地区：[1]武汉理工大学土木工程与建筑学院,武汉430070

出　　处：《武汉理工大学学报》2005年第4期58-61,共4页Journal of Wuhan University of Technology

基　　金：教育部科学技术研究重点项目 (10 4 135 );霍英东教育基金 (710 18)

摘　　要：人工神经网络作为一种评价边坡稳定性的半定量方法,由于影响边坡稳定性的因素较多,必然会产生信息量的重叠及失准,从而导致神经网络的规模庞大和泛化能力下降,影响其在边坡问题上的可应用性。针对这一问题,应用分段逆回归(SIR)方法实现BP神经网络的维数压缩,并在SIR方法中采用新的归一化函数进行标准化处理,最终达到减小神经网络规模、提高其泛化能力的目的。通过边坡稳定实例分析,说明该方法能达到满意效果。Being a kind of semiquantitative analysis, artificial neuro network (ANN) can get the non-line map between the independent variable and dependent variable through study, so ANN is an important analytic tool in the research on the slope stability. In this paper, sliced inverse regression (SIR) was proposed in ANN to reduce the dimension and advance the generalization of ANN. And in the process of SIR, a new normalized transfer function was adopted. In the end, ANN based SIR was applied to analyze the slope stability, and the result is satisfiable.

关 键 词：SIR方法 神经网络 维数压缩 边坡 稳定性 

分 类 号：P642.22[天文地球—工程地质学]