边坡稳定可靠度替代模型分析  

Analysis of alternative model for slope stability reliability

作　　者：蔡宁[1,2] 赵明华[1] 

机构地区：[1]湖南大学土木工程学院,湖南长沙410082 [2]湖南工程职业技术学院建筑工程系,湖南长沙410002

出　　处：《中南大学学报（自然科学版）》2014年第8期2851-2856,共6页Journal of Central South University:Science and Technology

基　　金：国家自然科学基金资助项目(51278187);湖南省教育厅项目(12C1025)

摘　　要：首先对多项式响应面、BP神经网络、径向基函数及支持向量机这4种替代模型的基本理论、适用范围及构造程序进行具体分析;然后,针对BP神经网络、径向基函数及支持向量机这3种替代模型中训练样本不易获取的情况,引入拉丁超立方试验抽取样本数据,并通过相应替代模型构建边坡可靠度求解的稳定极限状态功能函数近似表达式;最后,通过工程实例对各模型的计算工作量及计算精度进行评估。研究结果表明:对于地层结构较简单、随机参数较少的边坡可采用多项式响应面替代模型,而对于地层结构复杂、功能函数具有较强非线性的边坡建议采用支持向量机替代模型,这可为边坡稳定可靠度替代模型的甄选与构建提供重要依据。Four common alternative models for slope stability reliability analysis, i.e. polynomial response surface model, BP neural network, BF neural network and support vector machine, were systematically introduced, and the basic theory, application range and constructing program of these four alternative models were analyzed. In view of the situation that training sample of the latter three alternative models is not easy to get, Latin hypercube sampling（LHS） was introduced to extract sample data, and the approximate expression of the slope stability limit state function was constructed based on alternative models. Case studies were made to evaluate the calculation load and accuracy of these four alternative models. The results show that polynomial response surface model is applicable to the slope which has simple layer structure and less random parameters, and support vector machine is recommended for the slope which has complicated layer structure and strong nonlinear performance function, which can provide important basis for selection and construction of slope stability alternative models.

关 键 词：边坡工程 替代模型 可靠度指标 蒙特卡洛方法 拉丁超立方抽样 

分 类 号：TU47[建筑科学—结构工程]