边坡系统可靠度分析智能响应面法框架

Framework of intelligent response surface methods for system reliability analysis of slopes

多失效模式情况下,边坡体系可靠性分析较为复杂.提出了基于计算机试验结合机器学习理论建立边坡系统(体系)可靠度分析智能响应面法的一般框架.边坡体系可靠性分析智能响应面法步骤包括:样本生成、模型参数优化、智能响应面建立、蒙特卡罗模拟计算失效概率.对支持向量机、高斯过程等多种智能机器学习理论用于建立智能响应面的性能进行了对比分析.通过典型边坡的体系可靠度分析,验证了小样本情况下智能响应面法在边坡体系可靠性分析中的可行性和优越性.

System reliability analysis of soil slopes is a quite complex task under the multiple failure modes situation. This study proposes a general framework of intelligent response surface methods, which com- bine computer experimental methods with machine learning theories, for slope system reliability analysis. The main steps of the proposed method include, samples generation, model parameters optimization, intel- ligent response surface establishment and Monte Carlo simulation for calculating the failure probability. Performance of several intelligent response surfaces established by support vector machines（SVM） and Gaussian process regression are compared. The proposed methods are tested on three typical slope exam- ples with obvious system effect. The results show that the intelligent response surface methods with small samples are effective for slope system reliability analysis.