DE-SVM方法及在黄土滑坡体参数反演中的应用

DE-SVM Method and Its Application in Parameter Inversion of Loess Landslide

根据黄土滑坡体的特点,以及差分进化算法和支持向量机的优点,建立差分进化支持向量机(DE-SVM)反演模型。基于面向对象、自适应、可视化三大关键技术,利用MATLAB软件中的GUI模块编程,开发了滑坡体强度参数反演软件。通过均匀设计方法构造2因素20水平的计算方案,采用简化Bishop法计算出各剖面在不同c、φ值下的安全系数作为学习样本,反演延安新区杜家沟3号黄土滑坡体的强度参数。研究表明:反演得到该黄土滑坡体黏聚力和内摩擦角的值分别为:15.78 k Pa和19.6°;反馈计算得到的滑坡体安全系数值与稳定性评价指标值误差较小,验证反演参数的合理性,也表明所建立的DE-SVM方法适用于黄土滑坡体参数反演;开发的滑坡体强度参数反演软件稳定可靠。。

According to the characteristics of loess landslide and the advantages of differential evolution algorithm and support vector machine,the inversion model of DE-SVM was established. The inversion software of landslide strength parameters was developed based on three key techniques： object-oriented,self-adaptability and visualization. It was programmed by the GUI module of MATLAB software. Two factors and twenty levels of calculation programs were constructed according to the uniform design method. The safety coefficients of each section with different c and φ calculated by the simplified Bishop method were taken as the learning samples. And the strength parameters of No. 3 loess landslide of Du-jia gully in Yan＇an new district were inverted. The results show that： the cohesion and internal friction angle of loess landslide obtained by inversion are 15. 78 k Pa and 19. 6° respectively. The error between the safety coefficients of landslide calculated by feedback and the stability evaluation index is small,which verifies the rationality of parameter inversion and also shows that the established DE-SVM method is suitable for parameter inversion of loess landslide. The developed software is stable and reliable for the strength parameters inversion of landslide.