基于多输出支持向量机和遗传算法的岩土参数反演方法

提出了基于多输出支持向量机(M-SVM)和遗传算法(GA)的岩土参数反演方法.在进行参数反演过程中,根据正交试验产生一系列参数的可能组合;采用有限元法(FEM)得到坝体不同测点处的位移计算值.将不同的材料参数组合作为输入,相应的位移计算值作为模型的输出来训练M-SVM模型,以便模拟坝体位移和材料参数之间的复杂关系.采用GA来求解材料参数反演所对应的优化问题.在进行最优材料参数搜索的过程中,采用训练好的M-SVM模型来替代有限元计算,以获得坝体位移的计算值,从而大大减少了计算所需的时间.根据某面板堆石坝沉降的监测数据和有限元仿真模型,实现了以上反演过程.反演分析的结果表明,本文提出的岩土参数反演方法具有计算精度高和计算效率高等优点,可以应用于实际工程中.

Geotechnical Parameter Inversion Analysis Method Based on Multi-output Support Vector Machines and Genetic Algorithm

A geotechnical parameter inversion analysis method based on multi-output support vector machines （M-SVM） and genetic algorithm （GA） is proposed. Based on Taguchi design, some possible combinations of material parameters are generated within the admissible range of material parameters. Then with these possible combinations of material parameters, displacement of all the measuring points of the dam is calculated using finite element method （FEM）. Then, the different combinations of material parameters are taken as input and the calculated displacements are taken as output to train some M-SVM models to stimulate the complex relation between the material parameters and the dam displacements. The optimization problem corresponding to the inversion analysis is solved by GA. During the process of searching optimal material parameters, the dam displacements are calculated by the M-SVM mapping instead of the FEM, which may greatly reduce the computation time. Based on the observed settlements and FE model of the dam, the inversion analysis method above is implemented. The results show that the geotechnical parameter inversion analysis method proposed has high accuracy and computational efficiency and can be applied to practical engineering.