关键词: 粒子群算法, 小波核函数, 支持向量机, 反演

Abstract: In-situ testing and laboratory testing are two common methods for determining rock and soil mechanics parameters, but there are certain limitations of both methods. The rationality of the parameter selection greatly affects the effectiveness of design calculations and numerical simulation results. The support vector machine method shows obvious advantages on the theoretical basis and solving algorithm. To guarantee the rationality of rock and soil mechanics parameters, the support vector machine method is applied to conduct invert calculations of the rock and soil mechanics parameters. Firstly, the kernel function of the support vector machine is constructed using wavelet analysis theory, and then the support vector machine model parameters of Morlet wavelet, Mexico wavelet and RBF function are optimized using particle swarm optimization(PSO). Finally, the nonlinear mapping relationship between the inversion parameters and the displacement values is established through the wavelet support vector machine model. Based on the orthogonal test and uniform test, this study designs the rock and soil mechanics parameters, which need invert calculation. Meanwhile, by combining the calculation and analysis results through finite element software, the learning samples and the test samples are obtained. After the initial data is compared with the predicted results from the calculations of the Morlet wavelet, the Mexico wavelet, and the RBF function, respectively, it is found that the prediction result of the Morlet wavelet kernel function is more reliable and effective than those of the other two methods. The relative error between the calculated value and the actual monitoring value is no more than 8.1%, when parameters predicted by the Morlet wavelet kernel function are input into the Midas model to calculate the final settlement of the building. The research results show that this method presents good application value in the inverting calculation of geotechnical engineering parameters, and provides a new idea for the determination and verification of rock and soil mechanics parameters in the future.

Key words: particle swarm optimization(PSO), wavelet kernel function, support vector machine(SVM), inversion