论文部分内容阅读
结合单桩静力载荷试验的桩传力机理,提出了PSO-SVM算法与有限元联合反演模型。运用PSO优化的SVM预测算法解决了在选择惩罚系数和核函数参数方面的难题。同时基于正交试验与有限元分析获得了神经网络训练样本,建立起土体力学参数与桩顶竖向位移之间的高度非线性映射关系。利用实测位移预测反演得到了土体参数,将其带入正演运算,并与实测位移曲线进行了相似度计算。研究表明PSO-SVM算法在确定土体参数上有一定的适用性,同时PSO-SVM算法与有限元联合反演模型的建立也为参数反演问题的解决提供了一种新的思路。
Combined with the force transfer mechanism of pile in static load test of single pile, the PSO-SVM algorithm and FEM joint inversion model are proposed. The PSO-optimized SVM prediction algorithm solves the problems of choosing penalty coefficient and kernel function parameter. At the same time, neural network training samples were obtained based on orthogonal experiment and finite element analysis, and a highly nonlinear mapping relationship between soil mechanics parameters and pile top vertical displacement was established. The parameters of soil mass were obtained by inversion of measured displacement prediction, which was brought into forward calculation and the similarity was calculated with the measured displacement curve. The research shows that the PSO-SVM algorithm has some applicability in determining the soil parameters. Meanwhile, the establishment of the PSO-SVM algorithm and the FEM joint inversion model also provide a new idea for solving the parameter inversion problem.