土工基础

›› 2016, Vol. 30 ›› Issue (2): 196-200.

• 工程实录 • 上一篇    下一篇

基于粒子群优化-高斯过程回归的智能岩土体参数快速反演方法

黄 伟,刘 华   

  1. （中铁第四勘察设计院集团有限公司，武汉 430063）
  • 收稿日期:2016-01-19 出版日期:2016-04-20 发布日期:2016-04-27
  • 作者简介:黄 伟（1988-），男，助理工程师，研究方向为结构设计。

An Inversion Method Using PSO and GPR for the Rapid Determination of the Geotechnical Parameters

HUANG Wei, LIU Hua   

  1. （Urban Rail Transit and Underground Engineering Design Research Institute, China Railway Siyuan Survey and Design Group Co. Ltd., Wuhan 430063）
  • Received:2016-01-19 Online:2016-04-20 Published:2016-04-27

摘要： 针对盾构隧道结构设计中勘察报告提供各土层c、φ值不准确或缺失，岩土勘察结论不明确等问题，提出了一种粒子群优化（PSO）与高斯过程回归（GPR）机器学习方法的协同优化算法（PSOGPR）方法。该方法采用GPR模型来建立自变量与计算函数值之间的映射关系，并作为函数计算工具替代岩土力学计算中的高耗时的问题，实现PSO寻优加速。通过对南京地铁轨道交通1号线（七桥翁站到小天堂站区间）下穿宁铜铁路工程实例验证了反演方法的可行性，研究结果表明，对比基本PSO算法，该方法显著的减少了参数反演过程中的有限差分计算次数，计算效率明显较高，其对于复杂的岩土体参数反演问题具有良好的应用能力。

关键词: 岩土工程, 参数反演, 粒子群优化, 高斯过程

Abstract: To solve the problem of inaccuracy or missing information of the c and φ values for various soil strata in the geotechnical report for the shielded tunnel structures, a cooperative optimization algorithm based on the particle swarm optimization (PSO) algorithm and the Gaussian process regression(GPR) machine learning is presented. In the optimization process, the algorithm uses the approximation model of GPR to establish the relationship among the decision variables and the fitness functions. This relationship can be used as the approximate model of GPR as an evaluation tool of to resolve the time consuming problem in the engineering and can be the partial optimization during the global optimization process of PSO. This method was verified in the evaluation of the tunnel construction under cross the Ningtong Railway of the No 1 Line of Nanjing MetroRail Transportation Project. The results indicate that, compared with the basic PSO algorithm, the proposed method reduces the iteration numbers of finite difference for the back analysis and increase the efficiency. 

Key words: Geotechnical； Parameter Back Analysis, Particle Swarm Optimization (PSO), Gaussian Process Regression (GPR)