隧道围岩变形预测的进化高斯过程回归模型

摘要

隧道施工围岩变形预测是关系到隧道施工安全和工程质量的关键,至今已出现多种预测模型,但也存在各种问题.本文将高斯过程回归(GPR)引入隧道施工围岩变形预测以克服现有模型存在的问题,针对目前采用共轭梯度法获得GPR模型最优超参数的缺陷,将十进制遗传算法(GA)与高斯过程回归算法相耦合,采用遗传算法在训练过程中自动搜索GPR模型最优超参数,形成GA-GPR算法,并编制相应的计算程序.为了对比,采用遗传算法与支持向量回归(SVR)算法相耦合,形成GA-SVR算法,将这两种算法程序应用于黄榜岭隧道施工围岩变形预测.计算结果对比表明:本文提出的进化高斯过程回归算法明显提高了预测精度,并为类似工程提供借鉴.%Deformation prediction of surrounding rock is the key to guarantee construction safety and quality. By now,there have emerged many kinds of prediction models, however, various problems have been found with them. The Gaussian Process(GP)holds many advantages such as easy programming,self-adaptive acquisition of hyper parameters and prediction with probability interpretation. The Gaussian process regression ( GPR) is introduced to predict deformation of surrounding rock. Presently,the hyper parameters of GPR are got by maximizing the likelihood function of training samples based on the conjugate gradient algorithm. However, the conjugate gradient algorithm has the shortcomings of too strong dependence on the initial value in optimization, difficult determination of iteration steps and easy falling into the local optimum. -Aiming at overcoming those shortcomings,the genetic algorithm(GA)coded in the decimal system was used to search the optimal hyper parameters of GPR automatically in the training process so as to form the GA-GPR algorithm, and the corresponding codes were programmed in MATLAB. The genetic algorithm was coupled with the support vector regression algorithm to form the GA-SVR algorithm. Both the GA-GR and GA-SVR algorithms were applied in surrounding rock deformation prediction of the Huangbangling Tunnel. It is concluded through comparison that the GA-GPR algorithm improves precision of deformation prediction of tunnel surrounding rock obviously and so it provides reference for similar engineering projects.