论文部分内容阅读
学习机器性能是决定智能位移反分析效果的关键,针对现有智能反分析存在的问题,将高斯过程回归(Gaussian Process Regression,简称GPR)引入隧道工程计算模型参数的反演,并采用单一各向同性核函数之和作为GPR的组合核函数以提高其泛化性能。为克服传统共轭梯度法优化求取最优GPR超参数的缺陷,改用十进制遗传算法替代共轭梯度法在训练过程中搜索GPR最优超参数,并编制了相应的计算程序。结合北口隧道施工监测进行了算法程序的应用,并与进化–单一核函数高斯过程回归算法和进化支持向量回归(SVR)算法的应用结果作了对比,结果表明本文提出的进化高斯过程算法显著提高了反演精度,可以应用于岩土工程计算模型参数的反演辨识,并为类似工程提供了借鉴。
Learning machine performance is the key to determine the effect of intelligent displacement back analysis. Aiming at the existing problems of intelligent inverse analysis, Gaussian Process Regression (GPR) is introduced into the inversion of tunnel engineering calculation model parameters, and a single direction The sum of homosexual kernel functions serves as a combined kernel function of GPR to improve its generalization performance. In order to overcome the shortcomings of the traditional optimal conjugate gradient method to obtain the optimal GPR hyperoptimal parameters, the genetic algorithm instead of the conjugate gradient method was used to search the GPR optimal hyperparameters and a corresponding calculation program was compiled. Combined with the construction monitoring of Beikou tunnel, the application of the algorithm is compared with the application results of the evolutionary-single kernel Gaussian process regression algorithm and the evolutionary support vector regression (SVR) algorithm. The results show that the proposed evolutionary Gaussian process algorithm is significantly improved The inversion accuracy can be applied to the identification of geotechnical calculation model parameters and provide a reference for similar projects.