Abstract In order to overcome the problems (e.g. more parameters, excessive calculation amount) of the numerical calculation method in the process of solving the limit displacement of the tunnel, the Gaussian Process (GP) is introduced into the calculation of the limit displacement of the tunnel. Meanwhile, the Differential Evolution (DE) algorithm is adopted instead of the conjugate gradient method to search for the optimal GP hyperparameters during the training process so as to overcome the disadvantages that the conjugate gradient method is excessively dependent on initial values, difficult to determine iterations and easy to fall into local optimization, finally a limit displacement prediction method based on the DE-GP (Differential Evolution-Gaussian Process) collaborative optimization algorithm is presented. This prediction method greatly simplifies the solution process of the limit displacement while improving the generalization performance of the Gaussian Process with a single kernel function. Then, the limit displacement of a standard two-lane highway tunnel is calculated and the results show that the DE-GP collaborative optimization algorithm can significantly improve generalization performance compared with single GP and LS-SVM algorithm. The prediction results of the limit displacement are in good agreement with the calculated values and the calculation efficiency is significantly improved, which provides a new way for the solution of the limit displacement.
Abstract：
In order to overcome the problems (e.g. more parameters, excessive calculation amount) of the numerical calculation method in the process of solving the limit displacement of the tunnel, the Gaussian Process (GP) is introduced into the calculation of the limit displacement of the tunnel. Meanwhile, the Differential Evolution (DE) algorithm is adopted instead of the conjugate gradient method to search for the optimal GP hyperparameters during the training process so as to overcome the disadvantages that the conjugate gradient method is excessively dependent on initial values, difficult to determine iterations and easy to fall into local optimization, finally a limit displacement prediction method based on the DE-GP (Differential Evolution-Gaussian Process) collaborative optimization algorithm is presented. This prediction method greatly simplifies the solution process of the limit displacement while improving the generalization performance of the Gaussian Process with a single kernel function. Then, the limit displacement of a standard two-lane highway tunnel is calculated and the results show that the DE-GP collaborative optimization algorithm can significantly improve generalization performance compared with single GP and LS-SVM algorithm. The prediction results of the limit displacement are in good agreement with the calculated values and the calculation efficiency is significantly improved, which provides a new way for the solution of the limit displacement.
LI Xingsheng
.Prediction of Tunnel Limit Displacement Based on DE-GP Collaborative Optimization Algorithm[J] MODERN TUNNELLING TECHNOLOGY, 2021,V58(2): 78-85
2021, Vol. 58 