Prediction of the maximum ground settlement caused by shield tunneling based on the improved limit learning machine model

Abstract

Excessive ground deformation caused by shield tunneling of urban metro will seriously affect the normal use of surrounding structures, and even cause engineering accidents. In view of the problems that the data dimension in traditional prediction methods is too large, which easily leads to lower accuracy and complex calculation, this study proposes an extreme learning machine (ELM) prediction model based on the principal component analysis (PCA) algorithm and Harris Hawk optimization algorithm (HHO). Ten influence factors are preliminarily selected from the geological, geometric and shield parameters. PCA is used to separate and extract five principal component variables from the 10 dimensional arrays as the input of the model. HHO is used to optimize the input layer weights and hidden layer threshold parameters of the ELM model, and the optimal solution of the prediction model is obtained. The monitoring data of the Yiguang section of Kunming Rail Transit Line 5 are used for simulation verification, and the model is compared with the BP neural network, RBF and non-optimized ELM model. The results show that the root mean square error of the PCA-HHO-ELM prediction model is 0.1435, the average absolute error is 0.0262, and the determination coefficient R2 is 0.9596. Compared with other models, this model has better prediction performance. Compared with the non-optimized ELM, HHO can improve the prediction accuracy and generalization ability of ELM. The PCA-HHO-ELM model can reliably predict the maximum ground settlement induced by shield, and can provide a more feasible new idea for similar deformation prediction.

Keywords

• maximum surface settlement
• principal component analysis
• harris hawk optimization
• extreme learning machine
• settlement prediction