Reconstruction and prediction of tunnel surrounding rock deformation data based on PSO optimized LSSVR and GPR models

Abstract

Predicting the deformation of surrounding rock is an important task to ensure the safety of mountain tunnel construction.This study, set against the backdrop of an actual under-construction tunnel, reconstructed the missing surrounding rock monitoring data using a Particle Swarm Optimization-based Least Squares Support Vector Regression model (PSO-LSSVR), and subsequently predicted the tunnel surrounding rock deformation using the constructed Gaussian Process Regression model (PSO-GPR).The research results indicate that the average relative error of the PSO-LSSVR reconstruction model is 1.21 %, lower than the 4.82 % of the LSSVR reconstruction model and the 4.69 % of the BP reconstruction model. The relative errors of the PSO-LSSVR prediction model and the BP prediction model are 0.55 % and 2.9 %, respectively, both higher than the PSO-GPR prediction model. The PSO-GPR model considers three covariance functions: the Squared Exponential function (SE), the Rational Quadratic function (RQ), and the Matern function (Matern), with relative errors of 0.16 %, 0.15 %, and 0.23 % in the test results, respectively. However, PSO-GPR-SE has a computational efficiency advantage.Overall, PSO-GPR-SE is a suitable model for predicting the deformation of surrounding rock during mountain tunnel construction.

Keywords

• Surrounding rock deformation
• Data reconstruction
• Prediction model
• PSO-LSSVR
• PSO-GPR