RC-XGBoost-Based Mechanical Parameters Back Analysis of Rock Mass in Heavily Fractured Tunnel: A Case in Yunnan, China

The rock mechanics parameters are important indicators for risk assessment in tunnel construction. However, rock mechanical parameters estimated by traditional mechanical testing in the field are subject to considerable uncertainties due to limitations testing technology, which can directly lead to unreliable risk assessment results. Deformation data for parametric back analysis is a reliable and efficient method for estimating mechanical parameters of rock masses. Deformation monitoring data used for parameter back analysis is an efficient and reliable method for estimating rock mechanical parameters, but the current research generally suffers from inaccurate constitutive models, confusing selection of input data, and failure to consider the intrinsic correlation of mechanical parameters. In this paper, a deformation-based back analysis method for rock mechanical parameters was proposed by XGBoost with RC. Taking the Changning Tunnel in Yunnan Province, China, as an example, the construction process of the back analysis model was described in detail. To improve the reliability of the sample data, the Ubiquitous-Joint constitutive model was used and a more characteristic deformed data selection method was proposed. Furthermore, the performance of the back analysis model was improved through Bayesian Optimization of hyperparameters and order filtering of Regressor Chain, and the intrinsic correlation of the parameters was effectively considered through the Regressor Chain. The results showed that applying the rock mechanics parameters obtained from the back analysis model to numerical simulation can effectively predict deformation during tunnel construction, and the error rate of deformation prediction was within 10%, which demonstrates the reliability of the method in this paper. This method can be extended to other heavily fractured tunnels.