TBM performance prediction based on XGBoost models: a case study of the ghomrud water conveyance tunnel (Lots 3 and 4)

Abstract

Assessing machine performance is crucial for successful mechanized tunneling projects, particularly when using Tunnel Boring Machines (TBMs). This study presents an innovative approach to model TBM penetration per revolution (P_rev) by integrating novel metaheuristic algorithms with the XGBoost framework. Key geological (UCS, RMR) and machine (Fn, Fr) parameters were selected based on their impact on Prev. A statistical analysis based on a comprehensive database from various geological sections of the Ghomroud tunnel led to the development of Three empirical models, resulting in R2 values of 0.68, 0.67, and 0.70, respectively. Comparing the empirical models with existing ones based on the Ghomrud database, the quality of the developed models is promised. Building on these findings, the study also implemented a hybrid XGBoost model that integrates six optimization algorithms: traditional methods such as Bayesian Optimization (BO), Differential Evolution (DE), and Particle Swarm Optimization (PSO), alongside the newly developed ones: Archimedes Optimization Algorithm (AOA), Harris Hawk Optimization (HHO), and Geometric Mean Optimization (GMO), to improve Prev predictions. The training dataset yielded R² values between 0.91 and 0.95, while the test dataset produced values ranging from 0.93 to 0.97. Notably, while all algorithms effectively predicted Prev, both PSO and GMO demonstrated superior performance across all evaluation metrics. Furthermore, the empirical models proved effective for initial evaluations. Therefore, it is recommended to utilize a combination of the presented empirical models with PSO/GMO-XGBoost for similar projects.