Advanced GEP-probabilistic-based modeling for predicting tunneling-induced groundwater drawdown: A case study of the Uma Oya Multipurpose development project

Abstract

Groundwater drawdown from water ingress or tunnel seepage discharge is a critical challenge in tunneling operations, often causing ground settlement, increased costs, and excavation hazards. This study proposes Gene Expression Programming (GEP) models developed from 13 months of in-situ data to predict groundwater level (GWL) drawdown induced by tunneling. The models incorporate eight critical parameters, including Rock Mass Rating (RMR), rainfall, borehole distance, Poisson’s ratio, and four water ingress factors, applied to the headrace tunnel of the Uma Oya Multipurpose Development Project in Sri Lanka. A four-phase methodology was employed: data preparation, GEP model development with optimized expression trees, model performance analysis, and probabilistic integration using Monte Carlo simulations (MCs). The hybrid probabilistic-GEP model accurately predicts tunneling-induced GWL drawdown, achieving high reliability with an R² of up to 0.964 and low prediction errors (e.g., MAE of 2.20). Sensitivity analysis revealed that water ingress parameters and borehole distance significantly influence GWL drawdown, with a critical threshold at 600 lit/s. MCs enhance reliability by quantifying uncertainties. This approach provides tunnel engineers with a practical tool for mitigating environmental impacts and optimizing water resource management in complex tunneling projects.