Application of Surrogate Modeling for SEM Tunneling Simulation

This paper proposes a framework for applying surrogate-based approaches to evaluate SEM construction, which can both benefit the uncertainty analysis in the design stage and real-time calibration during the construction process. The performances of four surrogate modeling approaches, namely back-propagation neural network (BPNN), support vector regression (SVR), random forest (RF), and polynomial chaos expansions combined with Kriging (PCK), were evaluated and compared. A variance-based global sensitivity analysis (GSA) was performed on the surrogate predictor to quantify the impacts of geotechnical parameters on the deformation responses induced by SEM tunneling. The Regional Connector Crossover Cavern project constructed in Downtown Los Angeles was set as the benchmark problem to demonstrate the efficiency and reliability of the proposed procedure.