Two-stage surrogate modeling strategy for predicting foundation pit excavation-induced strata and tunnel deformation

Rapid and accurate prediction of strata and tunnel deformation caused by foundation pit excavation is an effective way to mitigate risks in the operation of urban underground spaces. This study proposes a two-stage surrogate modeling strategy, with the aid of predictive capabilities of numerical simulation-based surrogate model and deformation mechanism of underground structures embedded in the modeling procedure. Firstly, a parametric finite element (FE) model of foundation pit-strata-existing tunnel system is established. Subsequently, a first-stage kriging surrogate model is established based on Latin hypercube sampling (LHS) and the aforementioned parametric FE model. Then, another second-stage kriging surrogate model is constructed by utilizing the predictions of the first-stage surrogate model and the deformation mechanism of underground structures. The prediction performances of the two surrogate models and the improvement in performance from this two-stage surrogate modeling strategy were fully investigated via FE simulation examples. The prediction accuracy of strata deformation using the first-stage surrogate model was high at 96.00 similar to 96.25 %, but for tunnel deformation it was low at 88.75 %. However, after using the two-stage surrogate modeling strategy, the prediction accuracy of the tunnel deformation increased to 95.69 %. Finally, this two-stage surrogate modeling strategy was validated using measurement data from a real-world project, the final prediction results are basically consistent with both the field measurement data and the refined FE model. Consequently, this approach is deemed a viable substitute for numerical simulation, offering rapid and accurate predictions to support digital twin management of urban underground space.