Tunnel squeezing prediction using explainable GWO-XGBoost model

Abstract

[Objective] To achieve accurate prediction of tunnel squeezing, [Methods] an eXtreme Gradient Boosting(XGBoost) model tuned by Grey Wolf Optimization(GWO) was constructed for tunnel squeezing prediction. Training and testing of the GWO-XGBoost model were conducted on an imbalanced dataset with missing data that had undergone imputation and oversampling techniques. The input features of the GWO-XGBoost model included tunnel burial depth(H), rock tunnelling quality index(Q), diameter(D), strength stress ratio(SSR), and support stiffness(K). The performance of the GWO-XGBoost model was rigorously evaluated using a suite of metrics, including accuracy(ACC), the F1 score, the Kappa coefficient, and the Matthews correlation coefficient(MCC). [Results] The result indicated that the presented GWO-XGBoost model achieved an impressive prediction accuracy of 98.94% on both the training set and the test set. Moreover, on the test set, the cumulative value of the evaluation metrics soared to 5.913 1, underscoring the model's exceptional predictive capabilities. The average Shapley Additive exPlanation(SHAP) values for SSR, D, K, Q, and H were 3.06, 1.07, 0.82, 0.73, and 0.51, respectively, indicating that SSR was the most influential feature affecting the model's output result. [Conclusion] The application of the GWO-XGBoost model to the Huzhubeishan Tunnel and Muzhailing Tunnel has yielded squeezing predictions that closely align with the actual conditions observed, proving the high applicability and predictive accuracy of the presented model in tunnel engineering.

Keywords

tunnel squeezing prediction / XGBoost / grey wolf optimizer / model explanation / missing dataset / deformation / influencing factors