Machine learning prediction and parameter sensitivity analysis of top earth pressure of high-filled cut-and-cover tunnels

To quickly and effectively predict the unknown top earth pressure of high-filled cut-and-cover tunnels (HFCCTs), a prediction model of top earth pressure of HFCCTs based on machine learning was proposed. The data set was established by taking Poisson 's ratio, friction angle, cohesion, the ratio of groove width to HFCCT width, slope angle and filling height as the input parameters of ML models, and taking the maximum vertical top earth pressure of HFCCTs as the output parameter, and the correlation between the input parameters was analyzed. The Newton-Raphson-based optimization (NRBO) was used to optimize the hyper-parameters of the XGBoost model, and compared with the XGBoost, SVM, RF, BP models under grid search. The SHAP method was used to analyze the sensitivity of the input parameters of the NRBO-XGBoost model. Finally, based on the field measured data of an airport high-speed railway tunnel, the engineering applicability of the proposed earth pressure prediction model was verified. The results revealed that among the input parameters, the filling height was the most influential factor. The prediction performance of NRBO-XGBoost model was better than that of other traditional ML models, and it has good engineering applicability, which can provide an effective method and basis for judging the stability state of HFCCTs in practical geotechnical engineering.