Prediction and optimization of adverse responses for a highway tunnel after blasting excavation using a novel hybrid multi-objective intelligent model

Abstract

It is one of the main challenges of modern traffic engineering to monitor and control the adverse response of highway tunnels after blasting excavation. To that end, this paper proposes a novel hybrid multi-objective intelligent model to predict and optimize four adverse responses including tunnel roof settlement (TRS), linear overbreak (LO), underbreak area (UA), and the maximum block size (MBS) based on a case highway tunnel. This model is assembled by the support vector regression (SVR) and a multi-objective golden jackal optimization (MOGJO) algorithm. 95 monitored data after tunnel blasting containing 11 features of four categories are used to train and test the models. Other five classical intelligence models are also developed to compare the predictive performance of the proposed model. The results of model development indicate that the proposed model has the optimal predictive performance among all models when the ratio of training set to test set is 8: 2. The calculation results of performance evaluation indices (determination coefficient (R2) and root mean square error (RMSE)) demonstrate that the proposed MOGJO-SVR model is the best model for predicting four tunnel adverse responses among all models, i.e., TRS (R2: 0.9837; RMSE: 0.5591), LO (R2: 0.9711; RMSE: 4.5022), UA (R2: 0.9800; RMSE: 0.3803), and MBS (R2: 0.9535; RMSE: 5.9248). Besides, the model interpretation results based on the Shapley additive explanations (SHAP) method show that the contribution of blasting and explosive parameters to the adverse responses is significant. Then, a visualization procedure is generated to optimize the adverse responses for different tunnel excavation designs in real time. This work guides the prediction and control of adverse responses in highway tunnel construction.