Hybrid data-driven model for predicting the shear strength of discontinuous rock materials

Abstract

The shear strength of the rock discontinuities with different joint wall strengths (DDJS) is one of the important factors in the process of geotechnical engineering construction. This study presents a new data-driven model for predicting the shear strength of DDJS. This model uses conventional rock mechanics properties as inputs, extreme gradient boosting (XGBoost) model as surrogate model, and sparrow search algorithm optimized by levy flight strategy (LSSA) to optimize the hyperparameters of XGBoost model. Based on the collected database, the proposed model (LSSA- XGBoost model) establishes a nonlinear relationship between the shear strength of DDJS and the inputs. Then, the effects of data division ratio and different data preprocessing methods on the model are discussed. In order to verify the validity of LSSA- XGBoost model, it is compared with the original XGBoost model and SSA- XGBoost model. The results show that the LSSA- XGBoost model has high prediction accuracy with coefficient of determination (R) as high as 0.972 and root mean square error (RMSE) as low as 0.075. Moreover, the LSSA- XGBoost model avoids the disadvantage of SSA's optimization search falling into the local optimal value, and its running speed is significantly faster than that of the SSA- XGBoost model. For this database, the minimum-maximum normalization method and the 8:2 division ratio are the most suitable. The findings confirm the potential of this method and its superiority in predicting the shear strength of DDJS.