Hybridized random forest with population-based optimization for predicting shear properties of rock fractures
Jian Zhou,Zhenyu Wang,Chuanqi Li,Wei Wei,Shiming Wang,Danial Jahed Armaghani,Kang Peng
DOI: https://doi.org/10.1016/j.jocs.2023.102097
IF: 3.817
2023-06-18
Journal of Computational Science
Abstract:The fracture in rock mass is considered to be an important factor controlling the shear behaviors of natural rocks, and a clear investigation of its shear parameters is of great significance to maintain the underground structures stability. To this end, this study integrates random forest (RF) model with two population-based optimization algorithms including sine cosine algorithm (SCA) and whale optimization algorithm (WOA) to predict shear strength, peak shear displacement, and dilation angles of rock fractures. 84 direct shear data were utilized to train and test all developed SCA-RF and WOA-RF models. The prediction results showed that SCA-RF and WOA-RF models have obtained great effectiveness in predicting shear strength and dilation angles of rock fractures, as evidenced by their exceptional prediction accuracy in terms of the coefficient of determination (R 2 ), root mean square error (RMSE), and mean absolute error (MAE). To improve the accuracy of forecasting the peak shear displacement, a novel dataset division and a new fitness function were utilized to weaken the partiality of hybrid RF model training. Ultimately, the SCA-RF and WOA-RF models have demonstrated a satisfactory performance for predicting the peak shear displacement, with R 2 of 0.854 and 0.833. The above results demonstrated the applicability and rationality of the proposed models to estimate shear properties of rock fractures.
computer science, theory & methods, interdisciplinary applications