Hybridized random forest with population-based optimization for predicting shear properties of rock fractures

Abstract

Rock mass fractures play an essential role in controlling the shear behavior of natural rocks. The clear understanding of the fracture's shear parameters is of great significance in maintaining underground structure stability. A novel approach combining the random forest (RF) model and two population-based optimization algorithms respectively (the Sine Cosine Algorithm (SCA) and the whale optimization algorithm (WOA)) was proposed to predict the shear strength, peak shear displacement, and dilation angles of rock fractures. 84 direct shear data was divided into training set and test set for training and testing the proposed models. Coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE), three indexes were utilized to evaluate the model prediction accuracy. The prediction performance indicated that SCA-RF and WOA-RF models achieved high accuracy in predicting the shear strength and dilation angles of rock fractures. These two models also perform satisfactorily in predicting the peak shear displacement by optimizing the dataset division and fitness function. Comparing with Barton-Bandis and Asadollahi-Tonon empirical models, the proposed models have significant advantages in estimating the shear properties of rock fractures. It has important implications for the development of accurate and reliable models for predicting the shear behavior of natural rocks, that some data processing techniques were used based on the characteristics of the direct shear dataset.