Sparrow search algorithm enhanced multi-output regression for predicting rock fracture shear displacements: A metaheuristic-hybridized model

Abstract

This study presents a metaheuristic-hybridized model based on sparrow search algorithm (SSA) and multi-output least-squares support vector regression machines (SSA-MLS-SVR) to predict the continuous shear displacements of rock fractures, which is closely related to the geo-structure stability and safety. Based on the database including 258 continuous shear samples of rock fractures, two subdatasets recording continuous shear displacements in three (216 samples) and four steps (141 samples) are used to develop the proposed model respectively. In the model improvement phase, three kinds of nonlinear transformations are utilized to eliminate the low sensitivity of SSA-MLS-SVR model caused by value scale and data distribution. The experimental results show that the nonlinear transformations can significantly improve the prediction accuracy of SSA-MLS-SVR. Additionally, some characteristics of the continuous shear process of rock fractures are also discovered via extended experiments. First, the prediction accuracy of the first-step shear displacement can be greatly improved using the peak shear displacement as an input. And it is inferred that the beginning of the post-peak shear process (the first-step shear) may be closely related to the peak shear of rock fractures. Second, the post-peak shear process is a continuous random process, in which the displacement of each shear step is only highly correlated with the most recent shear displacements in the time dimension.