Abstract: In order to overcome the shortcomings of traditional methods to determine the parameter of terrane movement in underground metal mines, a new method of the support vector machine (SVM) to predict the motion angle of terrane (MAT) has been proposed. The main factors such as Pu’s coefficient of the hanging wall and foot wall of ore body, geologic conformation, mining depth, mining thickness and deposit angle have been selected as the input variables for the proposed model, and the MAT of hanging wall and foot wall as the output value for the proposed model. On the basis of the different mining conditions and 65 typical metal mines cases, the MAT prediction model of caving method and cut & filling method have been established respectively by using the powerful pattern recognition function of SVM with radial basis function (RBF) kernel. In order to improve the generalization performance and prediction accuracy, genetic algorithm (GA) has been adopted to choose the parameters for SVM model in the current study, thus the MAT with GA-SVM re-gression model for caving method and cut & filling method is established, and Sanshandao gold mine and Sizishan copper mine cases are to be validated for further study of the effectiveness and practicality of the proposed model. The results show that the establishment of SVM regression model prediction of the MAT in underground metal mine can achieve a high accuracy, which provides a new approach to evaluation of the MAT and can be applied to practical engineering, which provides a new approach to evaluation of the MAT.

Key words: underground metal mine, motion angle of terrane, support vector machine, genetic algorithm, prediction