Application of PCA, SVR, and ANFIS for modeling of rock fragmentation

Abstract

Fragmentation has direct effects not only on the drilling and blasting costs but also on the economy of subsequent operations. In the present study, two soft computing-based models, so called “support vector machines (SVM)” and “adaptive neuro-fuzzy inference system (ANFIS)” were used and compared with Kuz-Ram method. In this regard, six effective parameters including specific charge, stemming length, total delays per number of rows ratio, hole diameter, spacing to burden ratio, and blastability index were considered as input parameters containing a database of 80 variables from the blasting operation of the Chadormalu iron mine of Iran. Principal component analysis (PCA) was performed to clarify the effective parameters on the fragmentation. As statistical indices, root mean square error (RMSE), correlation coefficient (R ²), bias, variance account for (VAF), and mean absolute percentage error (MAPE) were used to evaluate the efficiency of the addressed models between measured and predicted values of rock fragmentation. The results confirmed the ANFIS and SVM as accurate predictive tools for rock fragmentation in open-pit mines. Correlation coefficient, bias, VAF, and MAPE generated by the ANFIS model (respectively 0.89, 0.257, 88.19, and 10.37) were higher than referred values for the SVM model (0.83, 1.87, 75.24, and 16.25, respectively) as well as Kuz-Ram inference.