Improved SVM Model for Predicting Pellet Metallurgical Properties Based on Textural Characteristics

Abstract

From the point of view that pellet microstructure determines its metallurgical properties, an improved support vector machine (SVM) model for pellet metallurgical properties forecast is studied based on the mineral phase characteristics, in order to improve the evaluation efficiency of pellet metallurgical properties. The forecast model is composed of a SVM with self-adaptive selection of kernel parameters and a SVM with self-adaptive compounding of kernel types. This not only guarantees the super interpolation ability of the forecast model, but also takes into account its good generalization performance. Based on 200 sets of original sample information, the quantitative relationship between the main characteristics of mineral phase and the grade labels of pellet metallurgical properties (reduction expansion index RSI, reduction index RI, low temperature reduction and pulverization index RDI) was determined by the improved SVM model. With the simulation results of RSI, RI, and RDI with the accuracy of 100%, 98%, and 100% respectively, the precise forecast of pellet metallurgical properties based on mineral phase is realized.