Forecast of sinter reduction degradation index and reducibility index and analysis of influencing factors using machine learning

Abstract

Reduction degradation index (RDI) and reducibility index (RI) of sinter are considered as important metallurgical properties for assessing the quality of sintered ore for blast furnace iron-making. For the sake of promoting the permeability of a blast furnace burden and ensuring the smooth smelting process, mathematical models for the prediction of RDI and RI were constructed using machine learning respectively and the effects of factors such as sinter composition on the RDI and RI of sintered ore were analyzed in this article. From simulation results, the precision of the CatBoost model for predicting RDI can reach 98.32%, and the precision of the XGBoost model for predicting RI can reach 93.47%, meaning that the models are effective for the models to forecast the sinter RDI and RI. Moreover, the influence of 16 factors on RDI and RI was analyzed separately based on the SHapley Additive exPlanations (SHAP) method and the accurate predictive models built.