Prediction of Sulfur Content in Electric Furnace Matte Using Machine Learning

Abstract

Sulfur content in Electric Furnace Matte is one of the key parameters in nickel matte process control at PT Vale Indonesia Tbk. (PTVI), a nickel matte smelter in Indonesia. Currently the compliance to standard specification is relatively low due to process variability and control limitation. The sulfur content in Electric Furnace Matte depend on sulfur addition and other operating conditions. In this research, a machine learning approach is used to predict sulfur content in Electric Furnace Matte based on selected predictors. Linear and support vector regression models were built on the training data and used to predict sulfur content on testing data. The performance of each models were evaluated and compared. The linear model shows a 0.5843 coefficient correlation between test data and prediction, with a mean square error (MSE) 0.4207. The support vector regression (SVR), a non-linear model, is built with the same predictors. SVR model improve the correlation to 0.9408 and reduce the MSE to 0.0762. The research has shown the practicality of applying machine learning in nickel matte processing and open opportunity for further research.