Developing oxygen amount prediction model of basic oxygen furnace steelmaking process with machine learning algorithms

Basic Oxygen Furnace (BOF) steelmaking is a highly complex multiphase physics and chemical reaction process, which aims to reduce the carbon content and increase temperature. During the process amount of oxygen blown into BOF is the most significant control parameter, which effects life of BOF refractory lining, production cost and producing steel in desired quality. In this thesis, the purpose is developing oxygen amount prediction of dynamic blow stage in BOF steelmaking process with machine learning algorithms. For finding the best prediction model, Linear Regression, Multi-Layer Perceptron (MLP), K-Nearest Neighbor (KNN), Extreme Learning Machine (ELM) and Support Vector Machine (SVM) based Sequential Minimal Optimization (SMO) Regression algorithms were compared using Weka and MATLAB. The results of the prediction models developed with gathered production data from BOF are close to each other. ELM and SVM based SMO regression predicted slightly better results than the other models.