Endpoint Temperature Prediction model for LD Converters Using Machine-Learning Techniques

Abstract

LD converters are used to improve the quality of melted pig iron. In particular, LD converters inject pure oxygen into hot metal in order to remove impurities via oxidation-reduction reactions. The changes in temperature caused by this redox process can lead to changes in the composition of the iron. Therefore, it is necessary to determine the proper amount of oxygen that would produce the desired composition. This paper provides a machine-learning algorithm for predicting the endpoint temperature of molten steel in a converter and shows relevant simulation results by comparing them with the real operating temperatures. In particular, we introduce a gradient boosting-based decision tree algorithm, a suitable pre-processing method for the data, and an ensemble learning method for several machine-learning models. We also discuss our future work on fringe cases in steel-making processes, which might pose problems for the algorithm to learn to predict accurately.