Per: Guilherme Frederico Bernardo Lenz e Silva (USP), THALES ARANTES KERCHE NUNES (USP), NATALIA PIEDEMONTE ANTONIASSI (USP), RONALDO ADRIANO ALVARENGA BORGES (USP)
Abstract:
There is great interest in accurately modeling the operational variables in the steel- making process on LD converters. However, this is not a simple task since the interation between the process variables is not fully understood, and many decisions within the in- dustry are taken from experience. Therefore, this work’s focus is to provide a reliable model that can, with good accuracy, assist the decision of engineers and technicians in the industry by presenting an estimate of the future behavior of the variables that involve steelmaking in the BOF furnace. Multivariate time series analyzes were used to achieve the objective, among them, the Vectorial AutoRegression models, ElasticNet, K-Nearest-Neighbors, Multiple Linear Regression and Long Short-Term Memory Neural network. These models were applied to a data set of three different steel production campaigns. A model with good performance was found to predict 35 of the 42 proposed variables, proving that it is possible to correlate most of the chosen variables