Data-Driven Approach Using a Hybrid Model for Predicting Oxygen Consumption in Argon Oxygen Decarburization Converter

Abstract

Accurately controlling oxygen supply in argon oxygen decarburization (AOD) process is invariably desired for efficient decarburization and reducing alloying elements consumption. Herein, a data-driven approach using a hybrid model integrating oxygen balance mechanism model and a two-layer Stacking ensemble learning model is successfully established for predicting oxygen consumption in AOD converter. In this hybrid model, the oxygen balance mechanism model is used to calculate the oxygen consumption based on industrial data. Then the model calculation error is compensated using an optimized two-layer Stacking model that is identified as (random forest (RF) + XGBoost + ridge regression)-RF model by evaluating different hybrid model frameworks and Bayesian optimization. The results show that, in comparison to conventional prediction model based on oxygen balance mechanism, the present hybrid model greatly improves the control accuracy of oxygen consumption in AOD industrial production. The hit rate and mean absolute error of the present hybrid model for predicting oxygen consumption are 84.8% and 330 Nm³, respectively, within absolute oxygen consumption prediction error ±600 Nm³ (relative error of 3.8%). This data-driven approach using the present hybrid model provides one pathway to efficient oxygen consumption control in AOD process.