Original Article

Endpoint carbon content and temperature prediction model in BOF steel-making based on dynamic feature partitioning − weighted ensemble learning

¹ Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming, PR China
² Yunnan Key Laboratory of Artificial Intelligence, Kunming University of Science and Technology, Kunming, PR China
³ Yunnan Kun gang Electronic and Information Science Ltd, Kunming 650302, PR China

^* e-mail: liuhui621@126.com

Received: 24 December 2024
Accepted: 23 March 2025

Abstract

The key to BOF steelmaking control lies in the precise prediction of carbon content and temperature. To address the challenge of dynamic changes in feature importance due to fluctuations in BOF steelmaking production data, a Dynamic Feature Partitioning-Weighted Ensemble Regression Strategy (DFP-WERS) is proposed. First, a genetic algorithm is employed to perform global feature selection across the entire dataset, resulting in global feature data. Second, a mutual information metric strategy based on similarity screening is introduced. This strategy enhances the accuracy and consistency of the correlation between features and target variables by calculating mutual information within a subset of samples similar to those being tested. Additionally, a dynamic feature partitioning strategy is implemented, which partitions global features dynamically based on their correlation with target variables and feature redundancy, thereby adapting to the dynamic changes in feature importance during the prediction process. Finally, a weighted integrated regression prediction method based on partition membership degree is applied. For each test sample, prediction results from each feature partition are dynamically weighted according to the sample’s membership degree. Simulations on actual BOF steelmaking data demonstrate a prediction accuracy of 83.50 per cent within a 0.02 per cent error range for carbon content and 83.00 per cent within a ±10°C error range for temperature.