Boosting algorithms for predicting end-point temperature in BOF steelmaking using big industrial datasets Boosting algorithms for predicting end-point temperature in BOF steelmaking using big industrial datasets

The application of machine learning was investigated for predicting end-point temperature in the basic oxygen furnace steelmaking process, addressing gaps in the field, particularly large-scale dataset sizes and the underutilization of boosting algorithms. Utilizing a substantial dataset containing...

Full description

Saved in:
Bibliographic Details
Published in:Journal of iron and steel research, international Vol. 32; no. 7; pp. 1856 - 1868
Main Authors: Zhang, Jian-bo, Khaksar Ghalati, Maryam, Fu, Jun, Yang, Xiao-an, El-Fallah, G.M.A.M., Dong, Hong-biao
Format: Journal Article
Language:English
Published: Singapore Springer Nature Singapore 01.07.2025
Springer Nature B.V
Subjects:
Accuracy
Algorithms
Applied and Technical Physics
Artificial intelligence
Basic converters
Continuous casting
Datasets
Engineering
Ensemble learning
Industry 4.0
Machine learning
Machines
Manufacturing
Materials Engineering
Materials Science
Metallic Materials
Multilayer perceptrons
Neural networks
Original Paper
Oxygen steel making
Physical Chemistry
Prediction models
Processes
Steel production
Support vector machines
Steelmaking
Basic oxygen furnace
Boosting algorithm
Machine learning
ISSN:1006-706X, 2210-3988
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The application of machine learning was investigated for predicting end-point temperature in the basic oxygen furnace steelmaking process, addressing gaps in the field, particularly large-scale dataset sizes and the underutilization of boosting algorithms. Utilizing a substantial dataset containing over 20,000 heats, significantly bigger than those in previous studies, a comprehensive evaluation of five advanced machine learning models was conducted. These include four ensemble learning algorithms: XGBoost, LightGBM, CatBoost (three boosting algorithms), along with random forest (a bagging algorithm), as well as a neural network model, namely the multilayer perceptron. Our comparative analysis reveals that Bayesian-optimized boosting models demonstrate exceptional robustness and accuracy, achieving the highest R -squared values, the lowest root mean square error, and lowest mean absolute error, along with the best hit ratio. CatBoost exhibited superior performance, with its test R -squared improving by 4.2% compared to that of the random forest and by 0.8% compared to that of the multilayer perceptron. This highlights the efficacy of boosting algorithms in refining complex industrial processes. Additionally, our investigation into the impact of varying dataset sizes, ranging from 500 to 20,000 heats, on model accuracy underscores the importance of leveraging larger-scale datasets to improve the accuracy and stability of predictive models.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1006-706X
2210-3988
DOI:10.1007/s42243-025-01454-z