TY - JOUR

T1 - Prediction of Sinter Chemical Composition Based on Ensemble Learning Algorithms

AU - Niu, Lele

AU - Liu, Zhengjian

AU - Zhang, Jianliang

AU - Sun, Qingke

AU - Schenk, Johannes

AU - Wang, Jiabao

AU - Wang, Yaozu

N1 - Funding Information: This work was supported by the [National Natural Science Foundation of China, 52204335], the [Beijing New-star Plan of Science and Technology, Z211100002121115], the [Central Universities Foundation of China, 06500170], the [Guangdong Basic & Applied Basic Research Fund Joint Regional Funds-Youth Foundation Projects, 2020A1515111008] and the [China Postdoctoral Science Foundation, 2021M690369]. Publisher Copyright: © 2023, The Minerals, Metals & Materials Society.

PY - 2023/7/26

Y1 - 2023/7/26

N2 - Effective prediction of sinter chemical composition enables more timely capture of production abnormalities and reduces the time for process parameters correction, thus optimizing sintering as well as blast furnace production. In this study, two ensemble algorithms, Random Forest and XGBoost, are used to model and predict the sinter chemical composition. The prediction results show that both ensemble algorithms are able to achieve small Mean Square Error (MSE) and Mean Absolute Error (MAE) values with the chemical composition of the raw materials as training parameters, and except for FeO, the prediction hit rate of all other components was above 85%. There is almost no significant difference between the prediction results of the two ensemble algorithms. With the addition of sintering process parameters and fuel/flux parameters as training parameters, reductions in MAE and MSE as well as increases in prediction hit rate were achieved for each chemical composition. Finally, the trained Random Forest model with chemical composition combined with process parameters or fuel/flux parameters achieved prediction hit rate of 94.45%, 83.39%, 95.29%, 89.44%, 90.96%, 97.72%, 99.39%, and 90.35% for Fe, FeO, CaO, SiO2, Al2O3, basicity, MgO, and P, respectively.

AB - Effective prediction of sinter chemical composition enables more timely capture of production abnormalities and reduces the time for process parameters correction, thus optimizing sintering as well as blast furnace production. In this study, two ensemble algorithms, Random Forest and XGBoost, are used to model and predict the sinter chemical composition. The prediction results show that both ensemble algorithms are able to achieve small Mean Square Error (MSE) and Mean Absolute Error (MAE) values with the chemical composition of the raw materials as training parameters, and except for FeO, the prediction hit rate of all other components was above 85%. There is almost no significant difference between the prediction results of the two ensemble algorithms. With the addition of sintering process parameters and fuel/flux parameters as training parameters, reductions in MAE and MSE as well as increases in prediction hit rate were achieved for each chemical composition. Finally, the trained Random Forest model with chemical composition combined with process parameters or fuel/flux parameters achieved prediction hit rate of 94.45%, 83.39%, 95.29%, 89.44%, 90.96%, 97.72%, 99.39%, and 90.35% for Fe, FeO, CaO, SiO2, Al2O3, basicity, MgO, and P, respectively.

KW - Ensemble algorithms

KW - Hit rate

KW - Ironmaking

KW - Prediction

KW - Sinter chemical composition

UR - http://www.scopus.com/inward/record.url?scp=85165699139&partnerID=8YFLogxK

U2 - 10.1007/s40831-023-00717-x

DO - 10.1007/s40831-023-00717-x

M3 - Article

AN - SCOPUS:85165699139

VL - 9.2023

SP - 1168

EP - 1179

JO - Journal of Sustainable Metallurgy

JF - Journal of Sustainable Metallurgy

SN - 2199-3823

IS - 3

ER -