To address the limitations of empirical models, which ignore nonlinear interactions among components and lack microscopic validation, as well as the dual challenges of “black-box” nature and data scarcity faced by pure machine learning models, a multi-scale research framework of “data drive, mechanism support, and microscopic validation” was established. A standardized database covering ternary to quinary slag systems was constructed. Ten supervised learning algorithms were compared to identify the optimal model for each system. To overcome the limitations of pure machine learning in the quinary system, a coupled GANPL-CatBoost model was proposed. This model, based on the CatBoost gradient boosting framework, incorporated corrected optical basicity as a physical prior feature and utilized a genetic algorithm to optimize sample weights and hyperparameters, achieving deep integration of physical constraints and data-driven. The mean absolute percentage error (MAPE) for prediction was reduced to 9.431%. Furthermore, molecular dynamics simulations were conducted using the large-scale atomic/molecular massively parallel simulator (LAMMPS). Based on the Born-Mayer-Huggins potential and the Nernst-Einstein relationship, the dominant charge conduction mechanism of Ca2+, the evolution of ion coordination, and the essence of component interactions were systematically quantified at the microstructural level. Consequently, a closed-loop system of “macroscopic prediction, microscopic elucidation, and multi-scale validation” was formed, providing an efficient tool for the intelligent design of CaF2-based slag systems.
在修正光学碱度理论指导下的数据扩充基础上,提出了GANPL-CatBoost(genetic algorithm with NPL-CatBoost)耦合模型.该模型的核心思想是采用在原始数据集表现最优的CatBoost模型对扩充后的数据集进行训练;然后,采用遗传算法调整超参数.结果显示:MSE=0.063,MAE=0.012,MAPE=9.431%(预测值和实验值对比如图7所示).
ZhangG H, ChouK C, PalU. Estimation of sulfide capacities of multicomponent slags using optical basicity[J]. ISIJ International, 2013, 53(5): 761-767.
[2]
ParkJ H, KangY B. Reoxidation phenomena of liquid steel in secondary refining and continuous casting processes: a review[J]. Steel Research International, 2024, 95(11):2300598.
[3]
WalekJ, OdehnalováA, KocichR. Analysis of thermophysical properties of electro slag remelting and evaluation of metallographic cleanliness of steel[J]. Materials, 2024, 17(18): 4613.
[4]
WenT J, RenQ, ZhangL F, et al. Evolution of nonmetallic inclusions during the electroslag remelting process[J]. Steel Research International, 2021, 92(6):2000629.
[5]
DuanS C, LeeM J, SuY, et al Evolution of nonmetallic inclusions in 80 -t 9CrMoCoB large-scale ingots during electroslag remelting process[J]. International Journal of Minerals, Metallurgy and Materials, 2024, 31(7): 1525-1539.
[6]
ChaoD S, JooL, SooK, et al. Oxidation behavior of boron in 9CrMoCoB steel by CaF2-CaO-Al2O3-SiO2-B2O3 electroslag remelting (ESR) type slag[J]. Journal of Materials Research and Technology, 2022, 17: 574-585.
[7]
YuZ G, LengH Y, LuoQ,et al. Geometrical modelling of the physicochemical properties of CaO-Al2O3-CaF2 slag at 1 873 K[J]. Ceramics International, 2020, 46(6): 8075-8081.
[8]
AllibertM, ChatillonC. Mass spectrometry measurements of CaF2 and CaO activities in CaF2-CaO-Al2O3 slags at 1 700 K[J]. Canadian Metallurgical Quarterly, 1979, 18(3): 349-354.
[9]
StovpchenkoG, TogobitskayaD, LisovaL,et al. Predictive models for molten slags viscosity and electrical conductivity based on directed chemical bonds concept[J]. Ironmaking & Steelmaking, 2022, 49(6): 572-580.
[10]
MidhunP M, BasuS. Influence of SiO2 on the structure and melting characteristics of CaF2-CaO-Al2O3-SiO2 slag for use in electroslag remelting[J]. Journal of Sustainable Metallurgy, 2025, 11(2): 941-960.
[11]
HeQ, WenJ H, YangG M, et al. Impurity aluminum removal in industrial silicon by CaO-SiO2-CaF2 slagging refining[J]. Silicon, 2023, 15(12): 5445-5453.
[12]
RokadeS, SinghK, DeshpandeV. Effect of addition of LiNbO3 on the electrical conductivity of Li2O-B2O3 system[J]. Solid State Ionics, 1986, 18/19: 374-377.
[13]
ZhangP, LiuJ H, WangZ, et al. Effect of B2O3 addition on electrical conductivity and structural roles of CaO-SiO2-B2O3 slag[J]. Metallurgical and Materials Transactions B, 2019, 50(1): 304-311.
[14]
ShanZ H, WenJ H, YangG M, et al. Inclusions removal in industrial silicon by SiO2-CaO-Al2O3 based slag system[J]. Silicon, 2023, 15(11): 4889-4896.
WangW, ChangJ S, LeeD J. Machine learning predicts properties of hydrochar derived from digestate[J]. Journal of the Taiwan Institute of Chemical Engineers, 2025, 167: 105862.
[17]
BirolB, PolatG, SaridedeM N. Estimation model for electrical conductivity of molten CaF2-Al2O3-CaO slags based on optical basicity[J]. JOM, 2015, 67(2): 427-435.
[18]
ShiG Y, ZhangT A, DouZ H, et al. Estimation model for electrical conductivity of CaF2-CaO-Al2O3 slags[J]. JOM, 2016, 68(9): 2365-2370.
[19]
LiuY X, LiuJ H, ZhangG H, et al. Experimental study on electrical conductivity of Fe x O-CaO-SiO2-Al2O3 system at various oxygen potentials[J]. High Temperature Materials and Processes, 2018, 37(2): 121-125.
[20]
LiY M, CuiS, WangA, et al. Innovative models for calculating electrical conductivity and viscosity of slag at 1 600 ℃: a case study on CaF2-Al2O3-CaO-MgO-SiO2 slag system[J]. Metallurgical and Materials Transactions B, 2025, 56(5): 5886-5900.
[21]
BaconG, MitchellA, NishizakiR M. Electroslag remelting with all-fluoride low conductivity slags[J]. Metallurgical Transactions, 1972, 3(3): 631-635.
[22]
SibouletB, HocineS, HartkampR,et al. Scrutinizing electro-osmosis and surface conductivity with molecular dynamics[J]. The Journal of Physical Chemistry C, 2017, 121(12): 6756-6769.
[23]
KubisiakP, EilmesA. Estimates of electrical conductivity from molecular dynamics simulations: how to invest the computational effort[J]. The Journal of Physical Chemistry B, 2020, 124(43): 9680-9689.
[24]
SharplessL, MoonJ, ChidambaramD. Optical basicity: a review[J]. Journal of Molecular Liquids, 2024, 394: 123803.
[25]
ZhangG H, XueQ G, LiL F, et al. Study on electrical conductivity and ion diffusion coefficients of CaO-Al2O3-SiO2 melts[J]. Journal of University of Science and Technology Beijing, 2012, 34(11): 1250-1255.
[26]
HaraS, HashimotoH, OginoK. Electrical conductivity of molten slags for electro-slag remelting[J]. Transactions of the Iron and Steel Institute of Japan, 1983, 23(12): 1053-1058.
[27]
HancockJ T, KhoshgoftaarT M. CatBoost for big data: an interdisciplinary review[J]. Journal of Big Data, 2020, 7(1): 94.