Computational Simulation and Prediction on Electrical Conductivity of Oxide-Based Melts by Big Data Mining

Abstract

Electrical conductivity is one of the most basic physical–chemical properties of oxide-based melts and plays an important role in the materials and metallurgical industries. Especially with the metallurgical melt, molten slag, existing research studies related to slag conductivity mainly used traditional experimental measurement approaches. Meanwhile, the idea of data-driven decision making has been widely used in many fields instead of expert experience. Therefore, this study proposed an innovative approach based on big data mining methods to investigate the computational simulation and prediction of electrical conductivity. Specific mechanisms are discussed to explain the findings of our proposed approach. Experimental results show slag conductivity can be predicted through constructing predictive models, and the Gradient Boosting Decision Tree (GBDT) model is the best prediction model with 90% accuracy and more than 88% sensitivity. The robustness result of the GBDT model demonstrates the reliability of prediction outcomes. It is concluded that the conductivity of slag systems is mainly affected by TiO2, FeO, SiO2, and CaO. TiO2 and FeO are positively correlated with conductivity, while SiO2 and CaO have negative correlations with conductivity.