Explicit modeling of mechanical property of hot-rolled strip steel based on data-driven and gene expression programming

In order to solve the black-box modeling problem and improve the prediction accuracy of model, two distinguished models for tensile strength (Ts) and yield strength (Ys) of hot-rolled strip steel are established based on the industrial hot-rolled data and the algorithm of gene expression programming (GEP). Firstly, the industrial data of hot-rolled strip steel are preprocessed using the Pauta criterion, so as to eliminate outliers. The key input variables that affect Ys and Ts are selected by using the method of the maximal information coefficient (MIC). Secondly, the explicit prediction models of Ys and Ts are established using GEP. Subsequently, the model results based on GEP are compared with those based on the support vector regression (SVR) and the back propagation neural network (BPNN). Finally, the mathematical expression models for Ys and Ts obtained by GEP are used to further analyse the specific relationships between the chemical composition and mechanical property. It is shown that the errors of Ys and Ts based on GEP are less than 4%, and the coefficient of determination (R²) of Ys and Ts based on GEP is above 0.9, which has strong prediction performance. The prediction accuracy of GEP can achieve the same level with SVR and BPNN. It is worth mentioning that the proposed model can not only show the explicit relationship between the chemical composition, production process, and mechanical property of strip steel, but also occupy high prediction accuracy, which can make reliable reference for strip steel product design and optimisation.