Optimization-based parameter search of support vector regression for high-temperature compression constitutive modeling of 25CrMo4 steel

An accurate intrinsic structural model is essential to describing the high-temperature deformation behavior of metal materials. Support Vector Regression (SVR) has strong regression analysis capabilities, but its application research in constructing constitutive models of 25CrMo4 steel still needs to be improved. In this study, we use grid search, particle swarm optimization, improved genetic algorithm, and improved gray wolf optimization to optimize SVR parameters. A constitutive relationship model for 25CrMo4 steel under high-temperature compression based on SVR was established through training using experimental data models. The predicted data of SVR constitutive models with different optimization algorithms were compared with experimental data. Statistical values, such as average absolute percentage error (AAPE), mean absolute percentage error (MAPE), and correlation coefficient (R2), were introduced to evaluate the accuracy of each model. The particle swarm optimization-SVR model achieved the best performance, with an AAPE of 0.455 38, MAPE of 0.489 09%, and R2 of 0.999 74. Furthermore, compared to other models, it requires the least time. This model has a higher accuracy than other commonly used instantaneous models. These findings can provide a basis for selecting appropriate deformation parameters and preventing hot working defects of 25CrMo4 steel, thus helping to improve the manufacturing process and material properties.