A physics-guided GA-BP neural network constitutive model and hot deformation behavior of 12Cr2Mo1V steel

12Cr2Mo1V steel is a core material for manufacturing heavy-duty pressure-bearing components, and precisely characterizing its high-temperature hot deformation behavior is crucial for process parameter optimization and microstructural control. In this study, isothermal uniaxial hot compression tests were conducted at temperatures ranging from 900 to 1200 °C and strain rates of 0.001 to 1 s -1 to systematically investigate the flow behavior and microstructural evolution of this steel. Electron backscatter diffraction (EBSD) characterization confirmed that significant dynamic recrystallization (DRX) occurred during hot deformation, and continuous dynamic recrystallization (CDRX) and discontinuous dynamic recrystallization (DDRX) mechanisms co-occurred to refine the microstructure. At the physical mechanism level, the flow characteristic values were precisely extracted based on the work hardening rate curves, and a traditional Arrhenius constitutive model was established combining the evolution processes of dynamic recovery and DRX. Building upon this, a data-mechanism dual-driven prediction approach was proposed: a backpropagation (BP) neural network model was constructed using material composition and processing parameters as inputs and physical characteristic values as outputs, which was globally optimized by a genetic algorithm (GA). The results demonstrate that the GA-optimized BP model achieved a high coefficient of determination (R 2 ) of 0.95–0.98, significantly outperforming the traditional phenomenological method in terms of prediction accuracy and robust generalization capability within the investigated parameter domain. A multivariable processing map was constructed based on the dynamic materials model (DMM), identifying the optimal hot working parameter window for this steel grade as 1050–1150 °C and 0.01–0.1 s -1 .