<p>The thermal deformation behavior of FV520B stainless steel is investigated. Isothermal compression tests were conducted at temperatures ranging from 600 to 900&#xa0;°C and strain rates from 0.001 to 10&#xa0;s⁻¹. The true stress–strain curves were corrected for friction and temperature due to the drum shape and adiabatic heating. The comparison shows that there is a large difference between the stress before and after the correction, which proves that the correction is necessary. Five constitutive models were developed: the original Arrhenius model, the strain correction Arrhenius model, a new modified Arrhenius model, the back propagation neural network model (BPNN) and the dandelion optimization BPNN model (DO-BPNN). The DO-BPNN model showed the highest prediction accuracy though it was more computationally intensive than the other models. The new modified Arrhenius model performed a better predictive capacity than the strain correction version, while it showed a negligible increase in the number of parameters and computational time. Although artificial neural network-based models exhibit superior accuracy compared to the Arrhenius models, their application in finite element simulations still faces notable challenges.</p>