<p>This research investigates the deformation behavior at multi-step elevated deformation by the multi-step uniaxial elevated tensile test from 700 to 850&#xa0;°C &amp; 0.1 to 0.0001&#xa0;s⁻¹. The strain-compensated Arrhenius model, LSTM-Transformer, Transformer and TCN models are developed to predict the flow behavior under different conditions, and RF model is established to predict the grain size, DRX and LAGBs proportion during deformation. The range of peak stress and elongation is 46.35 to 463.55&#xa0;MPa and 46.25 to 242.55%, respectively. The pre-strain promotes generation of the dislocation and voids, resulting in the increase in peak stress and decrease in elongation. Increasing temperature and decreasing strain rate promote DRX, refine grain. But low strain rate also promotes the thermal effect, leading to the grain growth. Moreover, the grain orientation has no significant change during the multi-step deformation, which is different from the single-step deformation. The strain-compensated Arrhenius model only predicts the low-stress curves and cannot predict the flow behavior throughout the entire true strain range. The <i>R</i>² values of three deep learning models are 0.999002, 0.99787 and 0.9979, indicating the excellent performance. Introducing LSTM module into Transformer model enhances the prediction of details of flow behavior. The prediction accuracy from high to low is LSTM-Transformer &gt; Transformer ≈ TCN. The RF model with data augmentation and k-fold cross-validation predicts accurately the grain size, DRX proportion and LAGBs proportion during multi-step elevated deformation. The regularity and range of datasets can obviously influence the model performance. The DRX and LAGBs proportion exhibit the highest prediction accuracy, with that for grain size being marginally lower.</p>