The electric upsetting process of large-scale valves is an ultra-long and ultra-time plastic deformation process, it is difficult and important to design the loading path of electric upsetting to achieve coordinated control of deformation and grain size. Therefore, based on the above problem, a finite element model for the electric upsetting process of large-scale valves was constructed based on the electrical-thermal-mechanical multi field coupling analysis theory and the grain size evolution models. Then, the multistage loading paths involving current and upsetting force were designed, and the influence laws of different loading modes on the shape, temperature and grain size of formed parts during electric upsetting were studied. The results show that increasing current significantly increases the temperature and leads to grain coarsening, while larger upsetting force decreases the temperature, thereby refining the grain size. By coordinating the positions and levels of peak values for current and upsetting force, the electric upsetting parts with smooth shape and fine grain can be obtained. In addition, based on the BP neural network model and case-based reasoning method, a design system for the loading mode of electric upsetting process in large-scale valves is developed, and the intelligent design of loading mode for the electric upsetting process parameters is achieved.