The effects of hysteresis nonlinear characteristics and creep characteristics of PFSM on control accuracy are analyzed. The XGBoost model is applied to the dynamic hysteresis nonlinear feature modeling of PFSM. By extracting the voltage history characteristics of PFSM, the problem of one-to-many data in hysteresis nonlinearity is effectively solved. The extracted features include voltage of voltage value, frequency of change, direction of change, voltage increment, historical local maximum and local minimum values and time intervals, etc. The simulation results show that the model has good fitting effect on hysteresis nonlinearity and creep characteristics, and the fitting precision is increased by about 40 μrand compared with the static PI model. At the same time, it has the advantages of easy identification of model parameters and easy transplantation to embedded chip. The dynamic hysteresis nonlinear feature modeling method has theoretical significance and practical value.