Abstract: The winding is the core component of power transformer,and the mechanical degradation represented by the clamping force looseness can lead to a decrease in the short-circuit withstand capability of the windings. A dynamic model between structural stiffness and load current is established based on the nonlinear relationship between stress and strain of insulating materials under electromagnetic force excitation,and the operating winding structure is equivalent to a parametric vibration system. A recurrent neural network containing structural physical information is proposed for parameter identification,which is further used to extract the dynamic stiffness. The current and vibration data of the windings under different axial clamping forces are obtained by short-circuit load experiments. The experimental results show that when the winding clamping force is small,the dynamic stiffness of the structure under electromagnetic force is significant,and the dynamic stiffness characteristics can accurately reflect the state of the winding clamping force.