During the operation of the power transformer, the fault location often occurs in the winding, and the fault phenomenon caused by insufficient short‐circuit withstand ability of the winding is particularly prominent. Based on this problem, this article proposes an improved symplectic geometry mode decomposition (ISGMD) algorithm to evaluate the short‐circuit withstand ability of winding. The realization method is to use the algorithm to perform time‐frequency decomposition of the vibration signal of the transformer, calculate time‐domain eigenvalues of the symplectic geometry components to constitute the eigenvector, utilize principal component analysis to reduce dimensionality of the data. Finally, apply the obtained data to train and test the extreme learning machine. The correctness of the algorithm is analyzed by simulating vibration signals. At last, the experiment is designed and the evaluation standard is proposed, and the practicability of the intelligent evaluation model based on the method is verified. © 2021 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.