The transformer is one of the most common equipment in power systems, and winding deformation is the most frequent failure of a power transformer. The short-circuit impedance method is the most commonly used method to detect winding deformation. Aiming at the problem that the short-circuit impedance method does not easily detect the small deformation of the transformer, based on the existing ΔU⁃I to construct the Lissajous diagram, ΔP⁃U
2 is proposed to improve the on-line detection of transformer winding deformation Lissajous diagram. The principle of ΔP⁃U
2 improved Lissajous diagram is introduced, and the degree of change of Lissajous ellipse under the same degree of deformation is compared between the two by simulation test. For the influence of the secondary side load on the Lissajous plot, a load normalization algorithm based on image transformation is proposed to eliminate the influence of the secondary side load on the image. The results show that the sensitivity of the improved method is much larger than that of the ΔU⁃I trajectory method, and it has higher accuracy in the judgment of small deformation of the winding, which has a certain engineering value.