The invention discloses a transformer partial discharge identification method of affinity propagation algorithm based on manifold distance (AP-MD). The method comprises the steps of constructing threetransformer internal partial discharge models of corona discharge in oil, creeping discharge in oil and air gap discharge; using box dimension and information dimension for the extraction of the feature quantity of a gray-scale map; obtaining the definition of the manifold distance and its calculation formula; performing the method steps of the AP-MD; and obtaining the setting principle of the initial value of the k-nearest neighbor k. The method of the invention improves the shortcoming of difficulty in accurately identifying structural complex data by the traditional AP, and at the same time, is applied to the identification of the three discharge modes of the corona discharge in oil inside a transformer, the creeping discharge in oil and the air gap discharge. The experimental resultsshow that the results of the AP-MD are better than the results of the traditional AP, K-means clustering and fuzzy C-means clustering, and the accuracy of identification is increased.