Aim at the problem that the uncontrollable interference faced by the partial discharge identification in substation, and the initial parameters of the existing identification method are difficult to determine.Design defects that meet the discharge characteristics of the substation.Multiple sample datas are collected combined with the statistical characteristic parameters extraction method.Based on the Kohonen network with self-organizing competition recognition and strong anti-interference characteristics, new method suitable for partial discharge identification in substation is presented.By exploring the influence of the Kohonen network′s parameters on its recognition effect, the recognition effect is optimized.Then by comparing the network with the commonly used pattern recognition algorithm under the same conditions, high stability and high recognition rate of Kohonen network are proved, and excellent performance in partial discharge identification of substation is verified.