The status of gas insulated switchgear (GIS) determines the reliability of power equipment operation. Partial discharge is one of the important manifestations for various early-stage latent insulation failures. The traditional partial discharge pattern recognition method relies on expert experience to select the features. The tradtional methed has the disadvantages of strong subjectivity and high uncertainty. To solve this problem, deep learning technology is introduced into the field of partial discharge pattern recognition, which uses convolutional neural network and its extended self-encoding network to extract the characteristics of partial discharge signals and gives full play to the feature extraction ability of self-encoding network. Features are connected with classical classifiers, realizing the organic combination of traditional machine learning method and deep learning method. The basic parameters extraction, statistical feature calculation and discharge type identification of partial discharge signals are realized. The experimental results show that the features extracted by the proposed method significantly improve the classification accuracy and efficiency of partial discharge compared with the traditional artificial features, which has broad engineering application prospects.