To enable the diagnosis of the GIS partial discharge fault type, the images of resolved partial discharge (PRPD) and gramian angular field (GAF) were obtained from the GIS partial discharge test using the ultra-high frequency method (UHF) and ultrasonic signals method. A GIS partial discharge type identification algorithm was proposed based on an adaptive convolutional neural network. The hyperparameters of the convolutional neural networks were optimized. A GIS partial discharge type identification model framework was established based on decision-level multi-information fusion. The accuracy and intrinsic influencing factors of different GIS partial discharge type identification methods were studied. The results show that the decision-level multi-information fusion method based on UHF and ultrasonic signals can further improve the identification accuracy of GIS partial discharge types, with the identification accuracy above 98%. The identification accuracy of the proposed decision-level multi-information fusion method is higher than the feature-level multi-information fusion method. The research results can provide technical support for the identification of GIS partial discharge types.