With the popularization of partial discharge（PD） detection technology and big data technology

PD detection data of GIS on substation site becomes not only large in quantities

but also complex in sources. Therefore

the traditional partial discharge pattern recognition methods are hardly to meet the demands. Consequently

we proposed a pattern recognition method based on a deep convolutional neural network（CNN） for partial discharge under big data background. In this method

a CNN model applied for PD data pattern recognition was established

in which parameters were initialized by the deep auto-encoder network. The recognition accuracy is aimed to improve by using a multilayer convolutional neural network to extract the deep features of the complex sources data. Lots of PD detection data samples were obtained by means of the experiment on true GIS model

live detection on substation site and the typical interference experiment

to evaluate the proposed method. The results demonstrate that

compared with traditional pattern recognition methods

the proposed method possesses higher accuracy in dealing with pattern recognition targets for complex sources data samples. With the enlargement of data samples

the improved value will further increase

thus this method is more suitable for the engineering application of big data platform.