Abstract: In order to intelligently identify crosslinked polyethylene（XLPE）cable faults in high noise

environment，a method of XLPE cable fault identification based on deep residual shrinkage network was proposed.

In this method，the soft threshold was embedded into the deep structure of the network as a nonlinear

transformation layer，and the soft attention mechanism was introduced to optimize the soft threshold，so as to

enhance the ability of deep neural network to learn features from high noise partial discharge signals，and improve

the accuracy of cable fault diagnosis. Firstly，according to the experience of operation and maintenance，four kinds

of typical terminal faults were made，and a partial discharge test system was built to test the partial discharge data

under different voltage levels and add noise to them. Then，fault data feature extraction and classification under

different noise environments were completed through deep residual shrinkage network. Finally，compared with

other fault diagnosis methods. The results show that the method can effectively suppress the noise signal，greatly

improve the accuracy of cable fault diagnosis in high noise environment，and provide a practical method for the

subsequent engineering application.

      Key words: crosslinked polyethylene（XLPE）cable；partial discharge；deep residual network；attention

mechanism