In order to effectively prevent the maloperation of transformer differential relays in non fault transient conditions

a kind of fast identification method to distinguish internal and external faults and their transient processes in transformers is proposed. Firstly

a method to visualize the three-phase differential time series is proposed by using parallel coordinate plot (PCP) to address the issue of potential noise in on-site monitoring data

which

at the same time of containing all fault information

ensures signal noise resistance. Secondly

improvements are made to the CNN: an ECA attention mechanism module is added to the feature extraction section

and the optimized multi kernel support vector machine is used to improve the softmax classifier of the convolutional neural network (CNN)

enhancing the applicability of CNN to various fault scenarios classification and facilitating the diagnosis and identification of internal and external faults in transformers. Finally

a internal and external fault simulation model of transformer is set up to obtain such differential protection monitoring data as winding inter turn faults

wiring faults

excitation inrush currents

current transformer (CT) saturation ferromagnetic resonance and other faults. Based on this

a graphical fault sample set is set up

the proposed method in this paper is verified and compared with similar algorithms so to prove the superiority of the proposed method.