Research On Partial Discharge Pattern Recognition Of Converter Transformer Based On Deep Learning

As one of the key equipments of HVDC system,the insulation status of converter transformer is closely related to the safe and stable operation of HVDC system.Partial discharge is not only a sign of insulation deterioration of converter transformer,but also a key factor causing insulation deterioration.Effective recognition of the type and severity of partial discharge in converter transformer can detect some internal latent faults in time,provide a basis for diagnosis and state maintenance strategy,and ensure the safe and stable operation of the system.Therefore,it is of great significance to study the method of partial discharge pattern recognition for converter transformers.Fingerprint feature extraction and classifier selection are the key steps to realize partial discharge pattern recognition of converter transformer.The existing fingerprint feature extraction and classifier selection have some problems as follows:1)In the initial selection of partial discharge feature parameters,due to the lack of prior knowledge,the selected feature parameters have poor sensitivity,and the partial discharge characteristics cann't be fully described.2)In the process of extracting fingerprint features,the commonly used feature space dimension reduction method ignores the nonlinear characteristics of the partial discharge data and the difference of the feature subspace,resulting in more information loss in the feature space after dimension reduction,which affects the accuracy of classifier recognition.3)In terms of classifier selection,the commonly used classifiers are mainly shallow learning methods such as BP neural network and support vector machine.Because of their limitations,these shallow learning methods have limited recognition ability and limited promotion space.These problems lead to the low accuracy of partial discharge pattern recognition of converter transformers.Aiming at the above problems,this paper proposes a partial discharge pattern recognition method based on deep learning.Firstly,based on the experimental data of partial discharge of converter transformer in laboratory,67,13,17 and 97 original partial discharge feature parameters were extracted from the established statistical diagram,the waveform scheme,chromatographic images and wholistic spectra including all above,and the original feature space for recognition research is constructed.Then,based on the three methods of kernel principal component analysis,correlation coefficient matrix and rough set theory,the original partial discharge feature space is reduced and the partial discharge fingerprint features are extracted.Finally,the mainstream model deep belief network in deep learning is introduced into the field of partial discharge pattern recognition of converter transformers.In this part,a deep belief network classifier is constructed according to its basic principle.A layered recognition model based on deep belief network classifier is established.And a partial discharge pattern recognition strategy is proposed to realize the effective identification of the type and severity of partial discharge.Compared with the traditional artificial neural network,the pattern recognition method based on deep confidence network classifier has strong feature learning ability,and the recognition effect is better as a whole.It also provides a reference for the further application of in-depth learning in the field of partial discharge pattern recognition.