Research On Transformer Fault Diagnosis Method Based On Multi-source Information Fusion

When performing fault diagnosis on a transformer,due to the intricate mapping relationship between fault characteristics and fault types,it is difficult for a single characteristic parameter to make a comprehensive and accurate assessment of its state.With the continuous maturity and gradual coverage of advanced sensor technology in the field of transformer status monitoring,the amount of transformer status data continues to grow,and the types of features are increasing.This has laid a good data foundation for transformer fault diagnosis based on multi-source information fusion.However,these multi-source features have the characteristics of non-linearity,heterogeneity,ambiguity and uncertainty.The traditional shallow intelligent classification algorithm alone cannot guarantee the diagnosis effect,and the results obtained by a single diagnosis algorithm are relatively one-sided.Therefore,it is necessary to study how to effectively analyze and process multi-source features and improve the diagnostic performance of the classifier.Aiming at the problem that the existing intelligent diagnosis methods only use oil chromatographic information for diagnosis,resulting in insufficient accuracy,the thesis proposes a feature-level fusion method of transformer multi-source fault information based on the concept of feature-level fusion in information fusion technology.The oil chromatographic data is used as the main transformer fault characteristic parameter,and combined with the electrical test and the oil chemical test data to construct a multi-source initial fault feature set to realize the complementarity of the multi-source information.Considering that the redundant information contained in the multi-source features may lead to the degradation of diagnostic performance,a comprehensive feature selection method based on the maximum information coefficient and random forest is proposed to fuse the multi-source features.At the same time,in view of the weak feature mining ability of the core extreme learning machine algorithm fault diagnosis and the difficulty of ensuring the diagnosis effect due to the sensitivity of the single core function,a deep multiple kernel extreme learning machine model is proposed for specific fault identification.Finally,the deep multiple kernel extreme learning machine and its comparison model are used to verify the effect of feature fusion.Aiming at the one-sidedness of the results obtained when a single intelligent algorithm uses multi-source features as input features for diagnosis,the thesis is based on the decision fusion concept in information fusion technology and uses evidence theory to improve it.It builds a deep belief neural network,multiple kernel support vector machine,The deep multiple kernel extreme learning machine performs preliminary diagnosis,and adopts the multi-level information fusion transformer fault diagnosis method of decision fusion through evidence theory,so as to avoid the problem of single information and single method that the diagnosis results are not comprehensive and accurate.The example shows that when the transformer fault diagnosis method based on the multi-source information feature level and the decision level multi-level fusion proposed in the thesis is used for diagnosis,the diagnosis accuracy rate is 94.2%,which is higher than that based on a single fault feature information and a single intelligent diagnosis algorithm.The fault recognition rate can achieve a more comprehensive and accurate assessment of the transformer status,thereby providing the necessary key technical support and scientific guidance for the promotion of the transformer status maintenance mode.