Research On Technologies Of Transformer Fault Diagnosis Based On Deep Learning

Since the types and contents of dissolved gas in oil are different when an oil immersed power transformer operates in different state, dissolved gas in oil analysis(DGA) has long been an effective means of judging the state of a transformer. Based on the analysis of the defects for the existed power transformer fault diagnosis methods, the paper proposed a new transformer fault diagnosis method based on deep learning neural networks(DLNNs),which was of strong learning ability, for the first time, so as to provide more accurate reference information for the maintenance of transformers.The paper proposed a new transformer fault diagnosis method based on deep auto-encoder networks(DAENs). Combining DGA data characteristics with corresponding transformer fault types, the paper built trasformer fault diagnosis model and described its realizing steps in transformer fault diagnosis in detail. The method proposed had strong ability of data feature conversion and could generate the fault diagnosis results in probability form.The paper also proposed a new transformer fault diagnosis method based on deep belief networks(DBNs). A deep belief network classifier was constructed and applied into transformer fault diagnosis, and detailed realizing steps in transformer fault judgments was described. The method proposed had strong ability of extracting features from a large number of data samples, which made it possible for taking fully advantage of unlabeled samples obtained by transformer oil chromatogram spectrum on-line monitoring devices, so as to effectively judge the fault types of transformers.The two diagnosing methods of oil immersed power transformers were tested by engineering examples and compared with that based on back propagtion neural networks(BPNNs) and support vector machine(SVM). The results showed that the methods proposed had better performance of fault diagnosis and extension, which could better meet the demands of practical engineering.