Research of Transformer Fault Diagnosis Algorithm Based on Stacked Sparse Autocoding

Oil immersed transformer is one of the most significant pieces of equipment in the power transmission and transformation system, and its safety is critical to maintaining the system's stability. A transformer PD platform built in this paper and the signals of high-frequency pulse current, UHF, as well as ultrasonic under PD of three typical defects is obtained. Three groups of characteristic parameters for the three signals to judge the fault type of the transformer are extracted. A stack sparse auto-encoding technique is built to fuse the three sets of feature and diagnose faults. The results show that the constructed algorithm's recognition accuracy is significantly higher than that of the traditional kernel principal component analysis method, and the recognition accuracy of multi-source information fusion is higher than that of single information, with recognition accuracy of 96.5% after the fusion of the three signals.