Transformer Fault Diagnosis Method Based on Self-Powered RFID Sensor Tag, DBN, and MKSVM

Abstract

A novel transformer fault diagnosis method using self-powered radio frequency identification (RFID) sensor tag, deep belief network (DBN), and multiple kernel support vector machine (MKSVM) is presented. The self-powered RFID sensor tag is applied to measure transformer vibration signals, which has advantages of convenience, long-term monitoring, low power consumption, and fast location. Then, the DBN is employed to extract features from the measured signals, and the DBN method’s extraction performance is improved by optimizing its learning rates and momentum factors. The extracted features of the same fault are highly centralized, and the features of the different faults are obviously separated. Based on the features, the MKSVM is utilized to construct a transformer fault diagnosis model, and the MKSVM’s penalty factor and kernel weights are optimized through a quantum-behaved particle swarm optimization (QPSO) algorithm. Finally, a mean signal deviation (MSD) metric is presented to locate the fault position. In this experiment, a 10 kV transformer is used to demonstrate the proposed transformer fault diagnosis procedure, and the diagnosis results reflect that the proposed method can effectively measure the transformer vibration signals, extract the essential features, construct an accurate diagnosis model, and locate the fault position.