Research on Transformer Fault Diagnosis Method Based on NRS-PSO-ANFIS

Aiming at the problem of uncertainty in the relationship between fault data and fault types in the traditional transformer fault diagnosis process and low diagnostic accuracy, this paper proposes a transformer fault diagnosis method based on neighborhood rough set (NRS) and adaptive neuro-fuzzy inference system (ANFIS). Firstly, the simplification model of neighborhood rough set is constructed, and the 18 gas ratios obtained by dissolved gas analysis method are taken as the initial feature quantity, and the optimal feature set is obtained by using NRS simplification. On this basis, the ANFIS model is established, and the particle swarm optimization algorithm (PSO) is used instead of the classical ANFIS hybrid learning algorithm based on the BP algorithm and the least squares method to train the model parameters, in order to overcome its shortcomings of easily falling into local optimization and to improve ANFIS performance. And real dataset experiments are conducted to compare the fault diagnosis accuracy of four different feature quantities under four different diagnostic methods, which shows that the method of this paper has high accuracy and reliability in transformer fault diagnosis.