Detection of early failures within traction transformers based on Gaussian-PSO

A novel self-adaptive RBF Neural Network algorithm is proposed in the paper to detect the early failures of electric locomotive traction transformers. In the algorithm, the initial node number and center vector of RBF neural network are obtained by fuzzy C - average (FCM) algorithm firstly, then together with connection weights are optimized by the Gaussian improved particle swarm optimization (PSO) algorithm. The self-adaptive RBF neural network is finally applied to the comprehensive test and fault diagnosis system for electric locomotive traction transformer. The results show that the proposed algorithm can effectively detect the faults that is misinformed and underreported by the original test system.