Applied Mathematics & Information Sciences

Author Country (or Countries)



To improve the fault diagnosis accuracy for power transformers, this paper presents a kernel based extreme learning machine (KELM) with particle swarm optimization (PSO). The parameters of KELM are optimized by using PSO, and then the optimized KELM is implemented for fault classification of power transformers. To verify its effectiveness, the proposed method was tested on nine benchmark classification data sets compared with KELM optimized by Grid algorithm. Fault diagnosis of power transformers based on KELM with PSO were compared with the other two ELMs, back-propagation neural network (BPNN) and support vector machines (SVM) on dissolved gas analysis (DGA) samples. Experimental results show that the proposed method is more stable, could achieve better generalization performance, and runs at much faster learning speed.