Fault Diagnosis of Transformer Based on Quantum-Behaved Particle Swarm Optimization-Based Least Squares Support Vector Machines

In order to overcome the deficiencies of artificial neural networks (ANN), such as low convergence rate, local optimal solution, over-fitting and difficult determination of structure, a proposed QPSO-LS-SVMs method is applied to fault diagnosis of power transformer. It takes five characteristic gases dissolved in transformer oil as its inputs and seven transformer states as its outputs, constructs a fault diagnosis model for power transformer based on least squares support vector machines (LSSVMs) and uses QPSO to determine parameters of LS-SVMs. The experimental results indicate that the recognition rate of QPSO-LS-SVMs is 18.8,14.3 and 6.0 percents higher than that of IEC three-ratio method, BPNN and PSO-LS-SVMs, respectively, and that the training speed of QPSO-LS-SVMs is 4.02 times faster than that of PSO-LS-SVMs. So, the correctness and effectiveness of our proposed method are proved, and QPSO-LSSVMs is a proper method for fault diagnosis of power transformer.