Fault diagnosis for power transformer using deep learning and softmax regression

In order to solve the problems such as availability of data extraction, better local optimum, gradient to dissipate more efficiently, this paper presents a new method of power transformer fault diagnosis based on deep learning and Softmax classifier. Power transformer fault diagnosis model is established based on stacked auto-encoders and softmax regression, then each restricted boltzmann machine of fault diagnosis model is optimized by the unsupervised raining with massive unlabeled samples based on k-step contrastive divergence and adjusts parameters of fault diagnosis model by the supervised algorithm. Finally, the power transformer fault types are determined by Softmax regression. Test results show that the proposed method has higher diagnosis accuracy and better adaptability than those based on the back propagation neural network and the support vector machine.