Transformer Fault Diagnosis Based on Neural Network of BPARM Algorithm

Abstract

Slower convergence and longer training time are the disadvantages usually mentioned when the conventional Back-Propagation (BP) algorithm are utilized in transformer fault diagnosis based on Artificial Neural Network (ANN). Consequently, an efficient acceleration technique- BPARM (Back-Propagation with Adaptive learning Rate and Momentum term) algorithm was proposed to reduce the training time, where the learning rate and the momentum term are altered at iteration. We implemented a system of transformer fault diagnosis based on Dissolved Gases Analysis (DGA) with BPARM. Training patterns were extracted from Refined Three-Ratio method. Test results show that the system has the better ability of quick learning and global convergence than other methods, and improves accuracy of fault recognition.