Partial Discharge Pattern Recognition of Transformer Based on OS-ELM

Abstract

The traditional partial discharge (PD) pattern recognition algorithms have low recognition accuracies and slow recognition speed in practical engineering applications because of their limitations, including a large number of parameters to tune, the difficulty in optimizing parameters and low learning rates. Therefore, we proposed a PD pattern recognition method for transformer based on the Online Sequential - Extreme Learning Machine (OS-ELM) algorithm. OS-ELM is an online-learning and improved algorithm of Extreme Learning Machine (ELM), and a new type of Single-hidden Layer Feed-forward neural network (SLFN). Meanwhile, a lot of experimental data were obtained from real transformer in the high voltage laboratory on the PD experiments based on Ultra High Frequency (UHF) detection method. In addition, OS-ELM is analyzed and compared with ELM, Support Vector Machine (SVM) and BP neural network (BPNN) in both pattern recognition effect and performance aspects. The results show that the accuracy of OS-ELM is 5.2% and 23.2% higher than that of SVM and BPNN, respectively. When reducing the size of the training samples, the fluctuation of OS-ELM recognition results is significantly smaller than that of SVM and BPNN, showing better generalization ability. Besides, the training time of OS-ELM is only 0.031 2 s, far less than that of SVM and BPNN. Therefore, OS-ELM is more suitable for engineering applications of large data volume samples.