Feature extraction and stage classification of cavity discharge development in transformer oil‐paper insulation

SUMMARY Characteristic study of cavity discharge is important in detecting and diagnosing internal failure of power transformers. Based on the kernel principal component analysis method, eight new feature parameters are extracted from 29 statistical parameters, which are commonly used in the characteristic study of partial discharges. Redundancy and computational complexity for applications are greatly reduced. Using cluster analysis method, new parameters are proved to be effective in the application of stage classification in the cavity discharge development. Four stages are classified: incipient discharge, unstable discharge, consistent discharge, and disruptive discharge. Copyright © 2013 John Wiley & Sons, Ltd.