Partial Discharge Pattern Recognition of Transformer Based on Deep Forest Algorithm

Abstract

In order to improve the accuracy of pattern recognition of transformer partial discharge type, solving the current pattern recognition based on machine learning algorithm requires artificial extraction of description features, poor adaptability and low recognition accuracy. In this paper, the deep forest algorithm is introduced into the partial discharge pattern recognition of transformers. In this method, the partial discharge images collected by partial discharge inspection instrument are processed by gray-scale and bilinear interpolation as input of deep forest model. The feature extraction of PD images is realized by using multi-grained scanning structure, and the classification of PD is realized by using cascade forest structure as classifier. By setting up the partial discharge experiment platform of transformer, the algorithm is tested with the sample data obtained from the experiment, and the results show that the pattern recognition accuracy of this method is high, and the recognition accuracy increases with the increase of sample data.

Dealing With Data Uncertainty for Transformer Insulation System Health Index

“…The study in [18] compared several machine learning methods to predict furan levels in transformer oil, where the RF model produced a good accuracy. Research in [38], [39] developed a partial discharge recognition model using a variety of machine learning methods, of which RF achieved the highest accuracy.…”