Abstract
—In order to remove interference signal and external discharge signal during partial discharge test and get the effective signal of partial discharge, this paper carries out the research of typical interference signal mining and identification method for transformer partial discharge test. Firstly, based on the experience of previous researchers in partial discharge research, 47 feature parameters including statistical, texture and shape features are extracted, and the original feature space is constructed based on them, which provides the data basis for the optimization of the feature space and signal classification identification later. Next, the random forest algorithm is used to measure and rank the importance of the feature parameters, and the original feature space is optimized according to the ranking results to obtain the optimized feature space. Finally, the optimized feature space is used as the input of linear discriminant analysis algorithm and K-nearest neighbor algorithm to classify and identify the local discharge signal and typical interference signal, and the recognition accuracy of these two classification and identification algorithms is compared and analyzed, and the K-nearest neighbor algorithm is selected as the final classification and identification algorithm.