Optimum Selection Strategy of Time Domain Dielectric Response Characteristic Quantity of Transformer Oil-Paper Insulation

In the current research of oil-paper insulation diagnosis, there is no reasonable basis for selecting the time-domain dielectric characteristic space. A two-stage time-domain dielectric characteristic optimization strategy is proposed, which combines Pearson correlation coefficient and random forest importance estimation. Firstly, 23 kinds of time-domain feature quantities which have been studied are summarized, and the measured data of transformer are collected to complete feature extraction. Secondly, a feature selection strategy is proposed to determine the optimal feature space: the first level feature selection takes Pearson correlation coefficient as the index to eliminate highly redundant features; the second level feature selection uses random forest bag data for feature evaluation, eliminating less important features. Finally, a variety of algorithms are used for diagnostic experiments and comparative analysis. The results show that the optimal feature space proposed in this paper has stronger applicability in all diagnostic algorithms. The method proposed in this paper is feasible and effective for the optimization of dielectric characteristics in time domain.