Aging State Evaluation for Insulation Paper of Traction Transformer Hotspot Region Based on FDS and Intelligent Algorithm

Oil-paper insulation materials are widely used in various types of traction transformers. However, the existing methods cannot effectively evaluate the aging state of hotspot insulation paper in traction transformers. To address this issue, this article prepared the uniformly and the nonuniformly aging oil-impregnated insulation paper samples, and their frequency-domain spectroscopy (FDS) and degree of polymerization (DP) are collected. The improved fractional dielectric model and the multiobjective cooperative coevolutionary algorithm (MO-CCEA) are proposed for fitting FDS and extracting feature parameters related to insulation paper's aging state. Compared with the traditional Cole-Cole model, the improved model reduces relative error by approximately 7%. The least squares fitting method is used to obtain quantitative equations between feature parameters and insulation paper's DP, achieving a fitting coefficient of over 0.97. Additionally, a feature parameters database is constructed. Based on the database, the training of the bootstrap aggregation classification and regression trees (BA-CARTs) classifier is completed, which enables the evaluation of the aging state of hotspot insulation paper. The evaluation results have been verified by laboratory and field transformers, proving the accuracy of the hotspot insulation paper aging state evaluation method.