Reinforcement Learning-Based Genetic Algorithm for Aging State Analysis of Insulating Paper at Transformer Hotspot
Zaijun Jiang, Jiefeng Liu, Xianhao Fan, Qingyin Wang, Yiyi Zhang, Thomas Wu
Abstract
The aging state evaluation of insulating paper at the transformer hotspot is a pain point in the industry. To address this issue, the modified dielectric response (MDR) model and the reinforcement learning based genetic algorithm (RLGA) are proposed to analyze the aging state of insulating paper at hotspot. Firstly, the aging state related polarization and depolarization currents (PDC) of transformer insulation are collected. Then the MDR model is reported to describe the PDC property of insulating paper. The RLGA is later proposed to search the optimal model parameters defined in MDR to characterize the aging state of insulating paper at hotspot. Verification results present the feasibility and validity of extracted optimal model parameters for aging analysis. Furthermore, the quantitative correlation between these parameters and the aging state of hotspot is analyzed. Regarding this, the present work provides a potential method for obtaining aging information of insulating paper at hotspot.