Prediction of Dissolved Alcohol Concentrations in Transformer Oil and Aging Evaluation of Cellulose Insulation

Abstract

Alcohol (methanol and ethanol) markers in transformer oil have progressively played a crucial role in assessing the aging state of cellulose paper. There is a lack of reported studies on the prediction of alcohol concentration in transformer oil. Given this, this work presents an approach that employs an atomic orbit search optimized support vector machine (AOS-SVM) to predict the alcohol concentration. Initially, oil-paper insulation samples are prepared, and the alcohol concentrations are determined. Furthermore, methanol and ethanol concentrations in oil were predicted using AOS-SVM, and the MSE of the prediction results reached 0.0428 and 0.0064, respectively. Eventually, a quantitative model combining methanol, ethanol, and degree of polymerization (DP) was constructed to evaluate the aging of insulating paper, which verified the reliability of the proposed model. In this regard, the proposed method exhibits reliable predictive performance in terms of alcohol concentration in oil, thereby contributing to the effective evaluation of the aging state of insulating paper.