Prediction of the lifetime of a power transformer using an optimized SVR

Predicting the lifetime of the transformer avoids a sudden cessation of its operation along with all technical, economic, and social consequences. The degree of polymerization of the cellulose which constitutes the transformer paper is a good indicator of aging. Since direct measurement of DP is not easy, DP can be determined from other transformer's quality parameters. One of the parameters used to determine indirectly the DP is the concentration of 2-furfuraldehyde. 2-furfuraldehyde is a product of cellulose degradation. In this article, we propose a support vector regression optimized by a Gold Sine Algorithm (Gold-SA) to predict the transformer's loss of life based on the 2-FAL concentration of its oil. The parameters of the model are calculated by the optimization algorithm operated by the model itself. As a result, the model automatically adjusts based on the data used for the designing of the model. The model was tested on nine transformers to assess performance. For DP prediction, the average error is 0.83%, the maximum error is 4.53% and the minimum error is 0.01%. For the transformer loss of life prediction, the average error is 0.90%, the maximum error is 4.46% and the minimum error is 0.01%.