Artificial neural network and non-linear models for prediction of transformer oil residual operating time

This paper presents two modeling techniques for the prediction and monitoring of the characteristics of transformer oil. The first employs artificial neural network (ANN) and the second employs non-linear modeling (nlm). The proposed techniques are implemented for predicting the transformer oil residual operating time (trot) which is defined as the service period after which the breakdown voltage (BDV) violates the limits given in the standard specifications.The selection of the most influential characteristics on residual operating time (trot) in the proposed techniques is obtained by statistical analysis. The non-linear model depends on linear combination of non-linear functions for each characteristic. The ANN technique for modeling these characteristics preserves the non-linear relationship between these characteristics and (trot). The results are compared with previously published modeling techniques namely multiple linear regression and polynomial regression models. Different evaluation indices have been used to justify the superiority of the proposed modeling techniques for predicting (trot).