الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 135 |  |  |
| | | from | 135 | | |
Abstract
Transformer is one of the most important components in any electrical power system. It is a common practice to use transformer oils as insulating liquids. Thus, it is very important to control the deterioration of transformer oil by periodically monitoring its characteristics. The prediction of transformer oil characteristics as a function of service period is necessary to plan a purification schedule for in-service power transformers. Hence, increase the reliability of system operation. This thesis introduces a support vector machine (SVM), an artificial neural network (ANN) and a non-linear (NLM) models for prediction of transformer oil residual operating time (trot). Transformer oil residual operating time is defined as the service period after which the transformer oil breakdown voltage violates the standard specification limit. The non-linear model depends on linear combination of non-linear models for the most influential characteristics on trot. Among many experimental results of transformer oil characteristics the most influential characteristics on trot have been determined by statistical analysis. These characteristics have been implemented for SVM, ANN and NLM models that preserve the non-linear relationship between their combinations for prediction of trot. The results of trot modeling techniques were compared with those obtained by the previously published models namely multiple linear regression, polynomial regression, non-linear regression, ANN and SVM models are compared using different evaluation indices. The NLM, ANN and SVM models have proved their accuracy and applicability while SVM outperforms other models when dealing with small numbers of training data. A NLM model and a SVM model were formulated for the prediction of transformer oil breakdown voltage. The results of these two models were compared with those obtained by the previously published models namely; multiple linear regression model, polynomial regression model and ANN model. This study is helpful for monitoring transformer oil characteristics and organizing maintenance schedules for in-site transformers