Electrical Engineering Department

PhD Final Oral Defense

EVALUATION OF MODELS FOR PREDICTING HOTTEST-SPOT TEMPERATURE

IN SUBSTATION DISTRIBUTION TRANSFORMERS

by

Oluwaseun Adeyemi Amoda

September 14, 2009

12.00pm

ERC 490

Committee:

Dr. Daniel Tylavsky (chair)

Dr. Raja Ayyanar

Dr. Gerald Heydt

Dr. Douglas Montgomery

Dr. Vijay Vittal

Abstract

To achieve the objectives of dynamic loading of a transformer, it is essential to be able to accurately predict hottest spot temperature (HST) in the transformer winding. The prediction of the HST in transformers can be achieved by using a thermal model developed for such purpose. This document presents an investigation into the acceptability of thermal models developed for predicting HST in transformers when the parameters of the models are to be determined from measured field data. The investigation uses numerical and statistical methods to examine the thermal models by fitting them to measured field data obtained from utility transformers.

Metrics to evaluate the models accuracy, adequacy and consistency in predicting HST were developed. These metrics were used to measure the reliability and acceptability of the models in a quantitative and graphical manner. The metrics were also used to compare the performances of the different models. The results of the metrics evaluations served as a measure of confidence in the models for practical applications purposes.

The following five models were investigated, the HST-rise model, modified HST model, nonlinear HST model, Swift’s model and Susa’s model. The HST-rise model was shown to be deficient and the modified HST model was proposed to correct its deficiency. The results of the parameters estimation and performance metrics were acceptable for the modified HST model. The nonlinear HST model was found not to be accurate and consistent. The values of one of the estimated parameters of Swift’s model showed that Swift’s model is not accurate, even though it had acceptable consistency and adequacy metrics. Susa’s model is a complex model that takes into account two dependency factors that the other models do not account for. Susa’s model, like the modified HST model, was shown to be an acceptable model.

The modified HST model was chosen over Susa’s model for further investigation. This choice was made because the modified HST model gave metric performance similar to that of the more complex Susa’s model. Two end-uses of the modified HST model were demonstrated in a prediction simulation and a Kalman filter application.