Reliability assessment of transformer thermal models.

摘要

This work presents methods and metrics to determine the reliability of transformer thermal models from two points of view: model adequacy and model accuracy. The model adequacy is measured by traditional metrics such us normal probability plot of the residuals, residual plots, R 2, Variance Inflation Factors (VIF) and Prediction Sum of Squares (PRESS). The model accuracy is measured by traditional metrics such as condition numbers, sensitivity numbers, parameter estimates, error duration curves and SSL prediction. These metrics will allow a quantitative comparison among different models. This includes the linear (linear top-oil model) and nonlinear (Swift et al., Susa et al. and the top-oil ANSI/IEEE) models. Numerical experiments using the proposed methods and metrics were performed using field data provided by Arizona Public Service and Salt River Project. The measured data used in this work corresponds to a 30-day data set DV6 and DV11 (APS transformers whose nominal rating is 167 MVA) and Corbell 3 (SRP transformer whose rating is 28 MVA) transformers. Comparison of the obtained metrics for the four models showed that the linear top-oil model performed better in comparison to the nonlinear models. However, further experimentation with larger data sets is required for the nonlinear models.