Multi-parametric Sensitivity Analysis of Improved Transformer Thermal Models Considering Nonlinear Effect of Oil Time Constant

This paper aims to develop an approach to investigating the effect of a particular parameter on the output accuracy of transformer thermal models, i.e. sensitivity analysis, which can not only reveal the most sensitive parameter of a thermal model but also improve model output accuracies. For the first time, the nonlinear time constant (NTC) of transformer oil is proposed to reshape three practical top-oil temperature models based on an expression of nonlinear thermal conductance: the modified IEEE clause 7 model, Swift's model, and Susa's model. Then, the multi-parametric sensitivity analysis (MPSA) is undertaken to reveal the effect of each parameter on the model output accuracy. Through onsite data validation, the results show that the accuracy performance of the proposed NTC thermal models are improved significantly by considering the nonlinear effect of oil time constant. Moreover, the derived sensitivity performances can clearly reveal the most dominant parameter of the model, so as to simplify the model parameter identification process by reducing the number of insensitive parameters. Finally, the heat-run test data is used as a reference to validate parameters optimized through a genetic algorithm (GA), which demonstrates that the proposed NTC IEEE model has not only one sensitive parameter but also superior accuracy performance.