Inversion Detection of Transformer Transient Hot Spot Temperature

This paper proposes an inversion method to estimate a 10 kV oil-immersed transformer transient hot spot temperature (HST). A set of transient feature quantities which can reflect the load change are proposed, those quantities as well as the real time load rate and feature temperature points on the transformer iron shell are taken as the input parameters of a machine learning model established by support vector regression (SVR), thus to describe their relationships with the transformer transient HST. K-fold cross-validation training method and grid search (GS) parameters optimization method are used to find the optimal parameters of the SVR model, the HST inversion results agree well with the transformer temperature rise test data which are conducted with short circuit method, and the HST inversion results outperform the results obtained with GA-BPNN method. The mean absolute percentage error (MAPE) is 1.66 %, and the maximum temperature difference is 2.93 degrees C.

Automated summary

This paper proposes an inversion method using a machine learning model to estimate the transient hot spot temperature of a 10 kV oil-immersed transformer, demonstrating that the inversion results are more accurate compared to other methods.