Regresar

Mechanical Stress in Power Transformer Winding Conductors: A Support Vector Regression Approach

Abstract:

A support vector regression model to determine the mechanical stress in winding conductors of a power transformer was developed in this research. Only the hyperparameter C, related to the regularization of the model, was varied during the design, thus simplifying the process and avoiding the need of using hyperparameter grids for performance comparison. The model was applied to a 400 MVA three-phase power transformer. The training and validation data were obtained from finite element simulations. The Python’s library scikit-learn was used for the design process. The input data of the model were the currents circulating through the windings and the output data were the mechanical stresses of the conductors. Transient fault currents were used for the sample cases, with fault impedances that varied from 15 Ω to 80 Ω. Cases outside of the training data were used to test the model, giving accuracies from 1% to 20%. The greatest errors appeared for high fault impedances, i.e. for low fault current values. In a global analysis, it was seen that 99% of the errors were lower than 2.5%.