Discharge voltage prediction of UHV AC transmission line–tower air gaps by a machine learning model

Full-scale discharge tests of transmission line–tower air gaps are costly and time-consuming, and they cannot exhaustively simulate all the gap configurations in practical engineering. In this paper, a machine learning model established by support vector machine is introduced to predict the switching impulse discharge voltages of the ultra-high-voltage (UHV) AC transmission line–tower air gaps. The three-dimensional finite element models of a UHV cup-type tower and a UHV compact transmission line were established for electric field calculation, and some features were extracted from the hypothetic discharge channel and the shortest path between the bundled conductor and the tower. These features under a given voltage were normalised and input to the SVM model, while the output is two binary values, respectively, representing gap withstanding or breakdown. Trained by experimental data of one type of the UHV transmission line–tower gaps, the SVM model is able to predict the discharge voltages of another gap type. The mean absolute percentage errors of the two engineering gap types, under different gap distances, are 8.31 and 4.86%, respectively, which are acceptable for engineering applications. The results provide a possible way to obtain the discharge voltages of complicated engineering gaps by mathematical calculations.