Prediction of Positive Lightning Impulse Breakdown Voltage Under Sphere-to-Barrier-to-Plane Air Gaps Using Machine Learning

Abstract

Barrier, solid insulator, is inserted between conductors to make compact power equipment. Prediction of the dielectric strength is significant owing to nonlinear effect of barrier. In this paper, positive lightning impulse breakdown voltages are predicted under sphere-to-barrier-to-plane air gaps using machine learning algorithms including a support vector regression (SVR), Bayesian regression (BR), and a multilayer perceptron (MLP), which are rarely used to derive breakdown voltages. Previous studies have generally considered background electric fields in field arrangements that lacked barriers. In contrast, electrostatic features are suggested based on the electro-geometric equivalency of each electrode, electric field distributions between sphere and barrier or between barrier and plane, and a condition for stable penetration of discharge channels, influencing background fields and discharge propagation characteristics in air gaps. SVR yielded more precise Breakdown voltages than BR or MLP. Predictions from algorithms were in good agreement with experimental results, regardless of geometrical parameters such as spherical radius, gap distance and barrier width. In particular, the SVR-predicted voltages were even more accurate than the calculated voltages from streamer propagation method in strongly inhomogeneous field with barrier. Our proposed method derives breakdown voltages without the need to consider geometrical parameters affecting streamer propagation.

Keywords

• Bayesian regression (BR)
• barrier
• lightning impulse breakdown voltage
• multilayer perceptron (MLP)
• support vector regression (SVR)
• sphere-to-barrier-to-plane