Intelligent Electric Field Feature Selection and Insulation Strength Evaluation for Phase-to-Earth Long Air Gaps on UHV Transmission Lines

Air gap breakdown constitutes the primary failure mechanism in external insulation of power transmission and transformation systems, and its dielectric strength is the key consideration for insulation coordination. This paper presents an insulation prediction methodology that integrates quantum-behaved particle swarm optimization (QPSO)-guided intelligent electric field (EF) feature selection framework with the random forest (RF) algorithm to accurately evaluate discharge voltages of phase-to-earth long air gaps on ultra-high voltage (UHV) transmission lines. Through finite element analysis of EF distribution, the 34-dimensional EF features were mined from the interelectrode primary, auxiliary paths and equipotential rings to characterize the gap configurations. After QPSO-driven dimensionality reduction, the optimal feature subset was input into the RF classifier to model dependence mapping of EF features with insulation strength. The proposed prediction approach, developed using engineering air gaps on extra-high voltage (EHV) transmission towers as training data, achieves accurate discharge voltage evaluation for 21 samples of phase-to-earth air gaps on UHV transmission lines. The predicted results of the RF model after QPSO-based feature selection are in good agreement with the experimental values, with a maximum relative error (e_MAX) of 8.2% and a mean absolute percentage error (e_MAPE) of 3.5%, which performs better than other dimensionality reduction methods. The results of the study can provide guidance for selection of insulation distance for transmission line-tower air gaps.