Synergistic Method for Machine-Learning-Based Diagnosis and Gradient-Based Optimization of Lightning Damage Mechanisms

This study presents a data-driven framework that integrates machine-learning-based diagnosis with gradient-based optimization for lightning protection of transmission corridors. Six standardized features — ground resistance, insulator-string length, tower height, protection angle, ground inclination, and ground-flash density—are normalized to construct tower state vectors for model training and inference. The diagnostic model outputs risk scores and cause attributions, identifying ground-flash density as the dominant hazard (27.9%), followed by excessive ground inclination, increased tower height, and shortened insulator strings that elevate wrap-around strike risk. Using these diagnostic results, a gradient-based optimizer minimizes state deviations within engineering constraints to generate actionable protection measures, including arrester configuration and parameter adjustments. In a 220-kV case study covering 39 towers, 30 satisfied the standard and 9 were recommended for arrester installation; for the high-risk Tower #034, the trip-out rate decreased to 1.632 times/(100 km·a). The proposed approach improves objectivity and efficiency relative to manual practice. Future work will incorporate cost weighting to enhance practical deployment.