AI-Powered Fault Prediction in 1100-kV Power Transmission Systems: From Algorithm-Driven Protection to Intelligent, Predictive Systems

Abstract Ultra-high-voltage (UHV) 1100-kV ac corridors carry 5-GW-class transfers over hundreds of kilometres, so a slow, missed or unnecessary trip has system-level consequences. Numerical protection is algorithm-driven: it reacts only after a fault and loses reach for high-resistance faults on long, shunt-compensated lines. This paper develops and evaluates an AI-powered framework that moves 1100-kV line protection toward intelligent, predictive operation. A 360-km, 8-bundle corridor is modelled from first principles (Carson line constants, sequence parameters, surge-impedance loading, shunt and neutral reactor sizing, IEEE 738 ampacity, symmetrical-component fault analysis) and by a 4.8-kHz electromagnetic-transient model. A two-layer architecture is proposed: a predictive layer estimates the day-ahead fault risk of each 20-km section from weather and condition data, and a protective layer classifies, locates and zones faults from a half-cycle window using physics-informed features, under supervisory logic that keeps certified distance protection in charge. On 5400 simulated cases the AI layer identified internal fault types with 99.2% accuracy (conventional selector 82.7%), raised Zone-1 dependability from 86.0% to 96.4% (80% vs 47% for ground faults through 100–400 Ω) with 1 external misoperation in 986, and cut the 90th-percentile decision time from 20.2 to 11.8 ms. A Takagi-residual locator reduced the mean error from 18.3 to 7.1 km. On two years of synthetic condition data the risk model flagged 62% of faults up to 24 h ahead versus 42% for threshold rules at equal alarm rate, and risk-based insulator washing cut expected pollution flashovers by 30%.