This paper introduces an advanced EEG-GRA cross-sequence feature extraction method for operator cognitive fatigue detection in industrial settings. Our research addresses key limitations in conventional approaches through three technical innovations: (1) an intelligent adaptive time-varying weight function system that continuously calibrates to operator cognitive states, (2) an advanced multi-scale analysis framework incorporating state-of-the-art wavelet decomposition, and (3) a sophisticated cross-sequence feature fusion mechanism that leverages spatial correlations across EEG channels. Comprehensive performance evaluation reveals significant quantifiable improvements: the system achieves a 45% reduction in processing time (from 100ms to 55ms), enabling genuine real-time monitoring capabilities; detection accuracy shows a remarkable increase of 17.5 percentage points (from 76% to 93.5%); and signal quality demonstrates a substantial improvement of 5.3dB (from 15dB to 20.3dB). These advances are achieved while simultaneously reducing computational demands, with algorithmic optimization decreasing complexity from O(n²) to O(n log n) and memory requirements reduced by 38%. Field implementation in a nuclear power plant control room involving 30 operators under rigorous operational conditions validated the system's exceptional reliability, maintaining 99.99% uptime during 12-hour continuous monitoring shifts. Statistical analysis confirms the significance of these improvements (p < 0.01), establishing a new benchmark for industrial safety systems across high-risk sectors.