• Multi-scale framework bridges load forecasting with topological resilience diagnosis. • Retrospective augmentation and spectral regularization enhance prediction accuracy. • Dimensionality reduction reveals universal collapse trajectories in power systems. • Framework distinguishes high-stress operation from structural failure states. • Integration enables real-time early-warning for distribution grid resilience. Power distribution systems face growing uncertainty from fluctuating demand and renewable integration, but existing methods suffer from scale mismatch: macro-scale forecasting cannot distinguish transient stress from structural degradation. We propose a two-phase framework bridging load prediction with resilience diagnosis. Phase 1 employs a Retrospective-Augmented Learner (RAL) with a Dynamic Relational Graph (DRG) and Spectral Fidelity Regularizer (SFR), achieving state-of-the-art prediction accuracy without redundancy parameters. Phase 2 applies Resilience Dimension Reduction (RDR) to expand macro-terminals into micro-network topologies (IEEE 14/118-bus), computing resilience parameter β to map system states onto universal curves and differentiate stable, pre-alarm, and collapse zones. Experiments demonstrate that this multi-scale coupling identifies when systems deviate and why, establishing a path toward physics-informed resilience reasoning for explainable energy infrastructures.
Chen et al. (Sat,) studied this question.