inZORi-PFΔ Phase 1: Streaming Power Flow Robustness under Multi-Shocks and N-1 Topology Events (118-bus, Nonlinear Surrogate)

We demonstrate inZORi's adaptive capacity in a challenging streaming power flow (PF) scenario inspired by the PF-Delta benchmark motivation. This phase uses a controlled nonlinear surrogate; real AC power flow results are presented in Phase 2. The problem: maintain robust convergence of a Newton-Raphson PF solver under extreme load/generation shocks, a limited iteration budget (nr_max=8), periodic state subsampling (pf_interval=5), and sudden topology changes (N-1 line outages). Traditional fixed-strategy baselines suffer significant convergence degradation under the severe_plus regime (S3 convergence ~84-86%), versus inZORi's ~99%. We establish this through: (1) world-level PF evaluation architecture, (2) fairness-controlled comparisons (equal PF budget and wall-clock time), (3) statistical validation (CI95, 30 seeds, FULL runs), and (4) a frozen-elites runtime extension showing offline-evolved genomes achieve comparable robustness without online evolution. Key results (118-bus, severe_plus + N-1): - S3 convergence (inZORi FULL): 99.1% (CI95 [98.9, 99.3]) - S3 convergence (Baseline A): 95.98% - S3 convergence (Baseline B): 83.94% - Improvement vs Baseline B: +15.2 percentage points (CI95 non-overlap confirmed) - Frozen-Top16-K1 (runtime): S3 convergence 100.0%, avg_k_used < 1.25 (O(1) cost) - Topology recovery: 1.82 steps (inZORi) vs 12.91 steps (Baseline B) — ~7x faster Note: This is Phase 1 of a two-phase study. Phase 2 uses real AC power flow (pandapower, IEEE 118-bus) and is published separately on Zenodo with a cross-reference link.