CodeGen‐Resilience: A Framework for Automated Generation and Validation of Complex Scenarios for Large‐Scale Power System Resilience Analysis

Abstract Power‐system resilience analysis requires executable simulation scenarios under changing risk conditions that translate disturbance descriptions into precise model modifications. This paper focuses empirically on steady‐state alternating‐current (AC) power‐flow scenario generation for physical‐layer power‐system disturbances; cyber‐layer and interdependent‐infrastructure couplings are retained as representational extensions, not validated use cases. Solver‐screened disturbance scenarios form the input layer for contingency screening, feasibility assessment, robust planning and digital‐twin studies. A persistent obstacle is the ‘scenario bottleneck:’ operational descriptions are often unstructured, ambiguous and difficult to encode consistently in simulation models. To address this problem, we propose CodeGen‐Resilience, a neuro‐symbolic framework that converts natural‐language event descriptions into validated resilience scenarios. The framework introduces a hierarchical event schema (HES) for formalizing disturbance operations, a template‐constrained scenario‐generation engine that maps HES effects to executable model updates, and a solver‐feedback verification and correction loop that detects execution failures and intent mismatches. Experimental validation on a constructed 102‐case IEEE 118‐bus benchmark achieves 97.9% event interpretation fidelity, 100% build success and 94.1% functional correctness. Separate scalability experiments with 39 cases per system report build success/functional correctness of 97.4%/92.3% on IEEE 118, 100.0%/97.4% on IEEE 300, and 94.9%/84.6% on PEGASE 1354, distinguishing the main benchmark from the scalability subset. The results show that unstructured event text can be transformed into auditable, executable and solver‐checked model modifications, with convergence status explicitly identified under controlled benchmark conditions.