An Explicit Memory-Driven Agentic Framework for Power System Simulation

Qinjuan Wang, Yongli Zhu

Keywords: LLM Agents, Explicit Memory, Power System Analysis, Trustworthy AI, Unlearning

Abstract: In rigorous domains like power system simulation, Foundation Models (FMs) relying on parametric memory or fine-tuning suffer from confident hallucinations, catastrophic forgetting, and privacy risks. To address this, we propose a decoupled, explicit memory-driven agentic framework that restricts the FM to a pure reasoning engine. By offloading knowledge, state, and skill management to external modules, our architecture facilitates more predictable execution and zero-shot skill reuse without weight updates. Experimental results show our approach achieves 87.35\% code fidelity accuracy, demonstrating improved performance over parametric baselines and maintaining stability in complex tasks.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 77