Power Grid Model Generation Based on the Tool-Augmented Large Language Model基于工具增强大语言模型的电网模型生成方法
英文 Abstract
Power grid models are essential for power system research and AI-based applications. However, due to security reasons, publicly available power grid models are very limited, making the generation of models with specific topologies and power flow properties crucial. As these models are typically stored in organized text, their generation can be framed as structured text generation, which can be effectively handled by Large Language Models (LLMs). Inspired by their remarkable performance, this letter proposes a novel method for generating power grid models based on the tool-augmented LLM, providing users with an intuitive way to generate power grid models according to their needs. Numerical results show that the proposed method can effectively generate power grid models that meet specified requirements expressed in natural language.
中文翻译
电网模型是电力系统研究和人工智能应用的重要基础，但受安全因素影响，公开的电网模型十分有限，因此需要生成具有指定拓扑和潮流特性的模型。由于电网模型通常以结构化文本保存，其生成可转化为大语言模型擅长的结构化文本生成任务。本文提出一种基于工具增强大语言模型的电网模型生成方法，使用户能够通过自然语言描述需求。数值结果表明，该方法可以有效生成满足指定要求的电网模型。
中文概述
该研究将电网拓扑生成、潮流数据填充和潮流约束调整封装为大语言模型可调用的工具链，并通过反馈回路纠正不可行结果。实验中生成模型的潮流收敛与物理约束满足率得到显著提升，整体成功率达到 95%，展示了 LLM Agent 在电网仿真数据与测试算例构建中的应用潜力。