An LLM-Assisted Framework for Synthetic Power Network Graph Generation

Due to privacy and security concerns surrounding actual power system structure data, many studies have focused on generating simulated power grids and providing corresponding datasets. However, existing methods and datasets often struggle with effective visualization and lack diverse power grid scenarios. To address these challenges, we propose a large language model-assisted workflow for small-scale synthetic power grid generation, verification, and visualization. Moreover, by taking into account the variety of node types and connection relationships inherent in power networks, our approach aims to enhance understanding of structural characteristics through detailed feature representation. We leverage the large language model to enrich the generated power grid structure data, uncovering potential internal structures. Furthermore, we employ advanced visualization techniques to generate power structure images, facilitating analysis by human experts and supporting potential learning tasks. Our results demonstrate that the generated power grid structures align well with the properties of real-world networks, and the visual outputs effectively highlight the intricacies of these structures. The code and examples are available at https://github.com/SYLLABLE604/PG-generator.