"Training Dataset for Small-Signal Modelling of Power Systems"

"This dataset presents a collection of structured training data designed for fine-tuning Tool-Augmented Large Language Models (TA-LLMs) in the power and energy domain. Specifically, it focuses on the small-signal modeling and simulation of AC-DC hybrid power electronic-defined power systems. The dataset for LLMs' training addresses challenges of the inefficiency of manual Grid Code Compliance assessments for Inverter-Based Resources (IBRs) and the scarcity of realistic network parameters due to confidentiality concerns. Organized in a hierarchical structure, the data covers fundamental mathematical control blocks, complex apparatuses such as synchronous generators and inverter-based resources, and multi-node AC-DC network topologies. The synthesis of this dataset utilizes a closed-loop framework that integrates physics-informed templates generated from online LLMs with an automated generation pipeline. All data entries are formatted in the standardized Alpaca-style JSONL structure, ensuring high machine-readability and compatibility with modern instruction-tuning workflows. To ensure technical rigor, the dataset's accuracy is evaluated through time-domain and power-flow simulations compared with public models. This article provides detailed documentation on data sources, formatting conventions, and usage instructions to facilitate the development of AI-driven digital twins and automated simulation tools in power systems."