ChatGrid: Intelligent Knowledge Q&A for Power Dispatching Control Based on Large Language Models and Retrieval-augmented Generation

As power systems become increasingly complex and intelligent, traditional power dispatching control systems struggle to effectively meet the demands of new digital and intelligent power grids. The issues of low digitalization, automation, and intelligence levels are becoming more pronounced. In response to these challenges, we propose an intelligent knowledge Q&A method for power dispatching control systems, named ChatGrid. This method leverages the robust language understanding and generation capabilities of Large Language Models (LLMs) and the comprehensive knowledge of the power dispatching control domain knowledge base, facilitated by Retrieval-augmented Generation (RAG). The process begins by vectorizing power dispatching control domain text to create a knowledge base, followed by preprocessing user questions and using similarity metrics for relevant vector retrieval. Subsequently, LLMs are utilized for command execution and answer generation, integrating domain knowledge and prompts. Experimental results indicate that our proposed method effectively enhances the intelligence level of knowledge Q&A in the power dispatching control domain and improves the context understanding, professionalism, interpretability, and accuracy of the answers provided by the LLMs. The proposed method has wide application prospects in the power dispatching control domain, offering digital, intelligent, and automated solutions for Q&A and command parts. Furthermore, this method can be extended to actual dispatching processes, providing dispatchers with key information retrieval and decision-making support.