Research on Intelligent Question Answering System for Electric Power Domain Based on LLMs and RAG

Large language models (LLMs) have demonstrated remarkable capabilities in conversation, reasoning, and knowledge retention. However, they still face challenges in handling knowledge-intensive tasks within the power domain, including insufficient factual accuracy, difficulties in knowledge updating, and a scarcity of high-quality domain-specific datasets. In order to address these challenges, this paper aims to design specifically a system for knowledge-intensive question answering tasks in the electric power domain. The proposed system incorporates an enhanced Retrieval-Augmented Generation (RAG) strategy that integrates a hybrid retrieval strategy and a finetuned model to provide more efficient knowledge acquisition and updating capabilities. Additionally, the system incorporates LLMbased prompt engineering techniques to enhance the coherence of responses to questions. Experimental results show that the question answering system effectively improves the accuracy of retrieving and answering power-related knowledge and reduces hallucinations. Furthermore, ablation experiments indicate that RAG module has the greatest impact on system performance, with the overall accuracy decreasing by 18.6 % without this component.