Synchronous Condenser-Based Intelligent Question Answering System Based on Large Language Model

The stable operation of power systems is crucial to modern society, where power regulators play a key role in voltage and reactive power regulation. However, the traditional forms of knowledge supply for the operation and maintenance of regulators are in urgent need of innovation. Based on artificial intelligence large model technology, an intelligent Q&A system is constructed to meet the knowledge service needs of power regulator operation and maintenance, thereby enhancing the operational efficiency and stability of the power system. By analyzing the needs of the power regulator field scene, the relevant operation and maintenance knowledge is divided into different topics, forming two major downstream tasks of the model: knowledge object recognition and knowledge Q&A. Combining automatic machine annotation and manual annotation methods, a high-quality small sample training corpus is constructed. By comparing the performance of four existing large language models: BaiChuan2-13B-Chat, ChatGLM2-6B, Llama-2-13B-Chat, and ChatGPT, the model with the best performance is selected as the base model. Following the research idea of “high-quality corpus + pre-trained large model + fine-tuning”, a deep neural network capable of semantic analysis, context association, and generation is trained to adapt to various downstream tasks, building a large model for knowledge Q&A of power regulators. Various strategies such as data optimization and retrieval enhancement generation technology are adopted to alleviate the hallucination problem of large models. The development of an intelligent Q&A system for power regulators generates high-precision, unambiguous operation and maintenance knowledge answers, while also supporting multi-turn Q&A for users.