基于RAG的煤矿安全智能问答模型
Intelligent Q&A model of coal mine safety based on RAG
洪亮 1郭瑶 1刘兴丽 2李宗雨2
作者信息
- 1. 黑龙江科技大学 管理学院,哈尔滨 150022
- 2. 黑龙江科技大学 计算机与信息工程学院,哈尔滨 150022
- 折叠
摘要
面向煤矿安全智能辅助预警决策需求,以瓦斯超限煤矿安全隐患知识为数据源,基于RAG检索增强生成,建立了一种煤矿安全智能问答模型.应用BGE-M3-Embedding、FAISS开源库等方法,构建瓦斯超限煤矿安全文本块的向量数据库,以及检索和生成模块,采用BGE-M3-Embedding模型完成Query文本块向量的检索,召回top-k向量检索的上下文,实现煤矿安全瓦斯超限的相关提问,构建Prompt微调提示词,增强大语言模型生成答案.结果表明,Ragas自动评测Rag检索增强三种大语言模型效果,Baichuan2-13BAR、F忠实度及上下文相关指标最优,分别为0.91、0.83 和0.87.搭建煤矿安全辅助决策的智能问答原型,验证了触发瓦斯超限煤矿安全隐患智能辅助决策的实效性、可靠性及迁移性.
Abstract
This paper aims to explore a coal mine safety intelligent question answering model in re-sponse to the demand for intelligent early warning decision-making for coal mine safety by based on RAG retrieval and enhanced generation by using the knowledge of safety hazards in coal mines with gas excee-ding limits as the data source.The study consists of using the methods of BGE-M3 Embedding and FAISS open-source libraries to construct a vector database of safety text blocks for coal mines with gas exceeding limits,and a retrieval and generation module;using the BGE-M3 Embedding model to retrieve the query text block vector;recalling the context of the top-k vector retrieval;implementing the relevant questions about coal mine safety with gas exceeding limits;building a Prompt fine-tuning clue words to enhance the generation of answers in the large language model.The results showed that Ragas automatically evaluates the effectiveness of Rag retrieval in enhancing three major language models,and the optimal indicators are Baichuan2-13BAR,F,and CR indicators,reaching 0.91,0.83 and 0.87,respectively.Building an intelligent Q&A prototype for coal mine safety assistance decision-making has verified the effectiveness,reliability,and transferability of intelligent assistance decision-making for triggering gas exceeding coal mine safety hazards.
关键词
煤矿安全/RAG检索增强生成/智能问答/智能辅助决策支持Key words
coal mine safety/RAG retrieval-augmented generation/intelligent question answering/intelligent decision support引用本文复制引用
基金项目
黑龙江省极薄煤层智能开采关键技术攻关与示范项目(2021ZXJ02A04)
黑龙江省省属高校基本科研业务费专项(2022-KYYWF-0569)
出版年
2024
国家科技期刊平台
NSTL
万方数据