基于大模型检索增强生成的水利问答架构设计OA

To address the challenges of knowledge hallucination,poor information timeliness and low localized deployment efficiency faced by Large Language Models(LLMs)in their applications to water conservancy and other knowledge-intensive industries,this paper proposes a Knowledge-Enhanced Retrieval-Augmented Generation(KE-RAG)system architecture that integrates vector retrieval with Knowledge Graph(KG)constraints.This architecture takes the domestically developed open-source model Qwen3-32B-AWQ as the generation core,constructs a multimodal water conservancy knowledge base containing laws and regulations,technical standards,engineering cases and expert experience,and achieves high-performance localized inference through the vectorized Large Language Model(vLLM)framework.Experimental results show that on the evaluation set,this architecture reaches a professional knowledge question and answer accuracy of 91.5%,an average end-to-end response latency of less than 500 ms and a throughput nearly four times higher than that of standard deployment,with all metrics significantly outperforming those of baseline models,which provides a reference solution for the application of Large Language Models in the water conservancy industry.