A Self-evolving Railway Standards and Specifications Retrieval Method Based on Multi-agent Collaboration

Railway survey and design standards and specifications serve as the technical guidelines and fundamental basis for railway engineering construction, playing an irreplaceable role in ensuring engineering quality, safety, efficiency, and sustainable development. Since survey and design standards and specifications are predominantly stored as unstructured documents, and complex cross-references and associations exist among multiple standards, significant difficulties arise in retrieving and utilizing clauses. To address these challenges, we propose a self-evolving multi-agent collaborative method for retrieving clauses from railway standards and specifications. First, a knowledge graph ontology structure oriented toward standards and specifications clause retrieval is constructed. Second, an automated knowledge base construction pipeline is developed that leverages prompt engineering to extract entities, relationships, and semantic information from unstructured documents. Third, a multi-agent retrieval system composed of five specialized agents is designed to achieve the goal through the division of labor and cooperation among multiple agents. Furthermore, the method incorporates a human-in-the-loop self-evolution mechanism that analyzes successful retrieval cases to identify and validate missing relationships for knowledge graph completion. Experimental results on real railway survey and design standards and specifications retrieval data demonstrate that the proposed method achieves significant improvements in retrieval accuracy, comprehensiveness, and efficiency.