# Emergency Operation Scheme Generation for Urban Rail Transit Train Door Systems Using Retrieval-Augmented Large Language Models

## TL;DR

This paper introduces a new method using AI to create better emergency plans for train door failures in urban rail systems, improving safety and adaptability.

## Contribution

A retrieval-augmented LLM framework for generating executable and verifiable emergency operation schemes for URT train doors.

## Key findings

- Hybrid retrieval with reranking achieves high retrieval quality (Recall@5 = 0.78; Coverage@B = 0.71).
- Fine-tuned generation improves usability metrics (SchemaPass = 0.88, RoleAcc = 0.91, CiteCov = 0.73).
- The full method outperforms pure LLM and RAG-only baselines in generating usable emergency operation schemes.

## Abstract

Urban rail transit (URT) train-door failures are safety-critical and can cause cascading service disruptions, yet existing emergency operation schemes (EOSs) are often static, difficult to adapt to evolving fault patterns, and hard to verify against updated regulations. This study proposes a retrieval-augmented large language model (LLM) framework for executable and evidence-traceable EOS generation. Multi-source heterogeneous incident evidence (structured work orders, operational impact records, and unstructured maintenance/dispatch narratives) is normalized into a structured incident representation, and a hybrid retriever (dense + BM25) with cross-encoder reranking selects compact regulatory clauses and historical cases under a fixed context budget. The generator is fine-tuned with structured objectives to enforce schema compliance, role assignment, and citation grounding. Experiments on 776 passenger-door incidents from Shanghai URT (2019–2024) show that Hybrid + rerank achieves the best retrieval quality (Recall@5 = 0.78; Coverage@B = 0.71; FirstHit/B = 0.46). For generation, the full setting improves operational usability, reaching SchemaPass = 0.88, RoleAcc = 0.91, CiteCov = 0.73, and UsableAns = 0.83, compared with 0.15 UsableAns for a pure LLM baseline and 0.26 for prompting with RAG only. These results indicate that combining high-utility retrieval with structure- and citation-aware fine-tuning substantially improves the executability and verifiability of safety-critical operation schemes.