A High-speed Train Operational Risk Prevention and Control Model Based on a Large Language Model

Yuan ZHAO, Shifeng LIU

Abstract

The high-speed train operational safety is a critical aspect of railway safety control, involving a multi-dimensional information analysis and complex decision-making processes. The traditional operational risk control systems face numerous challenges in areas such as the holistic risk perception, the real-time analysis of unstructured data, the evaluation of operator actions, and the generation of risk mitigation plans. Large Language Models (LLMs), with their exceptional general intelligence and robust reasoning capabilities, have sparked transformative changes across various fields. This study presents an operational risk prevention and control model for trains based on a LLM, called the train security large model (TSLM). The TSLM integrates techniques such as retrieval augmented generation (RAG), chain-of-thought (CoT), and model fine-tuning for performing a real-time analysis and risk identification in three train operation scenarios. The experimental results indicate that the TSLM performs well with regard to the risk identification accuracy, the interpretability of the obtained results, and the model`s generalization ability. The proposed model can effectively address various risk scenarios and provide reasonable risk mitigation suggestions. Even in unfamiliar scenarios, the TSLM demonstrates a high risk identification accuracy and rationality, reflecting its strong generalization capability and reliability. Thus, this model provides an innovative and practical approach to train dispatching safety management, with the potential for a continuous optimization and improvement in order to further enhance the safety and efficiency of railway transportation in real-world applications.

Keywords

High-speed train, Risk prevention and control, LLM.

View full article