This paper explores the method of domain specialization for Large Language Models in the field of railway track management, as well as their potential and practical effectiveness in knowledge-based question answering scenarios. Given the complexity of the railway track management work scenarios and the challenges associated with existing data acquisition and processing, this paper has proposed a domain-specialized large language model tailored for railway track management. Additionally, by integrating layout analysis, OCR image recognition technologies, and the self-instruct method, the efficiency and accuracy of data processing have been enhanced, and the training dataset has been expanded. Moreover, the paper employs knowledge retrieval augmentation techniques to mitigate the issue of hallucinations produced by the model, thereby improving the model's accuracy, relevance, and completeness in knowledge question answering. Furthermore, the paper has established a dataset and evaluation metrics specifically for question answering in the railway track management field and assessed them using a general Large Language Model. This research not only aids relevant personnel in their daily operations and enhances work efficiency but also provides valuable references for future applications of Large Language Models in similar domains.