高铁列控车载设备故障知识图谱构建方法研究

Research on construction method of fault knowledge graph of CTCS on-board equipment

作为列控系统的核心,车载设备有着结构复杂,模块间联系紧密的特点,若在行车过程中发生故障,将直接影响列车安全高效运行。为使维修人员准确掌握车载设备故障情况,借助智能化的手段对蕴含丰富经验信息的车载故障维修日志进行研究,有着重要的现实意义。研究通过分析该类日志特点,提出自顶向下与自底向上相结合的方式构建车载设备故障知识图谱。以车载故障维修日志实体关系转换为基础,将半结构化数据实体识别视为关键短语提取问题,提出词向量、主题模型与词典特征相结合的方法先获取关键词语,再通过Bi-gram模型将故障词语拼接为候选故障短语,其中评分最高者即为所需故障实体。实体间的关系则采用基于模式匹配的方法,构建车载故障关系模板,挖掘故障间的联系。对于识别实体的冗余和错误问题,利用实体向量间的余弦相似度计算,通过阈值设定实现实体融合,完成车载设备故障的知识挖掘。最后,以某铁路局2019~2020年车载故障维修日志为数据进行实验,累积抽取出故障实体339个,故障关系734条,据此构建车载设备故障知识图谱,并以可视化方式展示和检索车载设备故障间关系,有效提高了车载故障日志的知识发现能力,便于指导车载设备故障维修。

As the core of the CTCS, the on-board equipment has the characteristics of complex structure and close connection between modules. During the operation, if a fault occurs, it will directly affect the safe and efficient operation of the train. In order to make the maintenance personnel accurately grasp the fault situation of the on-board equipment, it is of great practical significance to study the fault maintenance log containing rich experience information by means of intelligent means. By analyzing the characteristics of such logs, the paper proposed a combination of top-down and bottom-up methods to construct a fault knowledge graph of in-vehicle equipment. Based on the entity relationship conversion of vehicle fault maintenance log, the entity recognition of semi-structured data was regarded as a key phrase extraction problem. A method combining word vector, topic model and dictionary features was proposed to obtain key phrases firstly, then use Bi-gram model to extract key phrases. The fault words were spliced into candidate fault phrases, and the one with the highest score was selected as the required fault entity. For the relationship between entities, the method based on pattern matching was used to construct the on-board equipment fault relationship template and mine the relationship between the faults. To solve the redundancy and errors of entities, the cosine similarity calculation between entity vectors was used and entity fusion was realized through threshold setting, so as to complete the knowledge mining of on-board equipment failures. Finally, an experiment was carried out using the 2019~2020 on-board fault maintenance log of a railway bureau as the data, and 339 fault entities and 734 fault relationships were extracted accumulatively.Based on the research, a fault knowledge graph of on-board equipment was constructed, and the graph was constructed and displayed and retrieved in a visual way. The relationship between equipment faults effectively improves the knowledge discovery ability of the on-board fault log and facilitates the maintenance of on-board equipment faults.