煤矿综采设备故障知识图谱构建

Fault knowledge graph construction for coal mine fully mechanized mining equipment

现有煤矿综采设备故障诊断方法缺乏对综采设备历史故障数据的系统化管理及应用,针对该问题,引入知识图谱技术对综采设备故障数据进行系统化管理。采用自顶而下的方法对综采设备故障知识进行本体构建,将综采设备故障知识归纳为故障位置、故障现象、故障原因、处理方法 4类,并进行规范化命名;采用通用的命名实体标注方法 BIOES对综采设备故障知识进行人工标注;将双向长短期记忆(BiLSTM)和条件随机场(CRF)相结合,构建BiLSTM-CRF模型,对已标注的综采设备故障知识进行命名实体识别,并通过人工抽取实体关系,从而实现故障知识抽取;结合BiLSTM-CRF模型的实体识别结果和人工抽取的实体关系,使用Neo4j图数据库存储综采设备故障知识,构建综采设备故障知识图谱。实验结果表明,相较于BiLSTM模型和BiLSTMAttention模型,BiLSTM-CRF模型精确率显著提高,为87%,F1值也有一定幅度上升,为69%。综采设备故障知识图谱的构建可为大规模、多域综采设备故障数据的有效分析、管理及应用提供支持。

The existing fault diagnosis methods for coal mine fully mechanized mining equipment lack systematic management and application of historical fault data of fully mechanized mining equipment.In response to this problem,knowledge graph technology is introduced to systematically manage the fault data of fully mechanized mining equipment.The top-down approach is used to construct the ontology of fully mechanized mining equipment fault knowledge.The knowledge of fully mechanized mining equipment fault is classified into four categories:fault location,fault phenomenon,fault cause,and treatment method.And the naming of the knowledge is standardized.The universal naming entity annotation method BIOES is used to manually annotate the fault knowledge of fully mechanized mining equipment.By combining bi-directional long short-term memory(BiLSTM)and conditional random field(CRF),the BiLSTM-CRF model is constructed.The marked fault knowledge of fully mechanized mining equipment is identified by the named entity,and the fault knowledge extraction is realized by manually extracting entity relationships.Combining the entity recognition results of the BiLSTM-CRF model with the manually extracted entity relationships,a Neo4j graph database is used to store the fault knowledge of fully mechanized mining equipment.A fault knowledge graph of fully mechanized mining equipment is constructed.The experimental results show that compared to the BiLSTM model and BiLSTMAttention model,the acurracy of the BiLSTM-CRF model is significantly improved,reaching 87%.The F1 value also has a certain increase,reaching 69%.The construction of fully mechanized mining equipment fault knowledge graph can provide support for the effective analysis,management,and application of large-scale and multi-domain fully mechanized mining equipment fault data.