基于知识图谱的风电机组诊断系统构建与应用

Construction and Application of Wind Turbine Diagnosis System Based on Knowledge Graph

针对风力发电机故障诊断与维修过程不明确以及历史故障数据记录大量遗留等问题,提出一种以知识图谱的方式构建的风力发电机故障诊断系统。首先,通过改进的命名实体识别模型BERT-BiLSTM-CRF对故障文本进行知识抽取。数据集采用了近10年来的风力发电机故障案例、事故分析等文本数据。实验结果表明:在风力发电机故障领域中,改进的实体识别方法相比于传统模型效果提升了2.54%。其次,对抽取的知识实体进行结构化分析,由于传统故障树在实际故障推理中缺乏目的性,且每个底事件相对于顶事件的重要性不同,提出以故障的特征属性为分支条件引入到故障树推理中,完成故障树定性与定量分析,并结合故障模式影响和危害性分析(FMECA)完善故障领域知识模型;再对知识结构完成本体化建模,使用Protégé开发工具对故障树结构完成了基于六元组概念的本体建模,使构建的本体知识库满足推理的前提条件。最后,通过Neo4j实现知识模型的可视化,并提升了知识数据的读写能力。

To address the precision problems in wind turbine fault diagnosis and maintenance processes,the lack of management of fault domain knowledge,and the large amount of historical fault data records left behind,a wind turbine fault diagnosis system was proposed to be constructed in the form of a knowledge graph.Firstly,knowledge extraction of fault texts was carried out by an improved named entity recognition model BERT-BiLSTM-CRF.The data set used text data of wind turbine fault cases and accident analysis in the past 10 years.And it was proved through experiments that the improved entity recognition method was 2.54%more effective compared to the traditional model in the wind turbine fault domain.The extracted knowledge entities were then structurally analysed.As the traditional fault tree lacked purpose in actual fault reasoning,and each bottom event had different levels of importance to the top event,it was proposed that the characteristic attributes of the fault was introduced,as branching conditions,into the fault tree reasoning,to complete the fault tree qualitative and quantitative analysis,and the fault mode impact and hazard analysis(FMECA)were combined to refine the fault domain knowledge model.Then Prot g development tools were use to complete the ontology modelling of the fault tree structure based on the six-tuple concept,so that the constructed ontology knowledge base could meet the prerequisites for inference.Finally,the visualization of knowledge model was realized by Neo4j,and the ability of reading and writing knowledge data was improved.