A FAULT DIAGNOSIS KNOWLEDGE GRAPH CONSTRUCTION FOR DISTANCE SUPERVISED RELATIONSHIP EXTRACTION

To address the issues of insufficient labeling of multi-source corpora and the impact of noisy statements on relationship extraction, we propose a fault diagnosis knowledge graph construction method. It designed a fault diagnosis knowledge ontology and defined a conceptual knowledge model. It presented a relation-aware attention enhanced piecewise convolutional neural network with reinforcement learning algorithm to extract relationships of entity pairs. As an example of island photovoltaic power plant fault diagnosis events, the experiments show that the proposed method can effectively predict the relationship for unlabeled data, and has higher accuracy compared with baseline methods that can provide engineers with effective fault diagnosis decisions.