Research on the Construction of Fault Knowledge Graph for Wind Power Hydraulic Equipment

Hydraulic systems are widely used in wind turbines. However, with the extension of the operation time, the failure rate of the hydraulic system gradually increases, which seriously drags down the operation efficiency of the wind turbine. Therefore, it is urgent to find a fast and accurate fault identification method to identify the abnormality of the wind turbine hydraulic system. The hydraulic system failure analysis documents collected over the past 15 years constitute a data set that has been processed by the BIO annotation technique to make it suitable for analysis. Based on the classic Bert-BILSTM-CRF entity recognition model, an optimized version is developed. Firstly, Bert model is used to collect and extract the features of scattered light; then, these features are fused with the output vector of BILSTM model; finally, the CRF model is used to complete the classification of labels. By embedding the strategy of adversarial learning in the BERT architecture, the robustness of the entity recognition model is successfully enhanced. Subsequently, we will analyze the overall architecture of the obtained triad information and save it in the Neo4j graph database to promote its adaptability. Finally, with the help of Python, we have created a system of fault knowledge mapping. The study ultimately revealed that the optimized model achieved a superior performance of 93 on the F1 score. Up by 1.8 percentage points. The proposed model exhibits an enhancement in performance of roughly 56 percentage points over the Bert-BILSTM-CRF model.