부산대학교 도서관

To solve the problem of transformer fault diagnosis, this paper proposes a transformer fault recommendation algorithm based on knowledge graph: the triplet in knowledge graph is transformed into a word vector according to the word embedding transformation method, and then the word vector is initialized, and the transformer components, fault types and phenomena are associated according to the distance translation model, and the confidence is obtained. The auxiliary push part assigns the weight according to the severity of the result caused by the transformer fault, and adjusts the proportion of the severity of the fault in the recommended result according to the range mapping to obtain the recommended proportion coefficient. When a fault is queried, the knowledge with high confidence is returned. If the fault is queried for the first time, the proportional weighting is recommended. Otherwise, the linear weighting is performed based on the historical query results to obtain the final output result. Through simulation, the AUC index of the model is close to 0.9, and the first three pushed can basically cover the failure and reduce the economic loss.