基于自适应模态分解与半监督时空融合网络的电力套管故障诊断方法OA

To improve the effectiveness of feature extraction in bushing fault diagnosis of 220 kV substation GIS equipment and alleviate the difficulty of model training caused by the scarcity of labeled samples,a fault diagnosis methods based on PKO-VMD feature enhancement and a semi-supervised spatiotemporal fusion network is proposed.First,the Pied Kingfisher Optimizer(PKO)algorithm is used to adaptively optimize key parameters of variational mode decomposition,achieving high-precision decomposition and sensitive mode enhancement of non-stationary monitoring signals.Then,a CNN-BiGRU-Attention deep fusion network is constructed to extract local features,bidirectional temporal dependencies,and key discriminant information of multimodal signals.Finally,a semi-supervised learning framework consisting of a supervised term,a consistency constraint term,and an entropy minimization term is introduced to collaboratively utilize a small number of labeled samples and a large number of unlabeled samples to complete model training.Experimental results on the 220 kV substation GIS equipment bushing monitoring dataset show that the proposed method achieves a fault diagnosis accuracy of 96.7％,demonstrating superior overall performance compared to comparative methods;even with only 20％labeled samples,it still maintains a diagnostic accuracy of 94.2％.The results show that the proposed method can effectively improve the accuracy,stability and generalization ability of bushing fault diagnosis for GIS equipment,providing a feasible technical approach for intelligent operation and maintenance of power equipment.