Edge–Cloud Collaboration Detection Approach for Small-Sample Imbalanced Faults in Power Lines

Abstract

Fault detection in power lines is crucial to the reliable operation of power systems. Recent progress in the development of the artificial intelligence-based (AI-based) fault detection for power lines has been receiving an ever-growing attention. Existing AI-based fault detection models mostly rely on the assumption that the power line faults collected in a substation are adequate and balanced. However, obtaining massive and balanced line fault samples is difficult because of the complexity of practical environments and the inherent nature of faults. As a result, the performance of the models gradually deteriorates when the small-sample imbalanced degree of the power line fault set increases. This study proposes a novel edge–cloud collaboration detection method based on transfer and federated learning to address the above issue. Specifically, a fault detection model is established on the basis of a convolutional neural network and pretrained by introducing a new fault set with a sufficient number of labelled samples. The pretrained model is then deployed to each of the substations and fine-tuned in a federated learning manner. By performing fine-tuning in substations and global aggregation in a cloud platform, the proposed method updates the model on the basis of the small-sample imbalanced faults of power lines collected in substations. Case studies verify that after training by using the proposed edge-cloud collaboration detection method, the model can accurately detect faults in power lines even if the fault training set has a small number of samples and is imbalanced.