Cloud-edge collaborative multimodal framework for galloping discrimination of overhead transmission lines in electric power systems

Galloping of overhead transmission lines under extreme weather conditions poses a serious threat to the operational safety and reliability of electric power systems. Existing galloping discrimination methods are still limited by constrained edge-side computing resources, complex environmental interference, and the insufficient representational capacity of single-modality data. To address these issues, this paper proposes a cloud-edge collaborative framework for galloping discrimination of overhead transmission lines in electric power systems. On the edge side, a lightweight vision-based perception model is developed for rapid feature extraction and motion analysis, enabling efficient local discrimination. On the cloud side, a multimodal galloping discrimination model with cross-modal attention fusion is constructed to effectively extract and fuse features from visual, micro-meteorological, and inspection text data uploaded from the edge side, thereby improving discrimination accuracy. In addition, a periodic model parameter optimization mechanism is designed based on the discrimination results from the edge and the cloud. To evaluate the proposed framework, a field-collected multimodal galloping dataset is constructed and comprehensive experiments are conducted. Experimental results demonstrate that the edge-side model achieves efficient galloping discrimination, while the cloud-side model exhibits stronger robustness under complex conditions, including snowfall, rainfall, and nighttime scenes, and the proposed cloud-edge collaboration supports periodic performance refinement for long-term overhead transmission-line galloping monitoring.