Federated Cloud-Edge-device Collaboration for Enhanced Risk Detection and Assessment of Anti-External Destruction in Cross-Province and Cross-Region Power Transmission Lines

• Yuankang He,
• Mingjie Xu,
• Congying Hou,

Vol. 19, No. 9, September 30, 2025
10.3837/tiis.2025.09.010, Download Paper (Free):

• Power transmission lines
• Risk Detection
• cloud-edge-device collaborative model
• deep feature fusion model
• risk map construction strategy

Abstract

Ensuring the safety and reliability of power transmission lines is critical for maintaining grid stability. A key challenge in achieving this goal is the timely detection of potential risks from external damage—harm caused to transmission lines by external factors such as large construction machinery, fires, or smoke, which may lead to line failures and compromise power system stability. Currently, most risk detection systems rely on manual inspections or intelligent devices like drones. While these methods have improved efficiency, they still suffer from high costs, insufficient accuracy, and scalability limitations. To address these issues, some researchers have proposed cloud-based solutions that centralize computational tasks to enhance detection accuracy. However, these approaches impose excessive pressure on cloud infrastructure and raise privacy concerns due to massive data transfers. In this work, we propose a cloud-edge-device collaborative model based on Federated learning, named FedTLRD, to achieve accurate and private power Transmission Line Risk Detection. Our approach consists of three key components: (1) DFNet, a Deep Feature Fusion model that improves risk detection accuracy in transmission lines; (2) RFed, a Federated Learning framework that enables distributed risk detection without raw data transmission, ensuring privacy; (3) a risk map construction strategy that provides authorities with a comprehensive overview of regional transmission line risks, facilitating timely maintenance. Extensive experimental evaluations demonstrate that FedTLRD outperforms existing techniques in detecting external damage risks while maintaining computational efficiency and data privacy.

Statistics

Cite this article