A Multi-Task Deep Learning Method for Power Equipment Fault Detection Based on Edge–Cloud Collaborative Architecture

Timely and accurate fault detection for complex and heterogeneous power equipment, such as transformer faults, line faults, and breaker malfunctions, poses significant challenges due to diverse signal characteristics and overlapping fault signatures. Timely and accurate fault detection in power equipment plays a crucial role in ensuring the safe and stable operation of power grids. Moreover, single-task models are inadequate for handling multiple concurrent fault detection tasks, and purely cloud-based processing suffers from high communication latency and poor real-time responsiveness, which are critical limitations in modern power grid scenarios with stringent resource constraints. However, conventional methods are constrained by limited computational resources and inflexible task processing schemes. Single-task models often fail to meet the demands of multi-task detection, and relying solely on cloud processing leads to high communication latency and insufficient real-time performance. To address these challenges, this paper proposes a multi-task deep learning approach based on an edge–cloud collaborative architecture, named ECMT-Net (Edge–Cloud Multi-Task Network). In this framework, lightweight models are deployed at the edge to perform initial feature extraction and task partitioning, while the cloud server handles global feature fusion and joint multi-task optimization. This design effectively balances local responsiveness and global accuracy. ECMT-Net is capable of performing fault type classification, severity assessment, and fault localization simultaneously, significantly improving detection efficiency and system responsiveness. Experimental results on multiple real-world power datasets demonstrate that ECMT-Net outperforms existing methods, validating its feasibility and effectiveness in practical applications.