An enhanced substation equipment detection method based on distributed federated learning

The inefficiency of manual inspections in substations struggles to meet increasing workloads amid power grid expansion, necessitating intelligent solutions for equipment monitoring. This study addresses two key challenges: detecting diverse equipment under scale variations, occlusions, and real-time constraints, and ensuring data privacy given geographically dispersed, sensitive substation data. We propose CWA-YOLO, a detection framework integrating multi-scale feature fusion and an enhanced small-object detection head into YOLOv8 to improve accuracy across variable conditions. Additionally, a federated learning (FL) system tailored for substations enables collaborative model training without centralized data sharing, addressing privacy concerns and data heterogeneity. The framework’s novelty lies in its dual focus: optimizing detection performance through architectural enhancements and ensuring secure, efficient distributed learning. CWA-YOLO achieves mAP scores of 0.918 ([email protected]) and 0.623 ([email protected]:0.95), surpassing YOLOv8l and YOLOv7l by 6.5% and 7.49%, respectively, in accuracy. For FL, the Federated Adaptive (FedAdp) algorithm reduces communication rounds by 62% compared to Federated Averaging (FedAvg), maintaining near-centralized accuracy while preserving data locality. These results confirm the method’s effectiveness in improving substation equipment recognition securely and efficiently. • Replaced the YOLOv8 backbone with ConvNeXt and introduced BiFPN, along with a specialized small object detection head (SOD) to enhance micro-object detection in indoor substation environments. • Developed and deployed a federated learning system based on NVIDIA FLARE, specifically designed for indoor substation scenarios, and proposed the FedAdp algorithm to enhance model performance and educe communication rounds.