Intelligent Equipment Inspection and Defect Detection Based on Federated Learning and YOLOv4

The current inspection points of substation intelligent equipment such as robots and drones are mainly preset by manual empirical, and it is important to study the inspection point evaluation index and evaluation method to guide the high quality, high efficiency and high intelligence inspection of intelligent equipment. Firstly, an automatic inspection point generation method based on shooting potential model is proposed for the shooting quality, light fit and inspection efficiency of substation intelligent equipment. Then the evaluation index system of inspection point deployment scheme is established, and a comprehensive evaluation method based on improved YOLOv4 method is proposed to determine the subjective weights of indexes by hierarchical analysis method and the objective weights by entropy weight method. Finally, an example calculation is carried out for a 110 kV substation, and the results show that the joint inspection point deployment scheme of robot and UAV based on the shooting potential model is the most suitable optimal scheme for this substation, and an auxiliary verification is carried out in Unreal Engine based on the real images of inspection points generated by each scheme.