Aiming at the problems of traditional distribution network monitoring system, such as lack of real-time performance, bandwidth pressure and early warning lag, this paper proposes an intelligent monitoring and risk diagnosis system for distribution network equipment based on edge computing. The system adopts a three-tier architecture of terminal perception-edge reasoning-cloud collaboration, and realizes high-frequency acquisition of electrical, mechanical, thermal and other multi-source data through Internet of Things sensors. In the edge layer, the attention mechanism is innovatively introduced to realize the adaptive fusion of multimodal data, and a lightweight Mobilene ETV 3-YOLO model is designed. By combining FP16 quantization and TensorRT optimization, the reasoning delay is reduced to 65ms, which significantly improves the real-time diagnosis ability of the edge. In the cloud, a dynamic risk assessment framework is built through Hidden Markov Model (HMM) and digital twinning technology to realize the early warning of equipment degradation more than 2 hours in advance. The experiment is based on the 12-month data of the real distribution network. The system is significantly superior to the traditional cloud solution in terms of end-to-end response delay (195ms), bandwidth consumption (0.28TB per month), fault recall rate (94.8%) and early warning time window (> 150 minutes), and the false alarm rate is reduced by 85%. The research results provide an efficient and reliable edge computing solution for intelligent operation and maintenance of distribution network.
Shu et al. (Sun,) studied this question.