Research on an Autonomous Fault Identification System for Unmanned Aerial Vehicle Distribution Networks Based on Deep Learning

Addressing the challenges of reliable multitask message transmission and stable formation communication for drones in distribution network inspection tasks, this paper proposes a deep learning-based method for autonomous identification of distribution network faults and assessment of message transmission reliability. First, the unmanned aerial vehicle (UAV) network is modeled as an undirected graph based on signal-to-noise ratio (SNR) and packet loss rate (PLR). A task message quadruple model [Formula: see text] is proposed to uniformly describe the packet length, quantity, delay requirements and loss tolerance of different types of messages. Second, combining M/D/1 queueing theory with TDMA fixed time-slot allocation strategy, a computational model for multi-hop transmission delay and success rate is established to analyze the impact of formation structure, formation scaling factor, number of UAVs, SNR environment, transmission rate and network load on message transmission reliability. In a real urban environment, 17 drone service stations and 100 task points were deployed to conduct multi-drone collaboration experiments, and the clustering contour coefficient was used to optimize task point allocation and path planning. The experimental results show that the proposed method can significantly improve message transmission reliability, reduce task latency and maintain formation communication stability under dynamic topology changes in complex urban environments, providing theoretical and engineering references for the application of drones in large-scale collaborative tasks such as power line inspections.