GNN-based Inspection UAV-assisted Computing Task Scheduling Method for Power Networks

Modern power systems impose dual challenges on data networks: increasing computational tasks and stringent low-latency requirements. Traditional cloud computing struggles to meet these real-time demands in power IoT, whereas edge computing and UAV-assisted offloading offer promising alternatives. This paper proposes an inspection UAV-assisted task scheduling method to enhance grid terminal throughput and fairness while reducing system energy consumption via optimized resource allocation and path selection. We develop a scheduling architecture integrating multiple UAVs and grid terminals, and formulate a joint optimization problem balancing energy consumption, throughput, and fairness. To solve this, we introduce DGDCMM—a multi-agent reinforcement learning algorithm that combines a Double-layer Graph Attention Network with a Discrete Communication Model. UAVs process local observations through a Feature Extraction Network (FEN) and share information via a Discrete Communication Network (DCN) under centralized training with distributed execution. Simulation results show that DGDCMM outperforms baseline methods in overall throughput, resource uniformity, and energy efficiency.