DRL-Based Joint UAV Trajectory, Nest Scheduling and Resource Optimization for Multi-UAV Aided Power Inspection System

Unmanned aerial vehicle (UAV)-aided inspection system is considered a promising technology due to the high mobility and real-time monitoring capabilities. In this paper, we consider a multi-UAV aided power inspection system with no-fly zones (NFZs), where a set of UAVs complete capturing images/video tasks near the inspection points. In particular, the mobile nests are scheduled to recharge the UAV when its battery falls below a certain threshold. The goal of this paper is to minimize the total inspection time of the system by jointly optimizing the multi-UAV inspection tasks, UAV trajectories, and mobile nest scheduling. On the one hand, as the number of inspection points <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$M$</tex> increases, the algorithm's complexity grows factorially, i.e., <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathcal{O}(M-1)!$</tex>. In addition, the formulated problem is a multivariable coupled non-convex problem, making it difficult to solve using traditional methods such as successive convex approximation or interior-point methods. To address these challenges, we first use the Kmeans++ algorithm to allocate the total tasks to each UAV, reducing the complexity of subsequent trajectory optimization. Then, for each UAV, we design its inspection trajectory based on an modified multi-step dueling double deep Q learning network (MSD-DDQN) algorithm and schedule the mobile nests for UAV recharging utilizing the A* algorithm. Finally, simulations validate the effectiveness of the proposed method. Compared to the baselines, the proposed approach accelerates the algorithm's convergence speed and reduces task completion time by nearly 37%.