In Wireless Rechargeable Sensor Networks, conventional long-range wireless power transfer technologies employed by mobile charging devices (e.g., Unmanned Aerial Vehicles (UAVs)) are inefficient. Intelligent Reflecting Surface (IRS) can effectively address this issue by enhancing channel gain through the adjustment of phase shifts. However, in most exsiting studies, the placement of IRS is constrained by spatial limitations. To address these problems, a model is formulated in which the IRS is mounted beneath the UAV and a two-stage algorithm based on Deep Reinforcement Learning (DRL) named “SCA-GMAPPO” is proposed. Compared to approaches that use DRL alone to simultaneously optimize UAVs trajectory the phase shift of IRS reflecting elements, the proposed approach significantly reduces computational time. First, Gated Recurrent Unit (GRU) is integrated within the MAPPO framework to accurately capture the trajectory variations and charging duration. Second, the Successive Convex Approximation (SCA) algorithm is employed to optimize the phase shift of IRS reflecting elements, which reduces computational overhead and enhances the energy reception efficiency of sensor nodes. The experimental results demonstrate that SCA-GMAPPO outperforms existing mainstream DRL methods in terms of system energy efficiency and energy fairness. Furthermore, in complex environments with obstacles, UAVs are able to accurately find safe trajectories.
Liu et al. (Wed,) studied this question.