Energy-Efficient Irregular RIS-Aided UAV-Assisted Optimization: A Deep Reinforcement Learning Approach

Reconfigurable intelligent surfaces (RISs) enhance unmanned aerial vehicle (UAV)-assisted communication by extending coverage, improving efficiency, and enabling adaptive beamforming. This article investigates a multiple-input–single-output (MISO) system where a base station (BS) communicates with multiple single-antenna users through a UAV-assisted RIS, dynamically adapting to user mobility to maintain seamless connectivity. To extend UAV-RIS operational time, we propose a hybrid energy harvesting resource allocation (HERA) strategy that leverages the irregular RIS on/off capability while adapting to BS-RIS and RIS-user channels. The HERA strategy dynamically allocates resources by integrating nonlinear radio frequency (RF) energy harvesting (EH) based on the time-switching (TS) approach and renewable energy (RE) as a complementary source. A nonconvex mixed-integer nonlinear programming (MINLP) problem is formulated to maximize EH efficiency while satisfying quality of service (QoS), power, and energy constraints under channel state information (CSI) and hardware impairments (HIs). The optimization jointly considers BS transmit power, RIS phase shifts, TS factor, UAV trajectory, and RIS element selection as decision variables. To solve this problem, we introduce the energy-efficient deep deterministic policy gradient (EE-DDPG) algorithm. This deep reinforcement learning (DRL)based approach integrates action clipping and Softmax-weighted Q value estimation to mitigate estimation errors. Simulation results demonstrate that the proposed HERA method significantly improves EH efficiency, reaching up to 85.8% and 69.8% in single-user (SU) and multiuser (MU) scenarios, respectively, contributing to extended UAV operational time. Additionally, the proposed EE-DDPG model outperforms existing DRL algorithms while maintaining practical computational complexity.