DEEP REINFORCEMENT LEARNING-BASED STUDY ON REACHABILITY SPEED FOR UAV WITH INTEGRATED RIS

In order to cope with the communication difficulties caused by the large user base and the transmission signals that are usually blocked by obstacles in the communication scenario of Internet of Things (IoT) devices, an airborne unmanned aerial vehicle (UAV) carrying reconfigurable intelligent surfaces (RIS) is designed to facilitate relay cooperative communication. Deep Reinforcement Learning (DRL) algorithms are used to jointly optimize the user's reachable speed and the UAV's energy consumption, ensuring the reachable speed for users. First, the system is analyzed and modeled considering the characteristics of multi-user mobility and mutual interference among users. Second, for the established non-convex optimization problem, a weighting function is designed for the user's reachable speed and UAV energy consumption, and joint optimization is performed using deep reinforcement learning. Finally, the simulation results show that, after reaching stabilization, the Softmax Deep Double Deterministic Policy Gradients (SD3) algorithm based on reachable speed and UAV energy consumption performs optimally in terms of reward value compared to other advanced deep reinforcement learning algorithms. Moreover, the system's reachable speed correlates positively with the number of RIS reflector units. Increasing these units significantly enhances communication link effectiveness by boosting signaling rates for users.