UAV Position Optimization for Servicing Ground Users Based on Deep Reinforcement Learning

Abstract A Deep Q-Network (DQN) algorithm is proposed for optimization of UAVs, which can increase the communication rate of multiple users within a certain area. The UAV is able to automatically adjust its position in 3D by means of the DQN Agent’s design and training, and then get the best UAV placement location. Furthermore, the Double Deep Q-Network (Double DQN) has been researched. It has been proved that this method has higher efficiency and higher convergence rate compared with traditional DQN for locating optimal UAV position. In this paper, we perform a complex simulation of a random distribution of users in a simulated region. The experimental results indicate that UAV is able to locate the optimal position with maximum mean transfer rate, and reduce the distance from central point to customer centre, thus increasing the communication quality. This study not only verifies the effectiveness of deep reinforcement learning in dynamic positioning optimization, but also provides new insights for the optimization design of future intelligent communication systems.