Deep Reinforcement Learning Based UAVs Trajectory Optimization for Maximum Communication Coverage of Users

With its flexibility and easy deployment, UAV can timely provide communication for areas with damaged infrastructure. However, due to some constraints of the UAV itself, such as limited energy and limited communication range, designing reasonable trajectory for UAV to improve the system performance has been studied. Considering UAV energy and user movement, we propose a deep reinforcement learning-based method of maximizing the number of users covered and saving energy (MUSE) to control UAVs. By observing system state, then the UAVs perform the optimal actions to maximize service users and prolong this network lifetime by maximizing the system reward. Simulation experiments show that the algorithm is superior to Q-learning in maximizing service users, moreover, with increasing number of UVAs, the system energy consumption is stable, which achieves the purpose of saving energy.