Deep Reinforcement Learning for Multi-UAV MEC System Trajectory Optimization: Minimizing Energy Consumption

In mobile edge computing (MEC) systems, unmanned aerial vehicles (UAVs) have garnered significant research and application attention due to their high flexibility and ease of deployment. However, given the limited energy resources carried by UAVs, a critical challenge is how to reduce system energy consumption while ensuring service delivery to users. Additionally, the dynamic nature of MEC network resources makes energy optimization even more challenging. Therefore, this study constructs an optimization objective in a disaster scenario involving multiple UAVs and multiple users by building energy consumption model. Based on this optimization objective, the study fully considers user mobility and proposes the MU-MADDPG algorithm to address this issue. In this system, each UAV acts as an independent agent, collaborating to optimize UAV flight paths. Furthermore, to accelerate the learning process and improve its efficiency, this study introduces a prioritized experience replay (PER) mechanism. Experimental results demonstrate that the proposed algorithm outperforms other algorithms in terms of convergence speed and system energy consumption.