A deep reinforcement learning-based method for reliability analysis of multi-agent communication networks

With the rapid and continuous advancement of unmanned aerial vehicles (UAVs) and communication technologies, UAVs have emerged as the preferred solution for executing long-range missions. Furthermore, UAVs have evolved from operating independently to collaborating within networks to collectively achieve mission objectives. However, increasing mission complexity and communication environments present significant challenges in assessing and optimizing network reliability. To address these challenges in evaluating and optimizing the communication network reliability of UAV swarms during mission execution, this study proposes a deep reinforcement learning-based algorithm. The UAV swarm is modeled as intelligent agents within a deep reinforcement learning framework to facilitate inter-UAV communication assessment and optimization. Initially, individual link reliability is computed using a path-loss model, followed by the calculation of overall network reliability employing ordered binary decision diagrams (OBDDs). A reward function incorporating node movement costs is then constructed. Q-learning is utilized to leverage historical experience, while deep learning trains the reliability optimization policy. The proposed algorithm aims to achieve enhanced system performance. Simulation results indicate that the deep reinforcement learning-based communication network reliability optimization algorithm effectively allocates UAV positions across diverse topologies and communication environments, thereby enhancing overall network reliability.