Fast connectivity restoration of UAV communication networks based on distributed hybrid MADDPG and APF algorithm

Abstract

The massive failures of unmanned aerial vehicles (UAVs) caused by malicious attacks or physical faults may lead to the UAV swarm splitting into several disconnected clusters. Each cluster cannot ascertain the statuses and positions of some other UAVs, making restoring connectivity face tremendous challenges. In this paper, a distributed connectivity restoration method for UAV communication networks is proposed, which combines the power of the deep reinforcement learning approach and the artificial potential field (APF) model. First, a connectivity restoration mechanism is designed, including UAV failure identification and UAV position relocation. The UAV position relocation involves an efficient exploration mechanism for outstanding performance in connectivity restoration. Subsequently, a hybrid connectivity restoration algorithm is proposed to train the agents to learn desired mobility strategies by applying multi-agent deep deterministic policy gradient (MADDPG) to accelerate connectivity restoration and APF for collision avoidance and connectivity maintenance within each cluster. The proposed algorithm is distributed and each UAV only utilizes the positions of other UAVs in the same cluster. Finally, the simulation results validate that the algorithm achieves faster connectivity restoration with shorter total motion distances of all operational UAVs than existing methods.