Topology Maintenance Optimization Algorithm Based on Deep Reinforcement Learning in High Dynamic Flying Ad-Hoc Networks

In the realm of high-dynamic-flight self-organizing networks, traditional proactive and reactive routing protocols confront challenges arising from the delayed and inflexible determination of the validity of neighbour nodes. An untimely assessment of node validity may lead to an escalated overall network packet loss rate, consequently causing a proportional decrease in throughput. To mitigate this issue, this paper introduces a Reinforcement Learning based Topoloty Maintenance Algorithm (RLTM). This algorithm integrates the mobility characteristics and remaining energy of neighbour nodes into a stability metric for these nodes. By conceptualizing the validity of neighbour nodes in topology maintenance as a Markov process and incorporating the stability metric as a dimension in the state space, the model undergoes training via Proximal Policy Optimization (PPO) to generate decisions for establishing the effective time of neighbour nodes. Simulation results demonstrate that, in comparison to other topology maintenance optimization algorithms, the proposed algorithm amplifies the average packet delivery rate and throughput of unmanned aerial vehicle (UAV) nodes in high-speed mobility scenarios.