GNNPPOR: A Proximal Policy Optimization Multi-Factor Joint Routing Approach Based on Graph Neural Networks in FANETs

Abstract

Given the significant challenges of low resource utilization, load imbalance, and difficulties in meeting quality of service requirements in Flying Ad Hoc Networks (FANETs) routing protocols, this letter proposes a Graph Neural Network (GNN)-based approach for proximal policy optimization routing (GNNPPOR). The approach aims to integrate traffic engineering into FANETs to effectively distribute network load and meet quality of service requirements. In GNNPPOR, we design a GNN model that first aggregates multi-dimensional network state information efficiently through a message-passing mechanism. Subsequently, the network state is updated in real-time using a gated recurrent unit to adapt to dynamic changes in the FANETs network state. Finally, a multi-factor joint decision-making approach is proposed to identify suitable routes for each traffic based on the current network state. Simulation results demonstrate that GNNPPOR outperforms existing methods in several key metrics. Specifically, packet delivery rate increased by 25.3%, while energy consumption and network jitter decreased by 12.8% and 24.9%, respectively.