GAT-based Optimization for Multipath Routing

Routing is the basis for the transmission of mobile ad-hoc networks (MANET). Global routing can achieve theoretically optimal transmission efficiency, and it is crucial to use network information to filter out the optimal routing. This paper proposes a multipath routing performance prediction model based on a graph attention network (GAT). The ability of GATs to model the structure and function of data makes it possible to determine the characteristics of links, paths, nodes, etc., and the complex relationships between them, to predict network throughput and stability. Finally, using a software-defined network (SDN), routing can be optimized by predicting the network performance of alternative routings. The simulation results showed that this method improves throughput by 5-25%, and network stability by 10-30%, under dynamic topology conditions compared with other SDN-controlled MANET routing protocols.