Abstract
This research introduces a new intelligent routing and resource allocation algorithm called Graph Sample and Aggregate-Multi-Agent Proximal Policy Optimization (GraphSAGE-MAPPO), which targets dynamic wireless mesh networks like those present in emergency communications. Aiming to address the emergency communication scenario where the network topology changes dynamically and the introduction of Artificial Intelligence (AI) model training services leads to more diverse user services and more dynamic node resource capabilities, a three-dimensional mesh network intelligent routing and resource allocation algorithm, GraphSAGE-MAPPO, based on Graph Neural Networks (GNN) combined with Deep Reinforcement Learning (DRL), is proposed. During the training process, the algorithm first uses GNN as a network feature extraction module to extract the resource capabilities and link status indicators of the nodes, thereby generating a hidden feature vector representation for each backbone mesh node; then, the feature vectors of each node are combined with the arrival service flow as the state input of the distributed multi-agent DRL model, supporting efficient routing and resource allocation decisions for service flows with different user Quality of Service (QoS) requirements. Simulation results show that in the face of dynamically changing network environments and user needs, the GraphSAGE-MAPPO algorithm proposed in this thesis can flexibly adjust routing strategies to better meet the QoS requirements of various services and has good generalization performance for network topology and resource changes. These results demonstrate that the algorithm has good flexibility and scalability in large-scale, real-world wireless mesh network environments.