| Wireless Mesh Network is a packet-switched network with a static backbone network.In recent years,it has become a good choice for WLAN to expand the public access range because of its low cost,automatic configuration,self-repair,scalability and robustness.The traditional routing algorithm of wireless Mesh network uses mathematical model to drive the routing discovery mechanism,but this method can not adapt to the real-time changing network state quickly,which may lead to uneven distribution of network resources and network congestion.In order to solve this problem,an intelligent routing algorithm based on deep reinforcement learning is proposed.By learning the data flow patterns and laws in the network,the algorithm can automatically select the appropriate routing path and realize the efficient transmission of network traffic.However,deep reinforcement learning technology can’t learn topology information when dealing with routing optimization problems.Once a node fails for some reasons,the network topology will change,and the input needs to be readjusted to train the model,resulting in a large processing delay.Graph neural network is a kind of neural network structure that can effectively deal with irregular topological information,and can update the node feature vectors of the network according to the changes of topological relations.Therefore,it is of great research significance to use graph neural network to solve the routing problem of wireless Mesh network.In order to verify the superiority of wireless Mesh network routing algorithm based on graph neural network,this thesis first realizes the shortest path algorithm of wireless Mesh network based on OMNet++.The Dijkstra algorithm of wireless Mesh network is implemented on OMNet++,and the shortest path from a designated wireless Mesh node(source point)to other wireless Mesh nodes can be found.The average delay and average throughput of the algorithm are analyzed by using the data generated by OMNet++ simulation software,and the relationship between them and networkrelated parameters is obtained,which lays the foundation for optimizing the routing algorithm in the next step.Aiming at the problem that the current routing algorithm can’t guarantee the differentiation of service quality of multi-network and cope with the change of network topology,this thesis further proposes a routing algorithm for wireless Mesh network based on graph neural network(GNN-DRL).Firstly,the algorithm uses graph neural network to generate the hidden feature vector representation of its own nodes,and then uses a fully distributed multi-agent DRL routing model to realize an efficient routing strategy supporting different Qo S requirements.The simulation results show that when the network topology changes,the GNN-DRL algorithm proposed in this thesis can flexibly adjust the routing strategy,improve the overall network throughput and reduce the average end-toend delay.In view of the complexity of the current wireless Mesh network operation environment,it is impossible to observe the network operation in real time.This thesis uses 3d-force-graph-master framework,database technology and front-end technology to build a 3D wireless Mesh network visualization platform.The platform can quickly obtain the attributes and parameters of field devices,intuitively perceive the operation data of the network,and realize the visualization of the network,resources and environment.Based on the network topology,it can also show the key parameters of communication between nodes of wireless Mesh network and query multiple available path routing reports with specified source and destination.The test results show that the platform can monitor the dynamic changes of wireless Mesh network in real time,optimize the communication parameters of the network,and improve the reliability and robustness of network operation. |
