Research On Routing Algorithms Of Wireless Mesh Network Based On Machine Learning

Wireless Mesh network has the advantages of flexible network configuration,high link bandwidth,non-line-of-sight multi-hop transmission,good compatibility,dynamic adaptability and low cost,and has been widely deployed to the Internet of Everything network scene.Routing technology is the core technology for wireless Mesh network routers to forward packets.The routing discovery mechanism of the traditional wireless Mesh network routing algorithm is driven by mathematical model,which can’t adapt to the changing network state in real time quickly,and can easily cause the phenomenon of uneven allocation of network resources.Machine learning routing algorithms are often data-driven,so they can infer the best decisions by analyzing past network statistics and performance data.In recent years,the classification of machine learning techniques used to solve routing problems mainly includes supervised learning and intensive learning.Supervised learning requires a large number of training data sets,which are difficult to obtain in an existing network environment.The routing algorithm based on intensive learning can overcome the deficiency that supervised learning requires a large number of training data sets,and is more suitable for routing problems.The existing enhanced learning routing algorithm based on distributed multi-hop wireless network mainly studies the reward function according to the optimization target,but the routing load balancing performance,including link interference and node load,is not sufficiently considered.Therefore,it is of great significance to design reward functions from the perspective of load balancing based on intensive learning to model wireless Mesh network routing problems.Aiming at the insufficient consideration of the existing distributed reinforcement learning routing scheme for load balancing routing performance including link interference and node load,the thesis designs a QLNLIA(Q-Learning Load Interface Aware)routing algorithm that meets the routing characteristics of wireless Mesh networks.The reward function including link interference and node load,so that it can continuously learn according to the changes of the network state to adapt to the dynamic changes of the wireless medium and network topology,and dynamically adjust the routing strategy to select the next hop relay node to forward data pack.The simulation results show that the QLNLIA algorithm effectively avoids the heavy load area of the network and achieves the effect of load balancing.The network joint point in the wireless Mesh network carries the traffic of the internal network interacting with the Internet network,and the wireless Mesh network research usually equips the wireless Mesh network with multiple network joint points to relieve the pressure of the bottleneck joint point.In view of the current multi-gateway wireless Mesh network scenario distributed intensive learning routing algorithm does not uniformly consider the gateway load and the quality of the arrival gateway path,the thesis proposes a QLGSLIA(Q-Learning Gateway Selection Load Interface Aware)algorithm.QLGSLIA algorithm based on the QLNLIA algorithm proposed in the previous chapter.Reward functions are designed for Internet business and non-Internet business separately,and the gateway’s listening range is divided into hotspots,so that non-Internet business actively avoids the gateway node;for Internet business,network parameters such as gateway load and link interference load are taken into consideration.The learning method is combined with the comprehensive load situation of the path to the gateway to select the gateway node reasonably to improve the network throughput as a whole.The simulation results show that the reinforcement learning method proposed in this research comprehensively considers the various routing indicators of the wireless Mesh network,and can flexibly adjust routing strategies,make full use of gateway resources,ensure service quality,improve overall network throughput,and reduce average end-to-end delivery delay.Finally,the thesis summarizes the research work of the whole text and looks forward to the next research work.