Machine Learning Empowered Intelligent Routing Algorithm In Wireless Networks

In recent years,with the rapid growth of network applications,routing protocols in ad-hoc networks should not only meet the needs of services transmission,but also enhance the adaptability of routing,reduce the complexity of routing mechanism,decrease routing overhead,and make efficient use of limited network resources.At present,the traditional routing protocols can be divided into table driven routing and on-demand routing according to the routing discovery strategy.Among them,table driven routing requires nodes participating in the routing to obtain the global network topology information,which will bring huge network load.The on-demand routing will bring high latency because it needs a routing operation before transmission of data packets.At present,using artificial intelligence method to optimize the routing is a very promising direction.Therefore,based on reinforcement learning and deep learning methods,this thesis studies how to solve the problems of current routing protocols,such as delivery delay,loss of packets and adaptability of routing,and proposes two innovative intelligent routing algorithms.Firstly,this thesis proposes RL-QAQ(Reinforcement Learning Empowered Qos-Aware Adaptive Q-Routing Algorithm)routing algorithm for the scenario that nodes can only obtain local routing information.The adaptive probability is designed to optimize the full echo search strategy of Q-routing to reduce the routing overhead of obtaining network state.The Qo S-aware reward function and Q-table are designed to support multiple Qo S transmission requirements of various services.In addition,the RL-QAQ routing algorithm is simulated and verified based on NS-3 simulation platform.The simulation results show that RL-QAQ routing algorithm can adaptively adjust the routing strategy according to the changing network environment to achieve low delay and low overhead,and can provide differentiated transmission for different services with different Qo S requirements to meet the various transmission requirements of services.Secondly,this thesis proposes DLA(Deep Learning Empowered Adaptive Routing Algorithm)routing algorithm for the scenario that there is a central node with global routing information.Convolutional neural networks(CNN)is used to identify network features,evaluate network performance and guide routing decisions.By combining the network traffic features and link state features into the network features matrix as the input of CNN,and designing a network performance evaluation method including a variety of routing information indicators,CNN can more comprehensively and accurately identify the network characteristics and judge the actual performance of the network,and then make better routing decisions.In addition,the DLA routing algorithm is simulated based on NS-3 and Tensorflow.The simulation results show that the DLA routing algorithm can accurately identify the characteristics of the current network,judge the actual performance of the current network,and adjust the routing strategy in time,which can avoid network congestion and meet the transmission requirements of services.