Research On Routing Protocol Of Flight Ad Hoc Network Based On Reinforcement Learning

With the reduction of UAV manufacturing costs and the continuous maturity of flight control technology,UAVs have been widely used in various fields.To build a stable UAV cluster system,it is essential to build a multi-UAV communication network with high reliability and low-latency.This article focuses on the research of routing protocols in the Flying Ad Hoc Networks(FANETs),introduces the idea of reinforcement learning,and designs a smarter link detection technology and a smarter routing protocol.First of all,this thesis proposes an adaptive link detection optimization algorithm based on reinforcement learning to solve the problem of network performance degradation caused by topology changes and link maintenance difficulties caused by the high-speed movement of UAV nodes.The node builds a reward function by sensing the degree of change in the number of neighbors and the degree of its business load in a dynamic environment and solves the optimal HELLO time slot to improve the node's link discovery and maintenance capabilities.Besides,for the problem of excessive states caused by continuous state space encountered in the process of constructing the reinforcement learning model,the Kanerva coding function based on the similarity mechanism is used to approximate the state space,reducing the complexity of the algorithm and increasing the stability of the algorithm.Simulation experiments show that this algorithm can effectively improve network throughput,reduce routing overhead,and has the characteristics of self-learning,which is suitable for UAV networking in a highly dynamic environment.Secondly,this thesis analyzes the shortcomings of existing routing protocols used in FANETs,and combines the ideas of trial and error with reinforcement learning and dynamic programming,and proposes a multi-agent routing protocol based on geographic location information.In this protocol,the node evaluates the value of neighbor nodes and guides route selection by measuring the relationship between the neighbor value and the weight of the distance from the destination node when data packets are forwarded.Experiments show that compared with other routing protocols based on geographic location,this protocol reduces overhead,delay,and improves throughput.When the network load is high,balance the link load,and reduce network congestion.