| As a non-center network with dynamic network topology,without infrastructure and central management,mobile Ad Hoc network is increasingly applied in the fields of transportation,military and emergency disaster relief with its advantages of flexible deployment,fast networking and high cost performance.Due to the mobility of nodes in Ad Hoc network,frequent changes in network topology structure,frequent breakage of communication links between nodes,and high complexity of routing,designing a dynamic routing protocol with strong adaptability to network topology and good stability is the key to ensure the communication performance of Ad Hoc network.Therefore,in this paper,Greedy Perimeter Stateless Routing(GPSR)is selected as the basis to design an improved routing protocol in order to improve network topology adaptability and data transmission efficiency.In this paper,the nodes of GPSR routing protocol have insuficient perception of the surrounding network topology structure and node distribution characteristics,insufficient use of network history information,and incomplete consideration of factors when selecting the next hop forwarding node during route establishment.Combined with deep reinforcement learning algorithm,data mining idea and swarm intelligence algorithm,GPSR protocol is optimized and improved.In order to improve the perception ability of GPSR routing protocol to network node distribution characteristics,make full use of network historical data information,strengthen the adaptability of GPSR protocol to network topology structure,and alleviate the problem that GPSR protocol is easy to fall into the routing void,An improved GPSR protocol based on Double Deep Q-Network(DDQN)algorithm is proposed.This protocol is based on reinforcement learning algorithm to model the routing process of Ad Hoc network and design reward function according to the characteristics of GPSR protocol.Based on the rough set theory,a potential communication node evaluation mechanism based on rough set and fuzzy comprehensive evaluation is proposed,and the degree of potential communication nodes calculated by this mechanism is taken as a part of the reward function,which improves the topology perception ability of nodes in the routing process.Finally,the real traffic node data is used as the basis of Ad Hoc network mobile model and node distribution model,and the modeling simulation and performance verification of the protocol are carried out.In order to solve the problem that GPSR has a single factor when selecting the next hop node,which leads to the high packet loss rate of the node,combined with swarm intelligence algorithm,a GPSR protocol based on Harris Eagle optimization algorithm is proposed.Firstly,a population initialization strategy based on chaotic mapping and an escape energy updating strategy based on nonlinear factors are proposed to solve the problem of the diversity of population initialization and the lack of global search capability of Harris Eagle optimization algorithm.Then,considering node energy,message buffer queue remaining capacity,neighbor node connectivity and other node and link characteristics,the routing cost function was constructed,which was used as fitness function,and the information fusion was carried out by the improved Harris Eagle algorithm to optimize the next hop node selection process.Finally,an Ad Hoc network model is constructed based on the traffic node data,and the simulation and performance analysis of the improved protocol are carried out. |
