Multi-agent Load Balancing And Qos Routing Based On Traffic Prediction

Low Earth Orbit(LEO)satellite networks have become an important way to achieve global communications because of the wide coverage,flexible networking,geographical constraints,simple access and other outstanding advantages.Howerver,LEO satellite networks also have the real-time changes of the Inter-Satellite Link(ISL)remaining bandwidth,the ISL package loss rate and the ISL end-to-end delay caused by the dynamics of the LEO satellite networks.These characteristics bring challenges to the design of LEO satellite network routing algorithm.Aiming at the problem that the current LEO satellite network routing algorithm has large network overhead,unsuccessful route update,unbalanced network load and Quality of Service(QoS)is difficult to guarantee,this paper proposes Multi-Agent Load Balancing and QoS Routing based on Traffic Prediction(TP-MALBQR).The main research contents are as follows:Accurate satellite network traffic prediction results reflect satellite load conditions,providing a reliable next hop routing strategy for load balancing and QoS routing.However,considering the complex self-similarity and nonlinearity of satellite network traffic and the limited computing and storage resources of satellites,traditional traffic prediction algorithms can not meet the requirements of prediction accuracy and prediction efficiency.This paper proposes a Gated Recurrent Unit(GRU)neural network traffic prediction algorithm based on transfer learning.In this paper,transfer learning is used to solve the problem of insufficient online satellite traffic data.The nonlinear,self-similar and long-correlation features of satellite traffic are obtained through GRU neural network.The particle filter online training algorithm reduces the time complexity of model training and improves the model training.The convergence efficiency and the distance comparison method and the optimized combination strategy overcome the problem of particle effectiveness and particle diversity in the particle filter algorithm.The simulation results show that the average relative error of the proposed traffic prediction algorithm is 35.80%and 8.13%lower than that of Fractional Autoregressive Integration Moving Average(FARIMA)and Support Vector Regression(SVR).The optimized particle filter algorithm is 40%faster than the gradient descent algorithm.The overall results show that the proposed algorithm has higher prediction accuracy and lower requirements for satellite computing storage resources,and provides good reliability guarantee for multi-agent load balancing and QoS routing algorithms.This paper designs a multi-agent load balancing and QoS routing algorithm based on traffic prediction values.The algorithm is a proxy-based satellite network routing architecture.According to the characteristics of polar orbit constellation,this paper designs the cost of inter-satellite link between orbit,and uses the traffic prediction result to design the cost adjustment factor to accurately reflect the satellite load.This paper designs two types of agent structures:mobile agent and static agent:the mobile agent is responsible for detecting the path with the least cost and QoS path,collecting routing information,and notifying the static agent to reserve resources;the static agent is responsible for traffic prediction,calculating the link cost,and routing table maintenance with update and QoS resource reservation.This paper considers three types of QoS parameters:delay,bandwidth,and packet loss rate.The path candidate set that satisfies the QoS parameters is detected by the mobile agent,and the path with the longest remaining time in the polar region is selected to finally satisfy the load balancing and QoS path.The simulation results show that the network throughput of the proposed routing scheme is increased by 8%to 20%,the packet loss rate is reduced by 9%to 25%,and the end-to-end delay is reduced by 6%to 15%.