Research On Delay Tolerant Network Routing Technology Based On Multi-agent Reinforcement Learning

Delay-tolerant network(DTN)is a network that’s designed to operate effectively in heterogeneous networks that may lack continuous network connectivity.Because DTN has the characteristics of uncontrollable transmission delay,intermittent communication,limited network resources and high dynamic topology,it is challenging to design an effective DTN routing protocol.The DTN routing protocol adopts the "store-and-forward" communication paradigm,and the traditional DTN routing algorithm largely relies on the greedy strategy.Such a scheme cannot guarantee that the data packet is finally sent to the destination node,and thus presents poor network performance.The way to design a routing protocol with good network performance in the case of limited resources is an urgent problem to be solved.This paper introduces the positive social attributes of nodes in the design of DTN routing.Firstly,the author introduces a community detection algorithm with a threshold for iterative optimization of modularity,divides DTN nodes into countable communities by connection history,and designs a multi-layer heterogeneous DTN routing architecture by introducing community service unit and global computing center.Based on this architecture,local centrality and global centrality indicators are designed.Furthermore,based on this architecture,considering the cooperation between nodes and fine-grained node buffer occupancy,this paper models the problem of selecting next-hop routing relay nodes as a distributed partially observable Markov decision process,and proposes a Multi-agent reinforcement learning QMIX-aided routing algorithm with centralized learning and distributed execution paradigm.In addition,for distributed network scenarios that lack centralized control,this paper also designs a DQN-based distributed DTN routing architecture,which models the routing problem as a Markov model.The DQN routing algorithm is trained in a distributed manner.By introducing two real wireless data sets and three classic DTN routing schemes,a large number of experiments demonstrate that the routing algorithm proposed in this paper has good performance in delivery success rate,mean delay and network overhead rate from the perspectives of training and testing.