融合强化学习和模糊逻辑的DTN路由算法

Routing Algorithm Combining Reinforcement Learning and Fuzzy Logicin DTN

在延迟容忍网络中,网络拓扑结构动态变化,节点间通信受限,如何选择优秀的中继节点进行消息转发以提高消息投递成功的机率是容迟网络领域重要的研究课题.本文提出了融合强化学习和模糊逻辑的DTN路由算法(A routing algorithm combining reinforcement learning and fuzzy logic in delay-tolerant network).该算法将DTN中消息传输的路径问题映射为有限的马尔可夫决策过程,针对传统算法对中继节点选择的盲目性、对节点和消息的评价缺乏全面性等问题进行改进.首先使用Q-Learning强化学习算法指导消息选取最佳的中继节点,其次通过模糊逻辑系统对节点和消息进行综合评价,并将评价值应用于路由转发.实验结果表明,与PROPHET,CARA,Epidemic,RLFGRP算法相比,RARF算法在低延时、低耗能的条件下有效提高了消息的投递概率.

Indelay-tolerant network,the network topology changes dynamically,and the communications are limitedbetween nodes.How to select excellent relay nodes for messages relaying to improve the probability of successful message forwarding is an important research topic in the field of delay-tolerant network.This paper presents arouting algorithm combining reinforcement learning and fuzzy logic(RARF):it maps the routing problem in DTN into a finite Markov decision process.Firstly,Q-Learning method in reinforcement learning is used to guide message selecting the optimal relay node.Secondly,the qualities of nodes and messages are comprehensively evaluated by fuzzy logic systems,and the evaluation of valuesare applied to the routing relaying process.Experimental results show that,compared with PROPHET,CARA,Epidemic and RLFGRP algorithms,RARF algorithm can effectively improve message delivery rate with low network overhead and low average delay.