| 摘要: |
| 针对网络服务质量的多目标优化问题以及难解的网络图结构问题，提出了一种基于多智能体深度强化学习联合图神经网络的工业软件定义网络智能路由算法。该算法主要采用多智能体系统通过分布式协同控制，优化业务时延要求、网络通信量、链路负载3个指标，针对一般无法实现网络场景通用化的模型，采用图神经网络进行图结构消息传递，同时采用动态权重配比方法对多目标问题进行整合，优化网络性能。实验结果表明，相对于深度Q网络（Deep Q-network,DQN）算法，所提算法在满足时延要求的业务流数量上平均增加了19.70%，在网络通信量上提高了17.35%，在链路负载平衡上实现了12.04%的改进，有效提高了网络服务质量和性能。 |
| 关键词: 软件定义网络(SDN) 多目标优化 多智能体深度强化学习 网络服务质量 |
| DOI：10.20079/j.issn.1001-893x.240513002 |
|
| 基金项目:国家自然科学基金重点项目(U19B2015) |
|
| Industrial SDN Intelligent Routing Algorithm Optimization Based on Multi-objective |
| ZHANG Xiaoli,LIU Xiaxi,LEI Yusheng,WANG Bin |
| (School of Communication and Information Engineering,Xi揳n University of Science and Technology,Xi揳n 710600,China) |
| Abstract: |
| A software defined network(SDN) intelligent routing algorithm based on multi-agent deep reinforcement learning combined graph neural network is proposed to solve the problem of multi-objective optimization of network quality of service and difficult network graph structure.The algorithm mainly adopts the multi-agent system to optimize three indexes of service delay requirements,network traffic and link load through distributed cooperative control.For the models that are generally unable to realize the universality of network scenarios,the graph neural network is adopted to carry out graph structure message transmission,and the dynamic weight matching method is adopted to integrate the multi-objective problems and optimize the network performance.Experimental results show that compared with the deep Q-network(DQN) algorithm,the proposed algorithm has increased the number of traffic flows meeting the delay requirements by an average of 19.70%,improved network traffic by 17.35%,and achieved a 12.04% improvement in link load balancing,effectively enhancing the quality and performance of network services. |
| Key words: software defined network(SDN) multi-objective optimization multi-agent deep reinforcement learning network quality of service |
《电讯技术》编辑部 