计算机工程

• •    

基于图神经网络的智能路由机制

  

  • 发布日期:2020-12-18

Intelligent Routing Scheme based on Graph Neural Networks

  • Published:2020-12-18

摘要: 随着当前网络规模的增加和网络应用不断增多,网络流量呈现出了复杂的动态性,使最优路由策略成为NP难问题。现有基于人工智能实现的网络路由方案在方案的泛化性能上有待改进。对此,本文提出了基于深度强化学习的智能路由策略SmartRoute,通过实时感知网络中流量分布状态,SmartRoute能够动态调整路由策略,从而提升路由策略性能。SmartRoute结合了图神经网络的拓扑信息感知能力和深度强化学习的自我训练能力,极大提升了网络路由策略的智能性。实验结果表明,SmartRoute比当前最优方案节省最多9.6%的端到端时延,并且具有更好的鲁棒性。

Abstract: With the increase of network scale and network application, the network traffic presents a complex dynamicity, which makes the optimal routing strategy an NP hard problem. Existing aitificial intelligence-based routing schemes have a bad generalization ability. In this regard, this paper proposes an intelligent routing strategy based on deep reinforcement learning called SmartRoute. SmartRoute can dynamically adjust the routing strategy by a real-time perception of the traffic distribution in the network, so as to improve the performance of the routing strategy. Smartroute combines the topology information perception ability of graph neural network and the self training ability of deep reinforcement learning, which greatly improves the intelligence of network routing strategy. Experimental results show that smartroute can save up to 9.6% of the end-to-end delay compared with the state-of-the-arts and shows a better robustness.