摘要
在控制与转发解耦的软件定义网络(SDN)架构中,控制平面通过全局感知和集中决策为路径调度提供了结构基础。深度强化学习凭借出色的状态感知与策略自适应演化能力,适用于高动态网络环境中的控制策略优化。文章提出一种基于深度Q网络的SDN交换机转发路径智能调度方法,从状态空间构建、动作定义、奖励函数设计、策略网络训练及路径执行5个维度构建完整的调度框架。通过融合经验回放和目标网络更新机制,该方法引导策略对路径时延、链路负载与跳数变化进行综合优化,从而有效提升交换机在动态业务条件下的路径自适应能力。
In the Software Defined Network(SDN)architecture where control and forwarding are decoupled,the control plane provides a structural basis for path scheduling through global awareness and centralized decisionmaking.Deep reinforcement learning is suitable for control policy optimization in highly dynamic network environment due to its excellent state perception and policy adaptive evolution ability.This paper proposes an intelligent scheduling method for SDN switch forwarding path based on Deep Q-network,which constructs a complete scheduling framework from five dimensions:state space construction,action definition,reward function design,policy network training and path execution.By integrating experience replay and target network update mechanism,this method guides the strategy to comprehensively optimize the path delay,link load and hop count change,so as to effectively improve the path adaptive ability of switches under dynamic traffic conditions.
作者
王迪
WANG Di(Armed Police Xiangyang Detachment,Xiangyang,Hubei 441000,China)
关键词
SDN
深度强化学习
Q网络
路径调度
网络控制
SDN
deep reinforcement learning
Q-network
path scheduling
network control
