摘要
设计了一种基于图神经网络(GNN)和优先经验回放的改进DQN算法(R_DQN)。该算法采用消息传递网络框架进行图结构节点间的信息传播,能够更好地适应网络拓扑这种图结构信息;同时采用优先经验回放机制去学习更有价值的信息,提高样本学习效率,进行算法优化。实验结果表明:该R_DQN算法在网络拓扑和业务模型方面都具有较强的泛化能力,对于训练期间没有见过的网络场景依然有较好的表现,在最大化通信量性能上有较大提升。
An improved DQN algorithm(R_DQN)based on graph neural network(GNN)and priority experience playback is designed.The algorithm uses message passing network framework to spread information between graph structure nodes,which can better adapt to the graph structure information such as network topology.At the same time,priority experience playback mechanism is adopted to learn more valuable information,improve the efficiency of sample learning,and implement the algorithm.optimization.The experimental results show that the R_DQN algorithm has strong generalization ability in network topology and service model,still has good performance for network scenarios that have not been seen during training,and has a great improvement in maximizing traffic performance.
作者
张晓莉
柳珍
郭庆
ZHANG Xiaoli;LIU Zhen;GUO Qing(School of Communication and Information Engineering,Xi’an University of Science and Technology,Xi’an 710000,China;College of Information Science and Technology,Beijing university of Chemical Technology,Beijing 100029,China)
出处
《传感器与微系统》
CSCD
北大核心
2024年第8期150-154,共5页
Transducer and Microsystem Technologies
基金
国家自然科学基金青年科学基金资助项目(61901358)
陕西省教育厅一般专项项目(20JK0757)。
关键词
深度强化学习
软件定义网络
图神经网络
deep reinforcement learning
software-defined network
graph neural network(GNN)
