摘要
该文结合强化学习方法提出一种QLCC算法,此算法是将网络拥塞过程进行简化之后描述为马尔科夫决策过程,在Q-learning算法应用的基础上创新设计的新型网络拥塞控制算法。研究过程中首先介绍强化学习方法,并对网络拥塞过程中马尔科夫决策过程的构建条件及假设进行探讨,之后从框架结构、参数结构及定义、参数离散划分和更新步骤几个方面介绍QLCC算法,并采取仿真实验方法对该种新算法的网络吞吐量、公平性、随机丢包环境下的吞吐量分别进行检测,通过与其他3种传统网络拥塞控制算法进行对比分析,证实QLCC算法具有吞吐量较佳、公平性最高、抗丢包性能最优的性能,说明其是一种具有较高应用优势的智能化网络拥塞控制算法。
In this paper,based on reinforcement learning,a QLCC algorithm is proposed,which describes the network congestion process as a Markov decision process and innovatively designs a new network congestion control algorithm based on the application of Q-learning algorithm.In the course of the research,the reinforcement learning method is first introduced,and the construction conditions and assumptions of Markov decision process in the process of network congestion are discussed,and then the QLCC algorithm is introduced from the aspects of frame structure,parameter structure and definition,parameter discrete partition and update steps.However,simulation experiments are used to test the network throughput,fairness and throughput of the new algorithm in random packet loss environment.By comparison and analysis with other three traditional network congestion control algorithms,it is proved that QLCC algorithm has better throughput,highest fairness and best anti-packet loss performance,indicating that it is an intelligent network congestion control algorithm with high application advantages.
出处
《科技创新与应用》
2024年第10期55-58,共4页
Technology Innovation and Application
基金
广州软件学院科研项目(ky202122,ky202123)。
关键词
强化学习
QLCC算法
网络拥塞控制
学习方法
仿真实验
reinforcement learning
QLCC algorithm
network congestion control
learning method
simulation experiment