摘要
利用BBR(Bottleneck bandwidth and RTT)算法虽可以实现在复杂网络中带宽的充分利用,但该算法对网络噪音所造成的丢包现象敏感,且该算法因存在协议内部不公平的问题而无法实现物联数据的实时高效获取。针对以上问题,提出了1种基于深度强化学习的起搏增益优化算法(Deep reinforcement learning of BBR,BBR-DRL)。首先,通过获取数据传输的往返时延、发送窗口大小和网络带宽等环境参数来实时感知网络状态;然后,结合环境参数,利用起搏增益进行动态调整,使得BBR算法能够及时与外部动态网络环境进行交互,从而降低丢包敏感度、提高不同往返时延(Round-trip time,RTT)流之间的公平性。实验结果表明,与经典BBR算法相比,所提出的BBR-DRL算法协议内部的公平性提高到了98.2%,丢包敏感性明显降低。
Although BBR(Bottleneck bandwidth and RTT)algorithm can make full use of bandwidth in the complex network,it is sensitive to packet loss caused by network noise,and the algorithm cannot achieve real-time and efficient access to Internet of Things data due to the unfairness in the protocol.Aiming at the above problems,a pacing gain optimization algorithm based on deep reinforcement learning of BBR(BBR-DRL)is proposed.First of all,the network status is sensed in real time by acquiring the round-trip delay,transmission window size and network bandwidth of data transmission;Then,the BBR algorithm can interact with the external dynamic network environment in time by dynamically adjusting the pacing gain in combination with the environment parameters,thus reducing the packet loss sensitivity and improving the fairness between different round-trip time(RTT)flows.The experimental results show that compared with the classical BBR algorithm,the fairness of the proposed BBR-DRL algorithm is improved to 98.2%,and the packet loss sensitivity is significantly reduced.
作者
刘鹏辉
琚贇
高维星
张彦彦
LIU Penghui;JU Yun;GAO Weixing;ZHANG Yanyan(School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China)
出处
《电力科学与工程》
2023年第4期20-27,共8页
Electric Power Science and Engineering
基金
国家重点研发计划(2020YFB0905900)。
关键词
电力物联网
拥塞控制
深度强化学习
起搏增益优化算法
Power Internet of Things
congestion control
deep reinforcement learning
pacing gain optimization algorithm
