天地一体化信息网络 ›› 2020, Vol. 1 ›› Issue (2): 57-65.doi: 10.11959/j.issn.2096-8930.20200208

所属专题: 专题:天地一体化信息网络体系架构

• 专题:天地一体化信息网络体系架构 • 上一篇    下一篇

基于改进LSTM算法的天地一体化信息网络流量预测

潘成胜1,2, 王羽夫1, 杨力1   

  1. 1 大连大学通信与网络重点实验室,辽宁 大连 116622
    2 南京信息工程大学,江苏 南京 210044
  • 修回日期:2020-11-10 出版日期:2020-12-20 发布日期:2020-12-01
  • 作者简介:潘成胜(1962-),男,大连大学通信与网络重点实验室教授,南京信息工程大学电子信息工程学院教授,主要研究方向为一体化网络系统与网络协议、一体化指挥系统的网络理论与技术。
    王羽夫(1996-),男,大连大学通信与网络重点实验室研究生,主要研究方向为一体化智能网络流量预测技术。
    杨力(1982-),女,大连大学通信与网络重点实验室负责人,主要研究方向为空间信息网络传输技术、无线通信网络协议理论与方法。

Traffic Prediction of Space-Integrated-Ground Information Network Based on Improved LSTM Algorithm

Chengsheng PAN1,2, Yufu WANG1, Li YANG1   

  1. 1 Communication and Network Laboratory, Dalian University, Dalian 116622, China
    2 Nanjing University of Information Science and Technology, Nanjing 210044, China
  • Revised:2020-11-10 Online:2020-12-20 Published:2020-12-01

摘要:

天地一体化信息网络由于存在流量突发性强、拓扑时变等问题使得通信易产生中断,流量波动不平稳导致其流量预测难度远高于地面网络。针对该问题,提出一种改进的LSTM算法,首先分析流量序列滞后变量对预测值的影响,判断流量自相关度;其次,采用以预测值代替中断的方式,消除训练集的噪声和断点;最后,使用Dropout算法减少了噪声和神经网络过拟合带来的影响,准确预测天地一体化信息网络流量数据。仿真实验表明,在OPNET仿真环境中,该算法相较于其他算法准确性提升了59.21%,算法训练速度提升了11.11%,能够为天地一体化信息网络统筹调度提供有效的数据支撑。

关键词: 天地一体化信息网络, 网络流量预测, 深度学习, 流量自相关性

Abstract:

The space-integrated-ground information network is easy to interrupt and the traffi c fl uctuation is not stable due to the problems of high traffi c burst and topological time-varying, which makes the traffi c prediction diffi cult much higher than the ground.In order to solve this problem, an improved LSTM algorithm was put forward.Firstly, the traffi c autocorrelation was judged by analyzd the infl uence of the lag variable of traffi c sequence on the predicted value; Secondly, the noise and breakpoint of the training set were eliminated by replacing the interruption with the predicted value; Finally, Dropout algorithm was used to reduce the impact of noise and neural network over fi tting, and accurately predict the traffi c data of the integrated intelligent network.The simulation results showed that in OPNET simulation environment, compared with other algorithms, the accuracy of this algorithm was improved by 59.21%, and the training speed of the algorithm was improved by 11.11%, which could provide eff ective data support for the overall scheduling of the integrated intelligent network.

Key words: space-integrated-ground information network, traffi c prediction of network, deep learning, traffi c autocorrelation

中图分类号: 

No Suggested Reading articles found!