大型IP网络流量矩阵分析预测的探讨研究
Research on analysis and prediction of traffic matrix for large-scale IP network
韦烜 1刘志华 1李青 2何晓明 1黄君雅1
作者信息
- 1. 中国电信股份有限公司广东研究院,广东广州 510630
- 2. 中国电信股份有限公司研究院,上海 200123
- 折叠
摘要
高效、准确的网际协议(internet protocol,IP)网络流量流向分析预测是网络规划建设的基础.通过部署流量采集分析系统,运营商可轻松获取网络总流量、节点流量、节点分方向流量等较完备的历史基础数据,为流量分析预测提供关键的输入.IP网络流量分析预测方法主要包括两类:传统统计模型和神经网络模型,近年提出的NeuralProphet模型因结合两者优点而得到广泛关注和应用.首次基于NeuralProphet模型对大型运营级IP网络源节点到 目的节点的流量流向进行直接预测,并采用改进的损失函数优化模型训练,预测结果表明Neural-Prophet 模型能够更科学、准确地预测IP网络流量矩阵,整体预测精度提升了 8.7％,同时模型扩展性和鲁棒性也具有更佳的表现,可以更好地满足IP网络规划建设和运行维护的实际需求.
Abstract
Efficient and accurate analysis and prediction of traffic flow direction for Internet protocol(IP)network are the basis of network planning and construction.By deploying a traffic collection and analysis system,operators can easily obtain comprehensive historical data such as network total traffic,node traffic,and node directional traffic,which provides key inputs for traffic analysis and prediction.Methods of traffic analysis and prediction for IP network are generally divided into two categories:traditional statistical model and neural network model.The NeuralProphet model proposed in recent years has been widely applied due to its combination of the advantages of the above models.It is the first time to directly predict the origin-destination traffic flow of large-scale carrier-grade IP network based on the NeuralProphet model,and adopts the improved loss function to optimize model training.The prediction results show that the NeuralProphet model can predict traffic matrix of IP network more scientifically and accurately,and the overall prediction accuracy was improved by 8.7％.Meanwhile,the model has better scalability and robustness,which can better meet the actual needs of IP network planning and maintenance.
关键词
流量矩阵/源节点到/目的节点流量流向/节点流量/预测模型/自回归Key words
traffic matrix/origin-destination traffic flow/node traffic/prediction model/auto-regression引用本文复制引用
基金项目
国家自然科学基金(62076179)
中国电信研究院专业能力级项目(T-2023-12)
出版年
2024
NETL
NSTL
万方数据