计算机应用

• 人工智能与仿真 •    下一篇

基于神经网络和自回归模型的网络流量预测

熊皓1,刘嘉勇1,王俊峰2   

  1. 1. 四川大学
    2. 四川大学计算机学院
  • 收稿日期:2020-10-19 修回日期:2021-01-11 发布日期:2021-01-11 出版日期:2021-01-27
  • 通讯作者: 熊皓

Network traffic prediction based on neural network and autoregressive model

  • Received:2020-10-19 Revised:2021-01-11 Online:2021-01-11 Published:2021-01-27

摘要: 互联网的急速发展在给人类带来了巨大便利的同时,也使网络中的网络流量出现了爆炸性的增长,预测网络流量对于网络的研究、管理和控制都具有很高的现实指导意义。为了减小网络流量数据的预测误差,提出了一种基于神经网络和自回归模型的网络流量预测模型——卷积神经网络(CNN)-长短期记忆(LSTM)网络+自回归(AR)。通过卷积神经网络和长短期记忆网络结合来同时获取数据的短期局部依赖特征和长期发展趋势,添加历史连接组件将网络流量的周期性考虑在预测中,完成对网络流量中非线性项的处理,利用自回归模型预测线性项,将两部分结果结合得到最终预测值。实验结果表明,对比传统的网络流量预测模型,在最好情况下所提出模型的均方误差(MSE)、均方根误差(RMSE)和平均绝对误差(MAE)分别减少了1.5604、0.1468和0.1405,这说明该模型有更好的预测表现,预测值与实际值的差距更小。

Abstract: The rapid development of Internet not only brings great convenience to human beings, but also makes the network traffic in the network appear explosive growth. The prediction of network traffic is of great practical guiding significance for the research, management and control of the network. In order to reduce the network traffic data prediction error, a network traffic prediction model based on neural network and autoregressive model — Convolutional Neural Networks (CNN)-Long Short-Term Memory(LSTM) Network+AutoRegressive (AR) was proposed. By combining CNN and LSTM network, the short-term local dependence characteristics and long-term development trend of the data are obtained at the same time, add a historical connection component to consider the periodicity of network traffic into the forecast to complete the processing of non-qualitative items of network traffic, autoregression model is used to predict linear terms, and the final predicted value is obtained by combining the results of the two parts. The experimental result shows that, compared with the traditional network traffic prediction model, in the best case, the Mean Square Error (MSE), the Root Mean Square Error (RMSE) and the Mean Absolute Error (MAE) of the proposed model is reduced by as much as 1.5604、 0.1468 and 0.1405, respectively, which indicates that the model has a better prediction performance and the difference between the predicted value and the actual value is smaller.

中图分类号: