基于非负矩阵分解的IP流量预测

为解决宽带多媒体卫星通信系统中的IP流量预测问题，首先使用多用户的IP流量作为训练数据，通过非负矩阵分解迭代方法将其分解为基向量矩阵和编码矩阵，之后再通过ARIMA模型在时间维度上对编码矩阵中的各个行向量进行预测，最后依照预测结果和基向量矩阵合成出各个用户的IP流量预测结果。由于经非负矩阵分解后，编码矩阵中的行向量个数小于用户个数，因此相对于原始的单个用户独立预测方法，新方法可以降低运算的复杂度。仿真实验证实了本方法预测的准确性。

Nonnegative Matrix Factorization-based IP Traffic Prediction

In the face of limited satellite bandwidth resources and users' increasing demands, it becomes a significant issue to realize reasonable and efficient bandwidth allocation among users for broadband multimedia satellite communicalions system. Traffic prediction plays an important role in system resource allocation and management. In the process, IP traffic for multiple users which is regarded as training data was decomposed into basis matrix and coding matrix based on NMF (Nonnegative Matrix Factorization). Then each row vector of encoding matrix was predicted on time dimension on the basis of ARIMA model. At the end, prediction results were combined with basis matrix to generate IP traffic prediction results of each user. After NMF decomposition, the native umber of row vector is fewer than the number of users. As a result, compared with traditional signal user prediction method, the method proposed in this paper can reduce computational complexity. Nests indicate the accuracy of this prediction method.