基于深度信念网络的网络流量预测模型

Network Traffic Forecasting Based on Deep Belief Network

为了提高网络流量预测精度,根据深度学习理论提出一种基于深度信念网络(DBN)的网络流量预测模型。该预测模型由受限玻尔兹曼机(RBM)单元组成,采用逐层无监督贪心算法训练参数,然后利用反向传播学习算法微调整个网络参数。最后基于该预测模型对收集到的真实网络流量进行预测和分析,并与传统神经网络预测进行对比研究,结果表明,该预测模型克服了传统神经网络容易陷入局部最优、训练时间长及函数拟合度不高等缺点,具有更高的预测精度。

The prediction of network traffic flow is the foundation of network performance and service quality; it plays an important role in network layout and management. To improve the network traffic prediction accuracy, a nonlinear network traffic prediction model based on Deep Belief Network （DBN） is proposed. The model consists of Restricted Bohzmann Machine （RBM） unit which optimizes the layer-by-layer parameters by unsupervised greedy algorithm, and then optimizes all parameters by Back-Propagation algorithm. Simulation results on real network traffic show that the proposed prediction model is more efficient, and has better precision compared with the traditional neural network model.