基于集成深度学习的网络中短期流量精准预测研究

Research on Accurate Prediction of Network Short-and Medium-Term Traffic Based on Integrated Deep Learning

网络中短期流量特征不同,依靠单一的方法预测流量,造成预测结果均方根误差较大。因此,提出了基于集成深度学习的网络中短期流量精准预测方法。运用BDS统计量检验方法建立非线性判断机制,提取网络流量内包含的非线性特征。依据小波变换理念分解和重构复杂的网络流量,引入集成深入学习理念,在Stacking集成策略的作用下,建立包含循环神经网络和卷积神经网络的集成流量预测模型。从网络结构和修剪过程两方面入手,优化预测模型结构,并通过正交最小二乘法求解网络中短期流量预测值。仿真测试结果表明:所用方法的网络中短期流量预测均方根误差为0.12,实现了网络中短期流量精准预测。

The short-and medium-term traffic in the network shows different characteristics that may result in great root-mean-square errors if predicted with a single method. Therefore, an accurate prediction method of network short-and medium-term traffic is proposed based on integrated deep learning. The nonlinear judgment mechanism is established by using BDS statistical test method to extract the nonlinear characteristics contained in the network traffic. The complex network traffic is decomposed and reconstructed under the idea of wavelet transform. With the introduction of the integrated deep learning, an integrated traffic prediction model containing cyclic neural network and convolutional neural network is established on the effect of stacking integration strategy. From the network structure and pruning process, the structure of prediction model is optimized, and the short-and medium-term traffic prediction value of network is solved by orthogonal least square method. The simulation test results show that the root-mean-square error of the proposed method is 0.12,which realizes the accurate prediction of short-and medium-term network traffic.