基于改进BP神经网络的网络流量预测  

Network Traffic Prediction Based on Improved BP Neural Network

作　　者：周华乔 祝宏亮 孙一凡 苏红艳 王康伟 倪敬一 ZHOU Huaqiao;ZHU Hongliang;SUN Yifan;SU Hongyan;WANG Kangwei;NI Jingyi(Jiangsu Industrial Cloud Edge Collaborative Technology Engineering Research Center,Huai’an Jiangsu 223001,China;Huaiyin Institute of Technology,Huai’an Jiangsu 223001,China)

机构地区：[1]江苏省工业云边协同技术工程研究中心,江苏淮安223001 [2]淮阴工学院,江苏淮安223001

出　　处：《通信技术》2024年第10期1059-1065,共7页Communications Technology

基　　金：国家自然科学基金(62001515);淮安市自然科学研究项目(HABL202215,HABZ202223)。

摘　　要：传统网络流量预测方法常面临精度不足、时间复杂度高的问题,且在处理复杂的网络流量时,难以达到理想的预测效果。为克服这些挑战,采用深度学习中的反向传播(Back Propagation,BP)神经网络算法,并结合改进的梯度下降法,通过引入动量项与变步长法相结合的方式,有效降低误差值,实现了对网络流量的准确预测。该方法旨在更好地满足用户需求,优化网络性能,并提升网络服务质量。实验结果表明,改进后的BP神经网络算法在网络流量预测中展现出良好的可行性与稳健性,同时达到了较高的预测精度。Conventional network traffic prediction methods often face the challenges of insufficient accuracy,high time complexity,and difficulty in achieving desired prediction effects when dealing with complex network traffic.To overcome these challenges,this paper adopts the BP(Back Propagation)neural network algorithm from deep learning and combines it with an improved gradient descent method,which effectively reduces the error value and achieves the accurate prediction of network traffic by introducing the momentum term in combination with the variable step size method.The method aims to better meet user needs,optimize network performance,and enhance network service quality.Experimental results demonstrate that the improved BP neural network algorithm exhibits good feasibility and robustness in network traffic prediction,while achieving high prediction accuracy.

关 键 词：网络流量预测 反向传播神经网络算法 梯度下降法 动量项与变步长 预测精度 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]