改进微分进化和小波神经网络的网络流量预测  

Network traffic prediction based on wavelet neural network with improved differential evolution algorithm

作　　者：林振荣[1] 黎嘉诚 杨冬芹[2] 伍军云[1] LIN Zhen-rong;LI Jia-cheng;YANG Dong-qin;WU Jun-yun(Information Engineering School,Nanchang University,Nanchang 330031,China;Electronic Information Engineering School,Jiangxi Industry Polytechnic College,Nanchang 330031,China)

机构地区：[1]南昌大学信息工程学院,江西南昌330031 [2]江西工业职业技术学院电子信息工程分院,江西南昌330031

出　　处：《计算机工程与设计》2019年第12期3413-3418,共6页Computer Engineering and Design

基　　金：江西省科技支撑计划重点基金项目(20151BBE50057)

摘　　要：针对网络流量的非线性和复杂性等特性以及传统网络流量预测模型精准度低的缺点,提出自适应微分进化算法(ADE)优化小波神经网络(WNN)的短期网络流量预测方法。以小波神经网络为基础,在神经网络训练过程中增加动量项,采用自适应微分进化算法优化小波神经网络原有的初始化参数的过程,有效解决小波神经网络中传统梯度下降算法易陷入局部极小解和对初始值敏感的缺陷,提高学习精度和收敛速度。仿真结果表明,相比对比模型,该方法具有良好的准确性、收敛性以及稳定性,是一种有效可靠的短期网络流量预测方法。Concerning the nonlinearity and complexity of network traffic and the defect of low precision of traditional network traffic prediction model,a short-term network traffic prediction method was proposed based on wavelet neural network(WNN)optimized by adaptive differential evolution algorithm(ADE).Based on the wavelet neural network,the momentum term was added to the training process and the adaptive differential evolution algorithm was used to optimize the original initialization process of the neural network parameters,which improved learning accuracy and convergence speed by effectively solving the defects that the traditional gradient descent algorithm in the wavelet neural network is easy to get the minimum solutions and it is sensitive to initial parameters values.The simulation results show that the proposed method is an effective and reliable shortterm network traffic prediction method with good accuracy,convergence and stability compared with the comparison models.

关 键 词：小波神经网络 微分进化算法 自适应 短期网络流量 预测计算 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]