基于RBF神经网络的自相似业务流预测研究  

Research on Prediction of Self-similar Traffic with RBF Neural Network

作　　者：欧家成[1] 吴援明[1] 

机构地区：[1]电子科技大学光电信息学院,成都610054

出　　处：《微处理机》2010年第3期80-82,86,共4页Microprocessors

基　　金：国家自然科学基金资助项目(60876051)

摘　　要：根据自相似业务流量非线性和非平稳的特点,采用极大重叠离散小波变换(MODWT)对网络流量进行预处理,简化了数据拟合过程。结合一个结构简单的RBF神经网络构建预测模型,对Bellcore实验室的原始数据进行预测仿真,通过与AR、FARIMA和BP等预测模型在不同时间尺度下的一步预测结果比较,提出的预测方法在信噪比(SNR)上提高了1-2个dB,而且在多步预测中也有更好的优越性。According to the nonlinear and nonstationary characteristics of self- similar traffic, the process of data fitting is simplified by processing the network traffic with maximal overlap discrete wavelet transform at first, then a prediction model is built combined with a RBF neural network of a simple structure. Forecasting experiments for actual traffic data from Bellcore Laboratory are carried out. Our prediction method increases 1 to 2 dB in SNR by comparing the one - step prediction results at different time scales with other models such as AR, FARIMA and BP, and also has advantages in mult - step prediction.

关 键 词：自相似 流量预测 小波变换 RBF神经网络 

分 类 号：TN915.07[电子电信—通信与信息系统]