摘要： 对原始数据进行R/S 分析得到Hurst 系数以选择合适的神经网络结构，重点分析了FIR 的阶及两种不同学习算法(Wan 和Back-Tsio算法)对预测结果的影响。结果表明FIR 阶的选择依赖于流量数据的变化周期，Wan 算法在Hurst 数接近1 的网络流量预测中具有更好的精确性
关键词:
神经网络；FIR；网络流量；预测
Abstract: To predict the Ethernet traffic, R/S analysis is done firstly to get the Hurst coefficient and choose the suitable network architecture, then,prediction results are compared while the order of the FIR filter or the training algorithms (Wan or Back-Tsio algorithms) are varied. The results show that the order of FIR filter depends on the cycle of the traffic and Wan algorithm is more accurate for predicting network traffic whose Hurst coefficient is near 1.
Key words:
Neural network; FIR; Network traffic; Prediction
林雪纲，郑成兴，窦旻，许榕生. 基于 FIR 神经网络的以太网网络流量预测[J]. 计算机工程, 2006, 32(8): 124-126,130.
LIN Xuegang, ZHENG Chengxing, DOU Min, XU Rongsheng. Prediction of Ethernet Traffic by FIR Neural Network[J]. Computer Engineering, 2006, 32(8): 124-126,130.