Abstract: To better solve the network traffic prediction problems, according to that the fraction function approximation nature and fitting ability in function approximation are much larger than linear function, and the process neural networks have the ability of non-linear transformation to time-varying function, a fraction process neural network model and its learning algorithm are proposed. The experimental result shows that the network model has flexibility approximation properties for singular value process function and sensitivity reactions near the area in the singular value better than the general process neural network. The model can be trained and be used to forecast flow using network measured data, and achieve good application effect.

Key words: structural formula process neural network, learning algorithm, network traffic, prediction

摘要： 为更好解决网络流量预测问题，依据函数逼近论中分式的函数逼近性质和拟合能力要远远大于线性函数的性质，以及过程神经元网络对时变函数的非线性变换能力，提出一种分式过程神经元网络模型及其学习算法。实验结果证明，该网络模型对具有奇异值过程函数的柔韧逼近性质和在奇异值点附近区域反应的灵敏性优于一般过程神经元网络，以网络实测数据对模型进行训练和流量预测，取得了较好的应用效果。

关键词: 分式过程神经元网络, 学习算法, 网络流量, 预测