摘要
网络时延的动态变化反映了网络路径的负载特征,对时延的精确预测是实施网络拥塞控制、路由选择的重要依据,建立了基于随机神经网络的时延预测模型,该模型克服了传统时间序列预测方法受随机干扰因素影响大、模型结构辨识过程繁琐,以及传统神经网络预测方法易于陷入局部极值、偏离全局最优的缺点。仿真实验表明,在提前单步和多步的预测中该模型比AR模型、RBF神经网络预测算法的准确度更高。
Network delay is an important performance metric of IP network which reflects network path＇s workload characteristics. The precise prediction for network delay is an important basis on congestion control and route selection. A new multi-step prediction method was proposed for network delay prediction based on the random neural networks, this method overcomes the disadvantage traditional time-series method and neural network method. Compared with traditional RBF network and AR model, the experimental result indicated that the proposed model has better accuracy for single steps and multi steps prediction.
出处
《计算机科学》
CSCD
北大核心
2009年第7期85-87,112,共4页
Computer Science
基金
国家自然科学基金资助项目(90204010)资助
关键词
网络时延
RNN神经网络
预测
Network delay, Random neural network, Prediction
