【作者】 许辉；

【导师】 谭敏生；

【作者基本信息】 南华大学 ， 计算机应用技术， 2010， 硕士

【摘要】 随着Internet及其应用的迅速发展,网络开始承载越来越多的应用服务,网络行为特征日趋复杂,这给网络规划、网络管理以及服务质量带来了越来越大挑战。IP网络流量建模及预测是带宽分配、流量工程、性能分析、路由控制及差错控制的基础和主要参考依据。本文首先分析了IP网络流量的主要特性,分析比较了几种传统的IP网络流量预测模型的优缺点,在此基础上提出了基于残差改进的灰色预测模型,该模型通过对灰色模型的残差序列做指数化处理,从而使得正负交替序列向非负序列转化。实验结果表明,改进后的模型具有较高的预测精度。针对常权组合模型的权值恒定不变,很难精确预测现实网络流量的问题,本文提出了模糊自适应的变权组合预测模型,该模型由改进后的残差灰色模型和BP神经网络模型组成,引入模糊决策机制和自适应机制,通过对数据的处理得到组合模型中单一模型的模糊权值和基本权值,然后计算得出该单一模型在组合模型中的权值。实验结果表明,模糊自适应的变权组合预测模型与常权组合预测模型相比,性能更优。但实验同时也发现,在小时间粒度网络流量预测中,当预测步长超过7步以后,预测误差超过20%。针对此问题,本文提出了动态变权组合预测模型,即在原模型的基础上引入动态机制。当预测误差超过设定阀值时,变权组合预测模型也随之进行重构,从而减少预测误差。实验结果表明,动态变权组合预测模型比原模型在预测步长上有所增长。更多还原

【Abstract】 As the rapid development and application of Internet, the scale of internet is becoming larger and larger, the features of network behavior are becoming complicated increasingly, which brings a great challenge to network planning、network management and quality of service. Predicting and modeling network traffic can bring out essential reference for bandwidth allocation, network traffic control, routing control, entry control and error control in network management.In the beginning, the article analyze the main character about network, then analyze and compare the advantage and disadvantage of some traditional network analytic model. After these analysis, a kind of grey predicting method based on error was given, which greatly improves the original residual error model via index processing on error sequence. And this new model was proved higher prediction accuracy by the experimental results. In the following, aiming at the shortages of combination Model based on Constant weight, a new model of combined forecasting method based on Fuzzy Adaptive variable weight was brought out, which uses fuzzy decision-making mechanism and the adaptive mechanism to gain the single model’s fuzzy weights and the basic weight, Experimental results show that the model of combined forecasting method based on Fuzzy Adaptive variable weight is better than the combination Model based on Constant weight in the performance .But at the same time, when forecasting more than 7-step in the small time granularity network traffic prediction,, the prediction error will more than 20%, and have become a big trend. On the basis of the original model by adding dynamic mechanism, when the prediction error reaches a certain characteristic value, the variable weight combination model also will be reconstructed to adjust, thereby reducing the prediction error. Experimental results show that the dynamic variable weight combination forecasting model is better than the original model on the step size increased.更多还原