面向IP网络扩容应用的复杂网络流量预测方法

A complex network traffic prediction method for IP network expansion applications

IP网络扩容是通信运营商保持网络平稳运行的一种常见维护方式,核心在于预测未来一段时间的网络流量走势。IP网络流量非常复杂,具有局部不确定性、突发性、异质性等,给预测带来困难。提出了一种针对复杂网络流量的预测方法,它采用编码-解码结构,即在编码层增加全局特征、在解码层增加全局特征和局部特征解析的方式解决局部不确定性;通过增加先验知识缓解突发性;模型采用样本均衡、归一化等方式尽量提取数据的共性,避开数据的异质性。模型整体的参数较少,具有较强的泛化性能;同时采用人工特征和自动特征结合方式保证了浅层网络的准确率。实验结果表明,所提出的方法具有准确率高、泛化性能强的特性。目前该方法已经在工程中大规模应用。

IP network expansion is a common maintenance method for communication operators to keep the network running smoothly.The core is to predict the network traffic trend in the future period.IP network traffic is very complex,with local uncertainty,burstiness,heterogeneity,etc.,which brings difficulties to prediction.A prediction method for complex network traffic was proposed.It adopted an encode-decode structure,adding global features to the encoding layer,and combining global features and local features in the decoding layer to solve local uncertainties.The model used sample balance,normalization and other methods to extract the commonality of the data as much as possible to avoid the heterogeneity of the data.And emergencies were alleviated by increasing prior knowledge.The overall model had fewer parameters and had strong generalization performance,at the same time,the combination of artificial features and automatic features ensured the accuracy of the shallow network.The experimental results show that the proposed method has the characteristics of high accuracy and strong generalization performance.At present,this method has been applied on a large scale in engineering.