The Network Traffic Prediction Based On Information Theory Learning

With the great development of the internet and its relative applications, the scale of the internet gets to more stronger day by day, these make the network service classes increase and boost the complexity of the internet, but at the same time, the internet happens failure more easy than before, such as the problems of the internet resource optimization allocation and network security become increasingly sharp prominent, the network management more difficulty, all these above problems had a bad effects on network quality of services. So model the internet for network traffic prediction have an positive significance, and many researchers had focus on this issue.The purpose of this study is for boost the precision of network traffic prediction, and its main work is as follows:First:This paper illustrated the Significance of network traffic prediction, showed the domestic and foreign development status about this issue, and introduced the basic theory of Information Theoretic Learning,the features of network traffic and its models based on network traffic prediction, all these mention above made a foreshadowing for the next studying work.Second:According to the features of network traffic, we proposed a new network traffic prediction model combining the Information Theory Learning and the Elman neutral network, that is for the first time we conducted the Maximum Correntropy Criterion (the traditional cost function is Minimum Mean Square Error, MSE) which comes from professor Principe’s Information Theory Learning as a cost function to train the weigh values of neutral network, and made a simulation and analysis for network traffic prediction.Third:Combining Maximum Correntropy Criterion (MCC) with Minimum Mean Square Error (MSE), we also proposed another type of cost function based on mixture mechanism(MCC-MSE) to make a simulation and analysis for network traffic prediction.Fourth:Because of the Correntropy in the Information Theoretic Learning is based on kernel function, and Combining of the MCC and MSE can bring into the weigh coefficient, so we again made a network traffic prediction simulation and analysis for different kernel sizes and weigh coefficients.