Combination Of Network Traffic Prediction Model System

With the increasing popularity of computers, the size of the network is becoming larger and larger, network management is also becoming more and more important.For the current and future network running status have a better monitor, network traffic analysis and modeling prediction is very necessary, and one of the research hotspots.In order to meet the development of network management, first this paper design and implement a combined model of network traffic prediction system, in this system there are the main process module, data acquisition module and the traffic prediction module. This paper have a thoroughly study on traffic prediction module.For the traffic prediction module, this paper presents a prediction based on wavelet transform and neural network models. First, this paper use the mutli-layer decomposition of the wavelet transform on flow sequceces,after the wavelet decomposition we can get approximate signal sequence and details signal sequence. For approximate signal, we use improved BP neural networks Prediction. For detail signal, we use ARIMA Prediction.For the improved BP neural network, first the number of neurons in the input layer and the number of neurons in the output layer neuron in the BP neural network is not a reasonable choice, the paper presents a new way that determine the number of neurons of the neural network's input and output layer according to relevance theory. This approach completely based on the sample and the sample correlation between the most closely to determine, this approach makes the flow of samples to learn more targeted and scientific.After that BP algorithm is further improvement in this papar, joined the dynamic learning rate, according to the size of the error dynamically change the rate of learning, the error can be very fast convergence.However, the convergence rate too fast can cause system instability, so to improve by adding momentum into the second item, according to the last moment the direction of weight change the rate of learning.Finally for the within of the neurons, propose a method of dynamic bias, according to the size of the bias error dynamic adjustment.For the detail signal's ARIMA(p,d,q) prediction, as the detail signal of single reconstruction has a certain regularity, the needed sequence can be extracted according to the laws, then analysis and Modeling them. This can significantly reduce the computation complexityFinally, the real network traffic data is uesed to fit the model, the results show that the model has higher predictive.