Network Traffic Prediction Based on Improved BP Wavelet Neural Network

Abstract

Considering that traditional BP wavelet neural network (BPWNN) is easy to take local convergence and has slowly learning convergent velocity. We apply a method based on adaptive learning rate to optimize it in accelerating the learning convergent velocity. In prediction, firstly, denoised the traffic time series with wavelet packet transform to improve the prediction precision, then compared the ability of BP neural network (BPNN) and improved BPWNN (IBPWNN) to the prediction of network traffic. The emulation experiment results indicate that in the case of one-step prediction, BPNN and IBPWNN have similar prediction precision, however, in the case of multi-step prediction; the BPNN has low prediction precision, while the IBPWNN still performs a good ability to prediction.