Multiresolution Analysis and Forecasting of Mobile Communication Traffic

Combined with Fourier spectrum prior knowledge, a novel wavelet multiresolution analysis and forecasting algorithm is proposed. It focuses on long term trend prediction of multi-periodic, non-stationary, mobile communication traffic series. New algorithm calculates the Fourier spectrum for multi-periodic series at first, and takes the prominent period components with definite physical notion as the prior knowledge. After that, it extracts more valuable time domain features with wavelet multiresolution analysis. Finally, it adopts a single model to predict each of them, and integrates the prediction results to gain the final trend prediction of traffic time series. Experimental results on real traffic data show that all isolated components in proposed multiresolution analysis deliver the distinct physical information in traffic data. Additionally, our algorithm can pick out most of prominent period components revealed in the Fourier spectrum and improve prediction accuracy.