Wavelet analysis-based NPR prediction of short-term traffic flow

Short-term forecasting for traffic flow is one of the key techniques in traffic route guidance and control.It is difficult for classical approaches to effectively forecast the short-term traffic volume with serious uncertainty.According to traffic flow signals showing different characteristics in different scale spaces,a short-term forecasting approach was proposed through combining wavelet analysis with non-parametric estimation.The principle was analyzed and described in detail.Firstly,original traffic flow time series were processed by wavelet multi-resolution decomposition,low frequency and high frequency signals were obtained and reconstructed;secondly,the reconstructed signals were predicted with the non-parametric regression(NPR) model.Finally,the forecasting results by the NPR model are integrated to acquire traffic volume prediction.The experiment results demonstrate that the proposed approach is effective and practical.