Nonlinear forecasting of daily traffic flow based on optimal embedding phase-space

Traffic flow prediction is an important application in ITS. This paper presents a new method to build a nonlinear forecasting model for daily traffic flow prediction. The method consists of three steps. First, a statistic is offered to determine whether a linear model or a nonlinear model is suitable for a given time series. Second, if a nonlinear model is suitable, then a new algorithm is approved to synchronously select the optimal embedding dimension and delay step of the time seriespsila constructed phase-space. Last, a local linear forecasting model based on the optimal embedding phase-space is build. The real daily traffic flow data are applied to test the new method.