Nonlinear Characteristics of Short - term Traffic Flow and Their Influences to Forecasting

To improve the forecasting accuracy and reliability of short-term traffic flow, the influence of length of historical data was reevaluated from the viewpoints of identification forecasting. Correlation dimensions and recurrence plots were calculated to analyze a freeway traffic flow. It is found that the traffic flow is chaotic and fractal under a large range of observation time windows. Correlation dimensions decrease with the increasing of time windows in minute scales. Different from the formerly deduction that fractal characteristics would vanish with the decreasing of time windows, the numerical estimation results show that correlation dimensions increase with the increasing of time windows in second scales. It is also suggested that a reliable forecasting requires that the length of historical data should be more than ten times of correlation dimensions of a short-term traffic flow.