Measurement analysis of traffic flow uncertainty on Chinese highway network

Extensive research has been done on traffic forecasting. However, performance of forecasting models is highly influenced by traffic uncertainty and predictability. Traffic uncertainty is important for road users and governors as well. With support of adequate real data from toll stations, we reveal laws in traffic flow uncertainty by employing dispersion coefficient. For further study, Hurst exponent and Approximate Entropy reflect temporal characteristics, indicating long-term randomness and short-term complexity respectively. These measurements all suggest that traffic flow uncertainty drops with the increase of time interval. Our study provides effective measuring methods of uncertainty and theoretical evidence for 15 minutes time horizon in short-term traffic prediction. Daily periodicity exists that highway traffic flow at night is more uncertain than in day time, and off-peak hour flows are more uncertain than peak hour flows. Finally, initial investigation into traffic predictability exhibits acme at 7 a.m. in our case.