抄録
This paper presents models to predict freeway link travel time in multiple periods. The relationships between traffic measures and average link travel time from simulation were explored using the Generalized Additive Model (GAM), a nonparametric regression technique. Parametric models to predict freeway link speed were then developed based on the relationships discovered by the nonparametric model. The models were developed on a set of simulation data and tested on another set of simulation data. The root mean square errors (RMSE) of the model developed to predict link travel time at current time-step were 88.63, 6.88, 2.38 seconds for morning peak hours, off-peak hours, and afternoon hours, respectively. The RMSE of the model to predict link travel time at 5, 10, 15 minutes forwards were 91.20, 101.86, 122.08 seconds, respectively.
