Statistical Model for Forecasting Link Travel Time Variability

Abstract

In the field of advanced traveler information systems, travel time reliability contributes significantly to the utility of traffic information affecting the traveler’s choice. The exact estimation of the variance in travel times is fundamental to calculating reliability indices. A method for predicting the dynamic variance in estimated link travel times is described. The dynamic variance is allowed to vary dependent on variances for previous time periods, which is typically ignored in conventional time-series analysis. We adopt the autoregressive moving average-generalized autoregressive conditional heteroscedasticity (ARMA-GARCH) model in which the ARMA model and the GARCH model are combined. In parallel, the generalized Pareto distribution (GPD) is employed in the computation of percentile to overcome the asymmetry in travel time distribution. The autocorrelation of dynamic variance is identified in links located in urban congested areas. The use of the ARMA-GARCH model yielded statistically significant outcomes in estimating dynamic variances in travel times. In particular, for a link with higher level of congestion, the ARMA-GARCH model along with GPD has been proven to be more promising.