Application of the GARCH Model for Depicting a Driver Acceleration Behavior on Freeways

Vehicle activity studies are now providing detailed second-by-second speed and acceleration data from instrumented vehicle fleets. These driving traces provide a wealth of information about the interaction between the driver and the vehicle, and the vehicle and traffic conditions (when external measurements of traffic flow and speed are available from roadway monitoring systems). In both qualitative and quantitative aspects, the second-bysecond data obtained from the instrumented vehicles enable researchers to examine the accuracy of simulation model algorithms used in predicting car-following behavior and speed acceleration activity of modeled vehicles. In this paper, a large set of data from an instrumented vehicle was used in modeling driver acceleration behavior under typical Atlanta freeway conditions. In this modeling effort, time series analysis techniques are applied to the detailed data. In particular, the GARCH (Generalized Autoregressive Conditional Heteroscedasticity) model, which was originally developed for analyzing financial data that contain time-dependent variance, was applied to depict the sudden speed changes. The fitted GARCH models, which are combined with ARMA (Autoregressive Moving average) model, produced satisfactory results from the residual analysis perspective. Simulation results also suggest that the GARCH model may be acceptable as a convenient tool for depicting acceleration behavior of freeway driving when applied to much larger data sets that are forthcoming in Atlanta from more than 1600 instrumented vehicles. This is because the GARCH model requires only the acceleration series itself. Although the reported modeling effort is limited to one driver and one vehicle’s speed data, the approach adopted in this paper can be fully utilized for other type of acceleration data.