| Description |
Traffic simulations, which attempt to describe how individual vehicles move on road segments in a network, rely on mathematical traffic flow models developed from empirical vehicle trajectory data (position, speed, acceleration, etc.). Many of these microscopic traffic flow models are described as car-following models, which assume that a driver will respond to the actions of the driver/s or vehicle/s located in front of them (stimulus-response behavior). Model calibration can be performed using regression and/or optimization techniques, but the process is often complicated by uncertainty and variation in human behavior, which can be described as driver heterogeneity. Driver heterogeneity is conceptually based on the idea that different drivers may have different reactions to the same stimuli (interdriver heterogeneity), and an individual driver may react differently to the same type of stimulus (intradriver heterogeneity). To capture interdriver heterogeneity, car-following model parameters must be estimated for each driver/vehicle in the dataset, which are then used to describe a probability distribution associated with those model parameters. Capturing intradriver heterogeneity requires going one step further, calculating those same model parameters over much smaller time periods (i.e., seconds, or fractions of sections) within one vehicle's trajectory. This significantly reduces the amount of data available for calibration, limiting the ability to use traditional calibration procedures. |
