For traffic flow models, calibration and validation are essential. Cellular automaton (CA) models are a special class of models, describing the movement of vehicles in discretised space and time. However, the previous work on calibration and validation does not discuss CA models systematically. This study calibrates and validates a stochastic CA model. The authors use a simple CA model, which only has two important parameters to be calibrated. The methodology for optimisation is to minimise the relative root mean square error between two properties: The averaged velocity and the variation of velocities in a platoon at a given density. Three different sites are used as cases to show the methodology, for which different types of data (video trajectories or GPS data) are available. The authors find that the best model parameters vary for the different locations. This may result from various driving strategies and potential tendencies. Thus, it is concluded that for CA models, various traffic flow phenomena need to be simulated by various parameters.@en

In this paper, we discuss the mechanisms for discretionary lane-changing behavior in traffic flow. NGSIM video data are used to check the validity of different lane-changing rules, and 373 lane changes at 4 locations in US-101 highway are analyzed. We find that the classical lane-changing rules of rule-based model cannot explain many cases in the empirical dataset. Therefore, we propose one new decision rule, comparing the position after a time horizon of several seconds without a lane-change. This rule can be described as “to have a further position within 9 seconds”. The tests on NGSIM data show that this rule can explain most (76%) of the lane-changing cases. Besides, some data when lane changes do not occur are also studied. We find that most (81%) of non-lane-changing vehicles do not fulfill the new rule. Thus, it can be considered as one sufficient and necessary condition for discretionary lane-changing.

@en