A Revised Logit Model for Stochastic Traffic Assignment With a Relatively Stable Dispersion Parameter

The optimal value of the embedded dispersion parameter in the logit-based stochastic traffic assignment (STA) model significantly impacts predicted network flows. It also varies dramatically for different systems. This prevents its application, especially when real-world network flow data are unavailable for model calibration. To address the problem, this article proposes a revised logit model. Rather than the route travel time in the logit model, it characterizes route utilities by using the relative route travel time defined as the quotient of the estimated route travel time with respect to the mean route travel time for routes between corresponding origin–destination (OD) pairs. The model captures the impacts that the heterogeneous mean route travel time for OD pairs has on travelers’ perception errors. Thus, the dispersion parameter in the revised model is much more stable than that in the original one.

To reduce the revised model’s computation time, this article also proposes a node-based traffic assignment method that can simultaneously load the demand from all origins to one destination into the network. Calibrated results based on large-scale taxi GPS trajectory data in Nanjing and Beijing, China, show that the optimal values of the dispersion parameters in the revised logit-based STA model for the two networks are very close. The calibrated model can replicate observed traffic flows with high accuracy. Hence, the proposed approach can better assist traffic managers in predicting network flows, particularly when real-world data are not available.