Abstract
The availability of turning movement data at intersections is essential in carrying out traffic simulations with software such as NETSIM. However, data collection is very time-consuming even for a small network. To estimate turning movements at intersections, a logit-based stochastic user equilibrium (SUE) model was integrated into a genetic algorithm. Three derivative models were developed: (a) a doubly constrained estimator, (b) a singly constrained estimator, and (c) an origin-destination (O-D)-based estimator. In the first and second estimators, the SUE model was formulated as a joint trip distribution and traffic assignment (TD-TA) program, whereas in the third estimator, the SUE model was based on a standard O-D distribution program. These turning movement estimator models were examined by applying them to three road networks: a virtual road network with simulated data, a small road network in the field with data obtained by manual counting, and a road network with data measured with traffic detectors. Application of the models to the virtual road network proves that if prior information on the O-D distribution is available, the O-D-based estimator is most effective in estimating turning movements. Application of the models to real networks for which no prior information on the O-D flows is available shows that the doubly constrained TD-TA-based model is the most accurate and efficient. The turning movements on major links estimated with this model were in relatively good agreement with those actually measured. The correlation coefficient of link flow exceeded 0.90. However, the rationale of the range of unknown variables remains unresolved.
Get full access to this article
View all access options for this article.