Safwat and magnanti have developed a combined trip generation, trip distribution, modal split, and traffic assignment model that canpredict demand and performance levels on large-scale transportationnetworks simultaneously--that is, a simultaneous transportation equilibrium model (stem). Safwat and brademeyer have developed a globally convergent algorithm for predicting equilibrium on the stem. The objective of this paper is to investigate the relative computationalefficiency of the algorithm as a function of demand, performance, and network parameters for two small, sample networks and one large-scale, real-world network. The algorithm was found indeed to be sensitive to the values of several variables and constants of the model. Many of the results were as expected and could be generalized. As the values of demand parameters increase, the algorithm tends to take more iterations, on the average, to arrive at a given accuracy level. Beyond maximum "practically feasible" values, however, the algorithm may require a considerable computational effort to satisfy a given tight level of accuracy. Network configuration may have a considerably greater influence on convergence rate than network size. These results should further encourage application of the stem approach tolarge-scale urban transportation studies. This paper appears in transportation research record no. 1251, Transport supply analysis.
Abstract