The paper discuss the use of different distributions of the stochastic component in SUE. A main conclusion is that they generally gave reasonable similar results, except for the LogNormal distribution which use is dissuaded. However, in cases with low link-costs (e.g. in dense urban areas, ramps and modelling of intersections and inter-changes), distributions with long tails (Gumbel and Normal) gave biased results com-pared with the Rectangular distribution. The Triangular distribution gave results somewhat between. Besides giving the most reasonable results, the Rectangular dis-tribution is the most calculation effective.All distributions gave a unique solution at link level after a sufficient large number of iterations (up to 1,000 at full-scale networks) while the usual aggregated measures of convergence converged quite fast (under 50 iterations). The tests also showed, that the distributions must be truncated symmetrical at fixed values (zero and twice mean), while truncation only at zero or any type of redraw bias the results. It is also noted that the variance should be related to the mean link costs, while a relation to the square mean bias the results systematically towards long links. In addition, it is very impor-tant to use a proper random number generator.Compared with the ‘standard version’ of SUE, better results were experienced by us-ing link dependent speed-flow curves. This removed a bias towards smaller roads. An even larger improvement can be obtained by including intersection delays in SUE. In addition, the simulation of differences in road users preferences can improve the re-sults significantly.
|Title of host publication||Seminar F, Transportation Planning Methods|
|Publication status||Published - 1997|
|Event||25th European Transport Forum (PTRC Annual Meeting) - Uxbridge, UK|
Duration: 1 Jan 1997 → …
|Conference||25th European Transport Forum (PTRC Annual Meeting)|
|Period||01/01/1997 → …|