To meet local market requirements, the software provides two sets of convergence criteria for assignment with ICA.
With the classic variant, both results of successive iterations as well as values within an iteration are compared. As one goal of the procedure is to obtain a match between the results of subordinate assignment and the results after congestion and ICA calculation, this degree of match is used as a factor to determine convergence. The GEH value is used to compare volumes and relative deviations are calculated to compare waiting times. The respective convergence condition is considered fulfilled if the share of network objects defined in the convergence conditions reaches or falls below the GEH value or the minimum value for relative deviation. To calculate the share of turns with the classic variant, Visum uses only open turns at nodes with ICA calculation. To calculate the share of links, it takes all open links into account. Merely when comparing the congestion of successive iterations (convergence criterion: Maximum value for the mean change in queue lengths on links with congestion) Visum only considers links that are congested.
The convergence criteria used in the WebTAG-compliant variant were defined in accordance with the WebTAG guideline ("TAG unit M3-1 Highway Assignment Modeling", Department for Transport 2014). With this variant, you can define a maximum gap (corresponding to the %GAP in WebTAG) as a convergence criterion. This gap is calculated at the end of each iteration, i.e. after spillback and ICA calculation, in the same way as other PrT assignments (Convergence criteria of assignment quality). To compare the results of successive iterations, you need to define a relative deviation value for each convergence criterion. The share of objects is calculated based on links or turns in the network that are open. The assignment is considered converged if the maximum gap is reached, the share of objects for volume or cost changes is reached or exceeded and these criteria are fulfilled in the specified number of consecutive external iterations.
Volume and cost changes between successive iterations are considered a sign of assignment stability. In networks with high volumes, the volumes may even out, but the costs will continue to fluctuate. By contrast, using relatively flat VD functions can stabilize the costs, while the volumes will continue to largely fluctuate. These dependencies are taken into account with the OR condition used between the criteria for volumes and costs.
