The Transport Research Centre of Rijkswaterstaat is developing a new version of the Dutch National Model for Traffic and Transport (LMS - Landelijk Model Systeem verkeer en vervoer). This new version of the model (LMS 8.0) will be dynamic : it will account for 11 time periods of varying levels of demand. In a preparatory study for LMS 8.0, RAND Europe studied the possible dynamic traffic assignment algorithms suited for the new LMS. In addition, three dynamic models have been tested for their suitability to be the new dynamic heart of the LMS. The conclusion was that, of the three models tested, INDY had the greatest potential for application in the LMS. Additional research was found to be necessary to reduce the run time and memory usage of INDY, which were both found to be quite large (namely 10 hours run time and a memory usage of 696 MB without acceleration option). In the study described in this report, several acceleration options were investigated in more detail by performing 14 new runs. The effectiveness of these options in terms of run time and memory usage reduction was weighed against the effects on quality. The studied acceleration options are : reducing the O/D matrices; simplifying the LMS network; increasing the time step, with and without increasing the minimum link lenght; starting INDY by using a static equilibrium assignment solution as the initial solution and investigating the potential of using accelerated averaging techniques; removing links that are not used, applying different aggregation periods and reducing the maximum number of alternative routes. The effects of these options on run time and memory usage were determined, as well as their effect on the quality of the results of the assignment in terms of total kilometers traveled and cumulative delay. In addition to testing these acceleration options, an estimation of the run time and memory usage of INDY when applied in the context of the New Regional Model (NRM) was given. The NRM networks contain more zones and links than the LMS network. The estimation of this study made use of the 'worst case' NRM network, the Randstad NRM. Finally, based on the findings of this study, recommendations for the use of acceleration options for the application of INDY to the LMS network are given. The main conclusions based on these findings is that the run time of INDY can be decreased from 10 hours without any acceleration option to 30 minutes with the most beneficial acceleration options applied. (Author/publisher)
Abstract