The broad objectives of the project reported here are as follows: (i) to test whether our expectations of the future development of congestion and the effectiveness of policy measures depend on the assumptions within the National Model System (NMS) regarding assignment technique; and (ii) to investigate how much these expectations change under varying assignment assumptions. The practical implementation of the project is through a theoretical and practical comparison of assignments with different dynamic assignment techniques, with reference to a standard application of the current assignment model within the NMS: QBLOK. Two representative dynamic assignment techniques have been compared: 3DAS and CONTRAM. To avoid spurious effects in this comparison, the two dynamic models have been implemented in such a way that the fit with the original QBLOK model and with observed reality is as close as possible. To this end the speed-flow-density relationships used in CONTRAM and 3DAS have been estimated to fit close to those used in QBLOK. All 3 models are compared with observations (validated) for a 1994 base year. The policy measures investigated in the scenario comparisons are those defined for SWAB (SamenWerken Aan Bereikbaarheid). The comparisons concentrate on 3 scenarios: 1990 base year (note the difference with the 1994 validation year), 2000 reference and 2000 with SWAB measures. All comparison reported here must be seen in the context of the limitations in the application: for the NMS, with its associated network and demand representations; without feedback to the NMS demand module; with link performance functions estimated to ensure as high a level of consistency between the 3 assignment representations. For the 1994 validation base year "observed" data has been collected and analysed with respect to: (i) total level of congestion, distinguishing Randstad/other and main/secondary road network; (ii) congestion locations; (iii) build-up of congestion over time; (iv) composition of affected traffic (in terms of purposes); and (v) kilometrage travelled on main and secondary road network. The fit of the two dynamic models to both their static originator and to observable reality is reasonably good, particularly so at a more aggregate level (in time and geographically). All three models are disappointing in their ability to reproduce the observed Queue Top 20, but the observation bias in police congestion reporting will certainly play a role here. In general the dynamic models estimate a lower level of congestion in the Randstad than QBLOK. The fit of the dynamic models with observed queuing profiles at notorious hotpots is reasonable. The scenario runs shed light on the impacts of different assignment techniques on assignment outcomes and network performance (as proxy for the overall impact on policy decisions, given the lack of feedback to demand). An important observation is that the absolute network-aggregated results for the 1990 base year (particularly travel and queued time) differ considerably between the three packages, whilst also differences with the 1994 validation year are substantial. The overall impact of 10 years' traffic growth on network-wide travel times and distances is very similar between the packages, although both dynamic models estimate a considerably higher growth in queued time in the PM peak than their QBLOK equivalent, whilst in the AM peak this value is slightly lower than QBLOK. An important result in terms of policy implications is that CONTRAM forecasts considerably less effectiveness of the SWAB measures than both other models. A second important consideration is that, in terms of geographical spread of congestion, QBLOK estimates a greater proportion of congestion to occur in the Randstad and on the main road network than do the two dynamic models; surprisingly, all 3 models forecast the effectiveness of SWAB measures to be greatest outside the congested Randstad area. This may influence the prioritisation of congestion reducing measures. The report concludes with a number of recommendations: (a) within the constraint placed on the comparison, related to consistency in speed-flow-density curves and ignoring demand feedbacks, the case for replacing the existing QBLOK assignment capability with a dynamic technique has in our view not been established, so that we recommend continued use of QBLOK in the near future, for practical reasons related to data requirements and run times; (b) on the supply side representation, the road network, there are clear indications that in the NMS the current link characteristics, and particularly the outflow capacities, have not been established with a sufficient degree of accuracy. We expect that further fine-tuning of the network parameters (and possibly also network configuration) would allow significant improvements to be made; (c) enhancements to the assignment techniques should most likely be driven by the ability to represent measures and policies. For example, provision of information can currently only be dealt with in an ad-hoc manner. It would be worthwhile identifying the required directions of QBLOK enhancements in this context; (d) if we accept that in return for improved model performance extended run times are acceptable, but that the extra investment in computational terms of dynamics is not warranted, we may also consider other possibly valuable assignment model extensions. One direction is the representation in more detail of junctions, particularly in urban areas; (e) on the demand side, the OD matrices, the results obtained suggests that the used dynamic base matrix (which has been derived from the static 1994 interim base matrix using a simple conversion mechanism, typical of use in practice) is not of sufficient quality to obtain the expected benefits of dynamic assignment. This points strongly towards the need for research into improved methods to estimate dynamic OD matrices (ODT). We believe that the benefits that are expected from dynamic assignment can only be properly realised if and when the ODT matrix can be estimated with much improved accuracy. It is hoped that with the arrival of real ODT-measurements, such as obtained for instance from Floating Car Data, major improvements can be made; and (f) finally, the representation of demand/supply interaction within the assignment procedure, the speed-flow-density modelling, would benefit from further research. With the sorts and amounts of data which are now available from MONICA and other dynamic measurement and monitoring systems, it would be possible to calibrate improved empirical relationships for typical NMS application contexts, which we hope would lead to more accurate identification of bottlenecks. Link performance functions, rather than assignment technique, may well have the greatest impacts on policy decisions made with the NMS. (A)
Abstract