CORDIS Archive
This page has been archived. It will no longer be updated.
 |
  | ||||||||||||||||||||||||||||||
                   |
AUITO- FINAL SUMMARY REPORT
The AIUTO project addresses Task 5.1.2/5 in the Urban Transport sub-programme of the EU 4th RTD&D Framework Transport Programme (Transport Management / Technique Tools - Predictive Modelling).
and it involved 6 test-sites throughout Europe: Como (IT), Salerno (IT), Randstad (NL), York (UK), Thessaloniki (GR) and Geneva (CH). The main objectives of AIUTO were: Secondary objectives were also present, as follows: AIUTO had an application-oriented approach as it intended to develop common guidelines and methodologies for TDM policies and innovative systems implementation integrating macroscopic and microscopic approaches at a European level, starting from a review of existing models and methodologies. The focus points or strong statements of the AIUTO project are the following:
| ||||||||||||||||||||||||||||||
| Effectiveness |
| ||||||||||||||||||||||||||||||
| Efficiency |
| ||||||||||||||||||||||||||||||
| Quality |
| ||||||||||||||||||||||||||||||
| Equity |
|
Review of demand modelling approaches
Demand models may be classified according to different criteria. Some main streams for demand models can be identified as follows:
1. demand models can follow a "switching" approach (where, given a previous demand pattern, TDM measure impacts are predicted as "differences", which are "conditioned" by existing demand) or a "holding" or "synthetic" approach (where the model is intended to recalculate the demand pattern for the scenario that includes the policy measures);Regarding switching vs. holding approaches, the AIUTO experience has found that neither one of the approaches has shown an evident supremacy. In general, switching models (as the ones used in the Como test site) have been proved to have a generally easier analytical structure and to be less data intensive. In fact, only attributes related to changing vs. non-changing alternatives have to be estimated. On the other hand, at least in the particular approach used in Como, where switching has been applied only to car trips, the contribution of changes in other-modes is missed.
The "all-day" and "peak-hour" approaches have been used within the AIUTO project. A non-negligible evidence has been shown that, for general cases, the all-day approach is recommended when TDM measures have to be assessed and evaluated. This is due to two factors. On one hand, most of the TDM measures could be intrinsically time-of-day-dependent; on the other hand, the effects of (even all-day-constant) TDM measures on the all-day-evaluated indicators cannot be linearly or proportionally derived from peak-hour (due to the strong non-linearity of the relationships among the variables). Finally, it should be noted that one of the non-negligible (and desired) effects of TDM measures is to spread some of the peak-demand, and therefore an activity time choice model is recommended.
Within the AIUTO project one of the test sites (Randstad) has explicitly considered the long term effects of TDM measures, while the other test sites have focused their attention on short-medium term effects. Even if the TDM measures were not intended for their long-term effects, it could be extremely useful to state the contribution of TDM's to long-term changes in travel demand patterns. However, it seems that long-term prediction issues should be extended to transportation-land-use interaction models.
Another key feature for demand models regards disaggregation. Since the general aim of demand models is to interpret and/or simulate users' behaviour, it could be relevant to apply different systems of demand models to different groups of users, at least when those groups are expected to behave or react differently to TDM measures. Most of the AIUTO test sites deal with demand segmentation, where the users are grouped into demand segments (characterised by similar attributes with respect to the modelled choice/s). The disaggregate demand models used in AIUTO have shown the relevance of demand segmentation.
The AIUTO experiences have shown that a major modelling issue is related to demand model elasticity. Model elasticity should be regarded from two different aspects.
Clearly, the modelled choices should be sensitive to the TDM measures. One cannot establish a priori which of the travel choices is influenced by TDM measures, so, in principle, all choice levels should be sensitive to TDM's. The most demand modelling oriented of the test sites (Como, Randstad and Salerno) have faced the elasticity challenge by including in each choice model both attributes directly influenced by TDM measures and attributes which take into account the indirect influence that comes from other choice levels. Such an approach is consistent with the application of the nested LOGIT model.
The AIUTO experiences have shown that a suitable and effective way in which different travel choices can be integrated into a consistent theoretical structure is to use the nested LOGIT model. It allows the model to have an open structure in which non-standard choices (typically related to the expected impacts of TDM measures) can be consistently integrated. However, the data base needed to set up such a system of models should be sufficiently large and complex or of a particular type (e.g. choice based sampling) to cover all the possible alternatives for the included travel choices.
The number of choice dimensions that each of the AIUTO test sites have considered is quite different, in fact one of the test sites has not used demand models at all (York), while all the others have implemented at least a mode choice model.
Some of the differences of the overall modelling structure among test sites are due to the fact that they simulate the application of different TDM measures. For instance, not all the test sites simulate parking policies, so not all implemented a parking choice model. However, some of the choice dimensions are indirectly affected from TDM measures other than the direct ones. For instance, a traffic restriction imposed in a given area could affect the parking pattern in a neighbouring zone; a proper system of demand models should take into account such an effect.
Review of network assignment approachesAssignment models play a crucial role where TDM measures are simulated. The outputs of assignment models are travel times and flows. The link flows and speeds are used to calculate some of the MOE's that capture the impacts of the TDM policy measures. The travel times output are also the key input of the travel demand models. All the test sites used the assignment to predict link flows and all the test sites used the assignment model to calculate the travel times input of the travel demand models.
Loosely speaking, traffic simulation models can represent traffic flows in a microscopic, or macroscopic way. A microscopic model is one in which individual vehicles are separately modelled. In a typical microscopic model, vehicles have individual properties such as velocity and road position. For instance, in the DRACULA model used in the York case study different vehicles have different abilities to accelerate and different "styles" of driving behaviour. By contrast, in a macroscopic model, a flow of vehicles on a road is considered a continuous variable. For instance, the PFE model also used in the York case study, is an example of a macroscopic model. In PFE, vehicles are stored not as individuals but as flow-rates. The program then calculates the delays which would be experienced by a flow of that size on the road assuming a certain cost-flow curve. It is also possible to have a model in which some aspects of behaviour are modelled macroscopically and some are modelled microscopically. For instance, as it is considered that delays at junctions are more important than delays along links, some models use cost-flow curves along links and microscopic modelling at junctions.
Microsimulation is extremely computationally intensive when compared with macrosimulation. As available computer power increases, this will become less important but for the moment, this remains an issue worth considering when deciding which type of traffic simulation is most appropriate. Microsimulation models, however, have a number of advantages. Perhaps the most important of these is in air pollution modelling. However, general theoretical properties of microsimulation models are not so well established as in macrosimulation models.
Another main difference in assignment models is due to a static or dynamic approach. We can identify two areas where the "dynamic/static" issue is important. Some models simulate the different levels of congestion that are experienced at different times of day, as well as the time-dependent way in which flows propagate through the network. This type of model is referred to as "within day dynamic model". Other dynamic models try to represent the fact that the traffic patterns vary from day to day (e.g.because a new traffic scheme takes some time to reach its full effect and/or because the traffic pattern has implicit "inter-periodical" profile). Models of this type are referred to as "day-to-day dynamic models".
In within-day-dynamic models, the network performance, or supply model can be considered dynamic if the delays experienced by vehicles on the successive links of a route change according to the time at which they travel. Microscopic modelling, which was discussed above, is "naturally" dynamic since the vehicles are simulated individually, typically by using short time-slices (such as one second or even less).
Dynamic supply and assignment models capture several traffic characteristics which are not available in static models. A dynamic model is better able than a static model to capture "over-capacity" queuing because it follows the trajectories in time and space of the vehicles. This capability is accentuated by the fact that within-day-dynamic models also simulate the consequent spill-back problems.
Other issues related to within-day-dynamic models are:
One of the test applications within AIUTO considered the impact of a replacement of the static assignment model in an existing transport modelling system for the Randstad area by a dynamic version. Such an advance might be desirable for the simulation and assessment of many novel TDM measures, such as peak period pricing and information systems. A static equilibrium assignment model of the Randstad area and its surroundings for the 1994 base year network has been transformed into equivalent dynamic representations. Three dynamic models were tested. The models have been compared with real-life observations and checked for internal consistency, while some sensitivity testing with respect to dynamic inputs also has been performed. The following conclusions can be drawn:
In conclusion, a robust conversion of a static to a dynamic assignment model has proved possible, through a careful conversion of speed-flow-density curves, plus further fine-tuning of model-specific parameters. However, more work is needed on model refinements and validation.
Most of the supply and assignment models used within AIUTO are in various ways (and to varying degrees) based on the principle of network equilibrium. It is widely recognised that the equilibrium assumption provides a very useful and reliable basis on which to conduct transportation network analysis.
Equilibrium assignment models can be distinguished according to "deterministic" or "probabilistic" route choice model. They are respectively referred to as DUE (deterministic user equilibrium) and SUE (stochastic user equilibrium). Deterministic approaches represent the well-known Wardrop principle, while probabilistic models are generally based on PROBIT or LOGIT-type probabilistic choice models. Probabilistic assignment models are increasingly used in modelling practice, due to their more rigorous behavioural interpretation.
It should be noted, however, that assignment to S.U.E. is not necessarily a probabilistic process and assignment to D.U.E. is not necessarily a deterministic process. In the York study, for instance, the STEER model probabilistically approached DUE whereas the PFE model deterministically approached SUE. Furthermore, from the modelling of York, it was not possible to draw a conclusion on whether DUE or SUE was more realistic in the models as used.
Recommendations for the use of existing demand and assignment models to evaluate TDM measures in different urban areasBased on the results of the AIUTO projects, some general recommendations can be summarised.
In the following table different demand and assignment models are listed. The demand dimensions are shown and, for each of these, different modelling approaches are taken into account. Where a choice can be made on which type of model could be used, both an "absolute" and a "relative" recommendation level is indicated. The absolute level refers to the need of including that dimension in the model structure, while the relative level refers to the use of a given type of modelling approach for a given dimension.
Recommended modelling approaches
Conclusions and recommendationsN.A. = Not Applicable
+ = low effort * = not recommended
++ = medium effort ** = recommended
+++ = high effort *** = highly recommended
++++ = very high effort
§ York (the only site which compared SUE and DUE) drew no conclusions on the matter whether SUE should be generally more highly recommended than DUE.
The previous sections of this report have synthesised the evaluations of the models and methodologies that were conducted by the six AIUTO test sites. The AIUTO test site evaluations were very useful both because of their heterogeneity and their commonality. The heterogeneity stemmed from the policies that were evaluated and the modelling capabilities that were used to analyse them. The commonality was in terms of the overall modelling framework, a classification of policies, the overall modelling structure and the measures of effectiveness.
Based on the experience gained from the test site applications and from the evaluation of the results, we have also drawn conclusions concerning the practical and theoretical aspects of the alternative modelling approaches.
Some site-specific conclusions can be also highlighted. For example, from Randstad study some outcomes have been achieved, about the effect of long-term planning - some of its long term studies reverse the effects of the short term ones - thus, what is good "short term" may well be disastrous "long term".
In Como, incentive ("pull") measures, if applied alone, were found to be rather ineffective in terms of causing a modal change from private car, while a good performance was achieved by "push" measures (mainly involving road- or park-pricing).
The York case study highlighted highly divergent estimates of pollutants from two current models - one of which is widely used (SATURN). This points to an urgent need to further study possible inconsistencies in pollutant modelling, and makes one wondering if it is even possible to get accurate measures of pollutants using macro-simulation models. Actually, the extreme differences in pollution modelling outputs seem to indicate that the macro and micro-simulation approaches to pollution modelling are inconsistent. Since pollution modelling is extremely important, it is pressing to find out how this can be resolved, which is correct and whether the experience of the micro-modelling (assuming it is correct) can be used to better tune the macro modelling.
Furthermore, different supply models do not always agree on the same study: since good supply modelling underpins good demand modelling, it is urgent to investigate these modelling differences in order to provide a more sound basis for transport modelling in general.
Although the AIUTO project has demonstrated that adequate modelling capabilities to analyse TDM measures are generally available (in the form of disaggregating travel demand model systems and dynamic network assignment), from some site-results it appears evident that more validation tests would be useful for a better assessment of the accuracy of such models.
As part of this validation process, it would also be useful to conduct comprehensive sensitivity analyses to identify the most critical aspects and the key parameters of the models. However, the state of development of these tools is such that they are not yet readily available for wider dissemination or for quick policy analyses. In their existing state these capabilities require a long period of data collection and calibration, as well as heavy user intervention and too much manual data-handling from sub-model to sub-model. Thus, their application to a new study area would call for a major investment in terms of time, money and involvement of highly skilled persons trained and experienced in transportation demand and network modelling.
Our key recommendation for further work is to conduct an AIUTO II project. The purpose of this follow-up project would be to develop a set of standard user-friendly software tools that would be used for quicker and easier applications of the models recommended by AIUTO. We propose to first develop the required software tools, and then, unlike what happened in AIUTO, apply a common set of tools with the recommended models to a range of test sites.
|
| |
 |
Last Updated: 06-12-1999 |
|
|
