The Need of Combining Different Traffic Modelling Levels for Effectively Tackling Complex Project



The Need of Combining Different Traffic Modelling Levels for Effectively Tackling Complex Project

Authors

J Casas, TSS-Traffic Simulation Systems / Vic University, ES; A Torday, A Gerodimos, TSS-Traffic Simulation Systems, ES

Description

Instead of being compared, different modelling approaches should rather be used in function of the objectives of a project and even combined in order to get the best of each at each step of the project

Abstract

Transport engineers have, for decades now, been relying increasingly on the use of mathematical models and specialist software for analysing the performance of current and future transportation networks. Macroscopic software packages, generally based on static paradigms, pioneered the field, to be followed later by more disaggregated and dynamic models. Benefiting from the steadily increasing availability of affordable computing power, these more detailed models have become the tool of choice for operational studies, commonly in the form of microscopic simulators. Among other dynamic models, mesoscopic ones have more recently started to receive attention as a viable and interesting compromise between the macro and micro levels. With the introduction of new technologies, data of unprecedented quantity, detail and ultimately quality is set to become available making nanoscopic models a viable future prospect and an interesting research direction.

The proliferation of levels, approaches and software packages inevitably creates a temptation to compare. Comparisons quickly mutate into contests that focus on limitations; after all, those can be easily identified by taking a critical look at a model?s underlying assumptions. For example, a static model is, by definition, not appropriate for studying the impact of different adaptive control regimes. A dynamic equilibrium assignment approach is probably not the most realistic way of predicting driver response to a non-recurrent incident. Using a micro-simulator for a 35-year strategic plan without information on the location and capacity of roads - let alone traffic control plans, types of vehicles and driver behaviour ? is a likely waste of resources. Mesoscopic models, whether working with platoons or individual vehicles are not the most precise when dealing with merging, oversaturated flows, actuated detection and interactions with pedestrians at crossings. And the list continues: today?s fastest micro-simulator may be good enough to run a simulation of the entirety of Singapore faster than real time; but it is still way too slow for carrying out real-time traffic analysis in the entire Los Angeles metropolitan area. With its detailed modelling of a driver?s decision-making process every fraction of a second, a nanoscopic model seems a promising and more appropriate way of analysing aspects such as emission patterns or ADAS. But what about its (currently) disproportional calibration and computing time requirements?

The obvious conclusion is that there is no overall ?contest winner? and that each model has its limitations and strengths and those depend on the intended application, data availability, time horizon and evolution of computing and ITS technology. Specialist consultants typically adopt an impartial approach opting to acquire and learn a variety of tools and to use ?the right model for the right job?. From a practical point of view, it is both attractive and appropriate to devise informed rules of thumb for choosing a particular approach. While this is clearly less error-prone than a dogmatic approach, one might question whether it is actually possible to compartmentalise transport engineering projects in such a neat way. Is it really possible to speak of a ?static assignment project? and a ?micro-simulation project?? What if one needs both models within the project? And what about mesoscopic approaches? Should we look for ?mesoscopic project? opportunities? What?s more, what if one needs to use two types of models iteratively or even concurrently?

The model integration seems therefore the obvious solution. It is possible to implement it either within a single multi-level framework or by integrating modelling approaches originally developed independently. The second method relies on the exchange of information via files and lacks some of the convenience, possibilities and economy of the first method: multiple tools imply duplication of cost, effort and data and propensity for error. That said, the multi-tool approach is feasible and can be considered in projects where different models are used in sequence. Where we believe that a single model and software architecture has a distinct advantage is when models must be used concurrently or iteratively. Working inside the same software is not just a case of convenience for the user (or the developer): for one, the coherence of the two models forming a hybrid is a necessary condition for its robustness and fidelity and the ultimate reason why fusion is the way forward.

Publisher

Association for European Transport