How can new forms of mobility be modelled?

summary

  • The development of new mobility services reflects transformations in mobility modes: flexibility, use versus possession, intermodality, service economy.
  • Conventional traffic forecasting models have limitations. Multi-agent simulation, by taking into consideration a certain number of characteristics related to these new mobility services, is a solution to forecast their uses more accurately.
  • However, the use of these new models, which are complex and require much longer calculation times, must be the result of joint consideration with the contracting authorities, based on a cost-benefit analysis.

With the possibilities offered by digital technology, the development of new mobility services for users has accelerated in recent years: self-service bicycles, electric scooters, car-sharing… Access to these services broadens the range of solutions available to them for getting around, and thus contributes to the emergence of what is known as service-based and shared mobility, based on the use of a mode of transport that is uncorrelated to one’s possession.

This increasingly diversified offer implies profound changes in mobility practices with:

  • Self-service vehicles (cars, motorised two-wheelers, bicycles, scooters – with or without terminals)
  • Transport-on-demand (TOD) services
  • Autonomous transport-on-demand services (AMoD – autonomous shuttles, Personal Rapid Transit)
  • Vehicle-sharing services
  • Trip sharing services (car-sharing)

These services each have their own specificity: for short or long distances, for ‘door-to-door’ or ‘stop-to-stop’ journeys, in urban or sparsely populated areas. But what characterises them above all is their flexibility.

Mobility is increasingly determined by user demand. It is less routine, more flexible and less constrained by a fixed transport offer. So how can these new forms of mobility be modelled? How can their development be predicted? How can decision-makers be given the keys to organising the provision of services, often of a private nature, in their region?

Are conventional modelling tools still relevant?

Conventional traffic forecasting models are generally ‘4-stage models*’ or simplified choice models.

These models face several constraints which do not allow them to take into consideration the new mobility services in their entirety:

  • The spatial aggregation of journeys does not allow for the detailed consideration of short distance journeys, unless a very fine grid is established
  • Their temporal aggregation makes it difficult to estimate demand over all periods of the day Successive trips from home or secondary trips not linked to home are less well described
  • Multimodal trips are often neglected
  • Only travellers are modelled, without taking into consideration empty vehicles

To date, the use of conventional modelling has been encouraged because of the aggregated nature of the available data and the limitations of information processing systems.

However, the continuous improvement of computer tools and computing power now makes it possible to consider modelling movements at the individual level. Moreover, more and more data are available for the development of these models, which increases their efficiency.

The new approaches to travel modelling thus make it possible to overcome the limitations of conventional models. However, the result is, on the one hand, increasing model complexity and, on the other, much longer calculation times.

*The 4-stage model is a general conventional transport study plan enabling transport demand to be forecast by following four stages: the generation of trips (how many trips?), the distribution of trips (for which destination?), the choice of transport mode (with which transport mode?), and which way (using which itinerary?).

Multi-agent simulations at the service of new mobilities

Among the new modelling approaches, Multi-Agent Systems (MAS) are attracting a lot of interest.

In the MAS approach, each object (individual, vehicle, etc.) can be monitored individually. This approach makes it possible to explicitly model individual decision-making processes. It also makes it possible to track vehicles in a disaggregated (spatio-temporal) manner and to turn them into autonomous entities. This makes it possible to model all types of transport services, especially those requiring optimal fleet management.

The MAS approach also has other advantages: it enables dynamic simulations to be carried out and offers very broad analysis perspectives, going beyond the framework of transport systems, by integrating models of interaction with land use, systems for measuring environmental impacts, or regional economic analysis tools.

Nevertheless, there is a major prerequisite: MAS simulation requires very fine input data on transport mode use and user preferences. The dissemination of all the information on different media also requires a significant effort in terms of processing the input data. In addition, considerable computing power is required to be able to simulate passenger movements on the scale of a large city, in which several million people travel.

How can regions be supported in order to structure the available and future mobility offer?

In order to organise the mobilities offer in a region, today and in the years to come, it is necessary to evaluate the use of each service, the choice of transport modes and the expected development of uses.

In practice, the choice of modelling approach will depend on the characteristics of the new service, what is to be modelled, and the model’s ability to take this into account.

The table below shows the comparative analysis between the ‘classical model’ approach versus the ‘agent model’ approach, depending on the nature of the service to be modelled (car-sharing, shared micro-mobility, Vehicle-sharing services, PRT or autonomous shuttle).

ModeConventional model vs agent model
Car sharingCar-sharing can be developed effectively using conventional models.
Micro-mobilitiesMicro-mobilities are handled better by multi-agent simulations that take into consideration service flexibility. However, the implementation of such a simulation remains complex and data-consuming.
Vehicle-sharing servicesThe choice of vehicle-sharing model depends on the level of detail sought in the results.
PRT, Autonomous shuttle, Self-service vehicleMulti-agent simulations will allow a detailed consideration of PRT or autonomous vehicles whereas conventional models must be made more complex with new modules for an adequate consideration.

Currently, multi-agent simulations are the dominant approach to modelling new mobilities in the research world. The simulation of travel on this microscopic scale (traveller/vehicle etc.) is relevant for the detailed study of the impacts of new transport offers that are increasingly individualised.

Nevertheless, the data currently available are not always in line with the design of such simulations (low volumes in surveys leading to large margins of error, commercial and non-public ridership data, large and complex databases to be analysed, etc.). The resulting calculation times do not allow these tools to be used in a pragmatic way, with a view to multiplying tests and scenarios.

AIRBUS CASE STUDY

While MAS models are very powerful, they require a high investment to ensure robust output results. This is at least what we observed during a consultancy mission for Airbus, which was carried out in 2019 with the support of the ETH University of Zurich, on the modelling of a service offer for flying shared vehicles (Urban Air Mobility).

SYSTRA modelled the service offer for flying shared vehicles using a price-time model, based on matrices derived from a 4-stage model in the Greater Paris region. This approach was compared with the results from MatSim*.

The comparison of the volumes of all-mode flows between Greater Paris departments derived from the four-stage model of Greater Paris and from MatSim show real similarities. The comparison of the shares of each mode of transport showed the need to improve the calibration of MatSim with the data from the Household Travel Survey, which is very time-consuming.

Once calibrated, the 4-stage model of the Greater Paris region can be simulated in several hours while the MatSim simulation takes place over several days. The price-time model, once configured, is immediate. It is nevertheless very simplified and only offers a partial approach to the problem.

*Open source multi-agent simulation platform.

to conclude

For a public authority or contractor, the ways to deal with the new services and the choice of tools must be considered  according to the available data, the ability to obtain complementary data, and the time required to construct models adapted to the objectives of the study. SYSTRA accompanies project owners to assist them in choosing the best solutions, in line with their needs.

Don't miss anything, follow us on social media !

Warning: Illegal string offset 'title' in /var/www/prod/htdocs/wp-content/themes/systra/single-expert_insights.php on line 177

Warning: Illegal string offset 'url' in /var/www/prod/htdocs/wp-content/themes/systra/single-expert_insights.php on line 178