VACANCY INTERNSHIP PROJECT We are looking for students interested in carrying out their internship project at DAT.Mobility
Removing stochasticity from an agent-based travel demand model to allow for strategic applications Problem description Recently, DAT.Mobility has added the agent based travel demand model BRUTUS to OmniTRANS transport planning software (see https://tinyurl.com/y96m54kp, https://www.mobilitymodeling.com/brutus/ and https://www.dat.nl/omnitrans/ for more info). This model describes the number of trip chains along with their destinations and modes for all persons and households within the study area. Although behaviorally more accurate, BRUTUS in its current form has a disadvantage when it comes to strategic applications: it contains stochasticity. In strategic applications transport model outcomes are used by the model analyst who compares different future scenario’s or solution variants to a reference situation and translates insights from this comparison into advice to decision makers. When the model contains stochasticity, differences may be caused by differences in the inputs (which is desirable) or due to stochastic randomness (which is not desirable).
Internship assignment The goal of this research is to isolate the stochastic processes within the current BRUTUS implementation and replace them by deterministic methods, thereby eliminating stochasticity within the model whilst maintaining the level of accuracy as much as possible. BRUTUS is programmed in R, which provides a good environment for prototyping, allowing to directly see the effect of changes to the implementation. We have already identified three sources of stochasticity: the population synthesizer, the destination choice sampler and the method that determines which household member will use a household car. For the population synthesizer (multizone) iterative proportional fitting seems to be the most desirable method (Brederode and Waanders, 2013). For the destination choice sampler, we think the most desirable method would be to remove it altogether or introduce importance sampling (Miller et al., 2007). As these insights where gained 5 to 10 years ago, a literature research to the state of the art is required to update them. For the household car method, no alternative method has been identified yet. Once the most desirable methods have been identified, they should be implemented and tested for correctness in a synthetic 3 zone BRUTUS model and tested for applicability in the BRUTUS model of Utrecht or Almere. It is likely that eliminating all sources of stochasticity is not feasible within one internship / Master thesis project. Furthermore, more sources of stochasticity might be identified during the internship / Master thesis project. During the definition of the research scope and planning, these uncertainties and practical constraints need to be addressed. The student is encouraged to align the research scope and planning to his/her interests and will be required to show some flexibility.