Strategic Demand Model in Flanders Using Large-scale Agent-based Microsimulation: How to Handle the Results



Strategic Demand Model in Flanders Using Large-scale Agent-based Microsimulation: How to Handle the Results

Authors

Kurt Verlinden, Significance

Description

Flanders 4G models run agent-based demand models with microsimulation, and provide broad analyses on micro-level. Classic evaluation can be used, but the rich data offer more insights. Do’s and don’ts using micro-models are shown via applications

Abstract

As of 2014 the Flemish Administration responsible for the strategic transport modelling, shifted its development focus in the 4th generation framework towards full agent-based demand models, in so making use of the available observed and synthetic micro-data of the Belgian population. The developed demand modes inherently keep the well-known and approved choice model setup following general logit formulation, but now incorporate a most general joint formulation of all essential choices combined with the highest level of detail on individual and family characteristics of each separate agent. This approach allows for a significantly more powerful formulation of the explanatory drivers and variables on specific individual level, integrating interactions between choices as well as decision takers beyond the classical generalized or aggregated attributes. By using Monte Carlo-simulation techniques, the overall model operates on a strict discrete level, resulting in a final outcome that provides a complete description of a valid travel pattern of the whole population for a given day, and this on specific individual levels. Several advanced procedures are introduced in order to both limit and quantify the inherent simulation error, thus warranting stable results in model application and avoiding noise caused by the stochastic nature of the instrument.
Since the demand model takes a broad set of individual and family demographics into account, the 4th generation instrument offers an interesting assessment tool with regards to changing social conditions and compositions, going beyond the classic aggregate approach focusing on network schemes. Moreover, the agent-based design allows for a seamless combination of micro-level behaviour with large-scale strategic applications, offering opportunities to assess major or generic schemes from the bottom up.
The paper aims to illustrate the richness of the proposed modelling setup and resulting outcomes by means of presenting a set of schemes and scenarios addressing different topics like ageing of the population, decreasing car ownership, changes in employment or study-participation as well as dedicated parking or park-and-ride schemes. All evaluations show particular behavioural changes over different population groups due to the elaborate sensitivities of the demand model, and offer ways of pinpointing overall results towards the underlying explaining effects. Pitfalls however come up as well, since the level of detail in the discrete outcome tends to lead towards too detailed analyses, losing sight of the inherent simulation error that can cloud sound judgements. Conclusions lead to the insights that the offered possibilities on the one hand stretch the current evaluation framework so that, for example, socio-economic cost-benefit analyses can be supported with useful segmentations and more. On the other hand, care must be taken not too divert into too detailed assessments of local or individual outcomes. Challenge lies in the weighing off of useful details against false or misplaced precision, which, in fact, technically comes down to getting a solid grip of the extent of simulation error and risk on particular outcomes.

Publisher

Association for European Transport