Micro-simulation with Discrete Choice Models: Application in Flanders
Kurt Verlinden, MINT, Cindy Puttemans, MINT, Michiel De Bok, Significance
Paper describes design of new generation Agent Based Model using micro-simulation in discrete choice framework and addresses methods developed to maximize benefits of micro-simulation and to mitigate the impact of simulation error and inefficiency.
The proposed paper aims to elaborate on the design and development of a new generation Agent Based Model using full micro-simulation in the discrete choice framework. In doing so it addresses methods and procedures developed to maximize the benefits and richness of micro-simulation on person and household level on the one hand and to mitigate the impact of simulation error and resource inefficiency on the other hand.
The newly developed 4th generation traffic model framework for Flanders incorporates travel demand data on individual basis, extracted from an advanced Population Simulator and address-based activity data. An extended multimodal network model integrating improved assignment for car and timetable based public transport provides the necessary level-of-service data in the highest feasible resolution. The demand model operates on a discrete tour-level, as drawn from an elaborate tour-frequency model handling individual persons and household. A simultaneous mode-destination-time choice model handles individual full tours based on all relevant agent’s characteristics.
Simulation based on the discrete choice framework uses Monte Carlo draws over the probability-spaces, and inherently introduces a degree of randomness in the results, see also the general review on this topic in the companion paper (de Bok, de Jong, Helder, Verlinden and Puttemans). Careful microseeding is used to stabilize repetition within model application. To minimize noise between different model runs, techniques of complex averaging combined with efficient multiple draws are introduced. As such, results remain in the discrete space with proper attention to accuracy and confidence of the overall discrete results. This allows for a broad and full set of individual results with relevant characteristics with the highest likelihood. Moreover, the complete system is designed to allow for flexible use of the internal choice models, offering full control of activating or disabling parts of the demand model, switching where necessary to various averaging techniques.
Starting from the design, the developed model system aims to efficiently perform all calculations in a most transparent way. In so all intermediate processes and results can be monitored to ensure valid proceedings. In the same context, in different stages and dimensions of the demand model, specific calibration parameters are introduced making feedback from network calibration possible using small corrections to the utility space of the internal discrete choice models.
Association for European Transport