aalto1 untyped-item.component.html
Reducing randomness in microsimulation travel demand models
Loading...
URL
Journal Title
Journal ISSN
Volume Title
School of Engineering |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
Department
Major/Subject
Mcode
Language
en
Pages
66
Series
Abstract
Stochasticity in activity-based travel demand models is often treated as an inherent feature of discrete choice models; however, limited attention has been paid to its underlying sources and to how uncontrolled randomness can compromise the stability and policy relevance of model predictions. This thesis investigates the sources of randomness in the Brutus simulation framework, a large-scale, data-driven, activity- and agent-based travel demand model implemented as a discrete Monte Carlo simulation. Within this framework, three sampling strategies, namely, Monte Carlo, Halton sequences, and Frozen Randomness, are evaluated for their effectiveness in reducing stochastic variability in the multinomial Logit model.
The models are tested across aggregated and small-grid spatial units, and their performance is assessed using variance behavior, directional consistency under policy scenarios, and stability of behavioral responses. The results show that most stochastic variation originates from the mode choice model, and that randomness becomes strongly amplified at fine spatial scales where population counts are low. Monte Carlo sampling shows the highest instability, while the Halton sequences had only limited improvement at the micro level. Frozen Randomness produces the most stable results. It can always accurately capture the expected policy effects in various scenarios and reveal the behavioral patterns that are hidden beneath the randomness of methods.
These findings are important for the development of activity-based and agent-based travel demand models. They show that understanding and controlling the sources of stochasticity is crucial to distinguishing between behavioral responses and simulation noise, especially when evaluating targeted policy interventions and small-scale planning measures. The results also show that variance reduction methods can significantly improve the interpretability of scenario analysis, and although Frozen Randomness has received less theoretical attention in the literature, it provides a practical and effective alternative to traditional random sampling. The thesis also emphasizes the necessity of scalable implementation methods and explores other methods to reduce randomness and achieve more repeatable results.
Description
Supervisor
Vosough, ShayaThesis advisor
Vosough, ShayaChen, Ruiwei