Agent-based Transport Models as a Tool for Evaluating Mobility
Licentiatavhandling, 2022

The transportation system is undergoing fundamental transformations through emerging technologies. Some of these innovations have the potential to contribute to the sustainable transformation of the transportation system, such as electric vehicles (EVs) and shared autonomous electric vehicles (SEAVs). Before enacting policies to support these technologies or limit the use of undesirable ones, decision-makers need to better understand these innovations and the consequences of the policy to be implemented. This insight can be provided with models that are capable of reflecting the dynamics of new mobility, and interactions of travelers with each other and the infrastructure. This thesis describes the development of the Synthetic Swedish Mobility (SySMo) model that represents the travel behavior of an advanced synthetic population of Sweden, using an agent-based framework. The SySMo model provides a scaffold to build decision support tools through which present and future mobility scenarios can be analyzed and thus aid decision-makers in formulating informed policies. The SySMo model comprises a series of modules that utilize a stochastic approach combined with Neural Networks, a machine learning technique to generate a synthetic population and behaviorally realistic daily activity-travel schedules for each agent.

The model first generates a synthetic replica of the population characterized by various socio-economic attributes using zone-level statistics and the national travel survey as input data. Then, daily heterogeneous activity patterns showing activity and trip features are assigned to each individual in the population with a high spatio-temporal resolution. To assess the SySMo model performance in each module, in-sample evaluations (i.e., comparing the model outputs with input data to measure the similarity of the results) and out-of-sample (i.e., comparing the model outputs with data never used in the model) evaluations are performed. The current model offers a valuable planning and visualization tool to illustrate mobility patterns of the Swedish population. The methodology can also be broadly applied to other regions with other relevant data and carefully calibrated parameters.

Machine learning

Daily activity pattern

Activity-based modeling

Agent-based modeling

Activity generation

EA-Hall, Hörsalsvägen 11
Opponent: Adj. Prof. Dr. Leonid Engelson, Department of Science and Technology, Linköping University, Linkoping, Sweden


Çaglar Tozluoglu

Chalmers, Rymd-, geo- och miljövetenskap, Fysisk resursteori

Ç. Tozluoğlu, S. Dhamal, Y. Liao, S. Yeh, F. Sprei, M. Marathe, C. Barrett and D. Dubhashi (2022a). Synthetic Sweden Mobility (SySMo) model documentation.

Ç. Tozluoğlu, S. Dhamal, S. Yeh, F. Sprei, Y. Liao, M. Marathe, C. Barrett and D. Dubhashi (2022b). The heterogeneous travel activity of a synthetic population.


Annan data- och informationsvetenskap

Transportteknik och logistik





EA-Hall, Hörsalsvägen 11

Opponent: Adj. Prof. Dr. Leonid Engelson, Department of Science and Technology, Linköping University, Linkoping, Sweden

Mer information

Senast uppdaterat