Uncertainty analysis of an activity-based microsimulation model for Singapore

Abstract

Transport models inherit uncertainties due to a variety of assumptions, inputs and their structural properties. With the increase in complexity of modern travel demand models, the uncertainty analysis becomes more important and it becomes a non-trivial procedure that requires a careful consideration for its investigation. This paper analyses the model uncertainty of the activity-based microsimulation (ABM) travel demand model including specification and parameter uncertainty. The ABM model predicts the entire day activity-travel schedule and was developed and calibrated using a variety of datasets from Singapore. The model is computationally heavy as it includes 22 sub-models, which follows multinomial and nested logit structure for different activity-travel decisions and in overall includes 817 parameters. The model specification uncertainty addressed in the study include simulation error and sample uncertainty, both measured by means of simulation techniques under various scenarios with running the entire model. The parameter uncertainty is estimated based on the simulation with sampling from a parametric multivariate distribution preserving the correlations between the sampled variables. The parameter uncertainty includes a sensitivity-based screening of the sub-models to identify the major contributors, followed by simulation runs of the entire model for the most influential parameters. The results showed that the order of magnitude of all considered kinds of uncertainty strongly depends on how frequently the alternative is predicted in the choice process. The parameter uncertainty is higher than the sampling uncertainty, and the sampling uncertainty is comparable with the simulation uncertainty. Moreover, this is the first study to compare the order of magnitude of the simulation, sampling and parameter uncertainties of an ABM model. The suggested method can be used to analyse both input and parameter uncertainties in computationally heavy models that have a hierarchical structure consisting of smaller sub-models. The uncertainty calculated for the model outcomes in this study will allow practitioners to choose a strategy for dealing with it.