Answers are grounded in this paper and cite their sources.
• Temporal traffic demand distribution offers key insights for transport planning. • Count-based TDOD estimation methods highly rely on observed traffic counts. • Theory-driven time-of-travel forecasting methods rely on stated/revealed preferences. • ML prediction of the temporal distribution of traffic demand proves effective. • It overcomes limits of count- and theory-based TDOD methods. . Temporal distribution of travel demand provides valuable insights for the planning and operation of transport systems. As a key input to dynamic traffic assignment (DTA) models, estimation of time-dependent origin-destination (TDOD) travel demand matrices across the modelled network gained attention in the 1980s, with significant advancements in methods and techniques since then. However, heavy reliance on observed traffic counts has long been recognised as a limitation of these approaches. In the travel demand modelling (TDM) domain, the time-dependency of travel demand has been associated with travellers' characteristics, typically implemented through theory-based choice modelling (CM) methods informed by stated or revealed preference datasets. CM's reliance on preference datasets introduces new shortcomings, e.g. heavy cost of reliable surveys, which constrain its broader application for temporal demand distribution prediction. In this paper, we demonstrate the successful application of machine learning in predicting the Origin-Destination (OD) temporal demand distributions based on TDM data, including sociodemographic and land information of origin and destination zones as well as origin-destination network statistics, e.g. travel time. We used a combined dataset incorporating TDM data and available count-based TDOD estimates to make separate train and test datasets. We train ML models that accurately predict the temporal traffic demand distribution at the origin-destination level for the test dataset. Our approach effectively addresses the limitations associated with count-based estimation and theory-driven choice modelling approaches.
No takes yet. Share an insight, caveat, or question.
Pourhassan et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: