Stochastic travel demand estimation: Improving network identifiability using multi-day observation sets

Stochastic travel demand estimation is essential to support many resilience and reliability based transportation network analyses. The problem of estimating travel demand based on sensor data often results in an ill-posed inverse problem, where solution uniqueness cannot be ensured. To overcome this challenge, effective utilization of more information/data, preferably from reliable sources, becomes critical. Conventional demand estimation methods often sacrifice system structural information during the process of compressing sensor data into its statistics. Loss of structural information, which captures critical relation between observed and estimated parameters, inevitably causes more dependence on unrealistic assumptions and unreliable data. Our model is designed to preserve all structural information contained from different observation sets and allow it to directly contribute to the identification of population parameters of travel demand. The proposed hierarchical framework integrates two traditionally distinctive identification problems, mean demand estimation and trip table reconstruction. Through mathematical analyses and numerical experiments, we show that the proposed framework improves parameter identifiability and leads to better estimation quality compared to conventional methods. The proposed framework is also flexible to accommodate a wide variety of travel behavior assumptions and estimation principles. As an example among many possible alternatives, Wardrop equilibrium based traffic assignment and generalized least square are implemented and tested using a case study based on a moderately large network.