Inferring origin-destination pairs and utility-based travel preferences of shared mobility system users in a multi-modal environment

• An expectation maximization methodology of inferring true OD and utility-based travel preferences is developed and applied. True OD is considered as the unknown latent variable and traveler preference is considered as the unknown variable. The methodology observes route/mode choice changes of users due to perturbations (pricing) in the share mobility system with repeated observations.
• The Selective Set Expectation Maximization (SSEM) is developed for data sets with repeated observation. SSEM only searches over choices consistent with all the repeated observations which increases the accuracy of inference results.
• A simulation framework is developed for bike sharing system analysis with heterogeneous travelers in a multi-modal travel environment.
• Promising computation results are obtained in estimating both true ODs and traveler preference distribution with disaggregate data.
• The inferred quantities can inform bike sharing system operations, facilitating inventory rebalancing.

This paper presents a methodological framework to identify population-wide traveler type distribution and simultaneously infer individual travelers’ Origin-Destination (OD) pairs, based on the individual records of a shared mobility (bike) system use in a multimodal travel environment. Given the information about the travelers’ outbound and inbound bike stations under varied price settings, the developed Selective Set Expectation Maximization (SSEM) algorithm infers an underlying distribution of travelers over the given traveler “types,” or “classes,” treating each traveler’s OD pair as a latent variable; the inferred most likely traveler type for each traveler then informs their most likely OD pair. The experimental results based on simulated data demonstrate high SSEM learning accuracy both on the aggregate and dissagregate levels.