OpFlow: Learning Opportunity-Conditioned Choice Potentials for Robust OD Flow Prediction

2026-07-03

AI Summary

Existing deep learning models based on raw origin-destination (OD) flows exhibit poor generalization under distribution shifts, primarily due to their inability to disentangle transferable choice mechanisms from environment-specific shortcuts. To address this, this work proposes OpFlow, a novel framework that decomposes OD flows into a transferable, origin-centered choice potential and origin-scale components. Theoretically, we establish the identifiability of this potential and show that classical spatial interaction models are special cases of our formulation. OpFlow employs mechanistic constraints within a deep learning architecture to model the mapping from spatial exposure to relative destination preferences, and reconstructs OD flows by combining this potential with independently calibrated origin scales. Experiments demonstrate that OpFlow significantly enhances prediction robustness under both synthetic and real-world distribution shifts.

Abstract

Origin-destination (OD) flow prediction is central to urban analytics, yet deep models trained on raw counts remain vulnerable to distribution shift. The core problem is that raw count supervision cannot distinguish transferable choice mechanisms from environment-specific shortcuts. Raw OD count mixes two objects: how much demand an origin produces and how that demand is allocated across destinations. We argue that the transferable object is the exposure-to-choice law that maps spatial conditions to relative destination preferences. We propose OpFlow, a mechanism-constrained framework that learns row-centered choice potentials and reconstructs flows by combining the induced allocation with a separately calibrated origin scale. Under distribution shift, spatial exposures and the induced allocations are allowed to vary; what transfers is the conditional map from exposure states to relative choice potentials. Theoretically, we characterize the identifiable row-centered potential and show that classical spatial interaction laws are restricted log-potential cases. Controlled synthetic shifts and a real-world experiment show OpFlow improves robustness under environment shifts.

Problem

• OD flow prediction
• distribution shift
• choice mechanism
• spatial interaction
• robustness

Innovation

• choice potential
• distribution shift
• origin-destination flow
• spatial interaction
• mechanism-constrained learning