Origin-Destination Demand Prediction Using Urban Radiation and Attraction

A research paper on arXiv proposes a new approach for origin-destination (OD) demand prediction by integrating physical radiation and attraction capacities with deep learning. Existing data-driven methods focus on spatial or temporal dependencies but neglect functional differences between regions, while knowledge-driven physical methods define capacities based on numerical factors like population, ignoring nominal attributes such as whether a region is residential or industrial. The paper also addresses the dynamic transformation of radiation and attraction capacities over time, e.g., a residential district's radiation in the morning becomes attraction in the evening. The study generalizes physical radiation and attraction models to improve prediction accuracy.