Demand Inference for Free-Floating Micro-Mobility: Accessibility and Availability

Abstract

Micro-mobility systems (such as bike-sharing or scooter-sharing) have been widely adopted across the globe as sustainable modes of urban transportation. To efficiently plan and operate such systems, it is crucial to understand the underlying rider demand --- where riders come from and the rates of arrivals into the service area. Estimating rider demand is not trivial as most systems only track trip data, which are a biased representation of underlying demand. In this report, we describe development of a locational demand model to estimate rider demand based on only trip and vehicle location and status data. We established conditions under which our estimators are identifiable and consistent. In addition, we devised an expectation-maximization (EM) algorithm with closed-form updates for efficient estimation. To scale the estimation procedures, this EM algorithm is complemented with a location-discovery procedure that gradually adds new locations in the service region that make the largest improvements to the log-likelihood. Experiments using both synthetic data and real data from a dockless bike-sharing system in the Seattle area demonstrated the accuracy and scalability of the model and its estimation algorithm.