An interpretable generative probabilistic framework for demand characterization and consistency checking in...

Area of Science:

• Urban Mobility and Transportation Science
• Data Science and Probabilistic Modeling

Background:

• Bicycle-sharing systems (BSS) are crucial for sustainable urban mobility but face challenges in demand modeling due to complex influencing factors.
• Existing data often offer incomplete insights into the true demand patterns of BSS.

Purpose of the Study:

• To develop an interpretable probabilistic framework for characterizing and generating synthetic demand for dock-based BSS.
• To provide a baseline for demand analysis, synthetic data generation, and data quality assessment in BSS.

Main Methods:

• Utilized trip-level data from Madrid's BiciMad system (2018-2019).
• Modeled trip distances using Gamma distributions and hourly trip counts using Negative Binomial distributions, conditioned on temporal and weather factors.
• Integrated probabilistic components with station-popularity profiles to generate synthetic origin-destination demand.

Main Results:

• The framework demonstrated calibrated uncertainty estimates, with 95% prediction intervals showing empirical coverage close to nominal levels.
• Generated synthetic demand data aligned well with observed data across various scenarios.
• Identified systematic timestamp misattributions in the dataset through an external consistency check.

Conclusions:

• The proposed framework offers a simple, interpretable, and uncertainty-aware method for BSS demand characterization and synthetic data generation.
• The approach is valuable for exploratory disruption analysis and ensuring data quality in dock-based BSS.
• It serves as a foundational tool, not a full operational simulator, for understanding BSS demand dynamics.