Real-time prediction of public bike sharing system demand using generalized extreme value count model

Public Bike Sharing Systems (BSSs) are becoming increasingly popular in recent times. Both the BSS operators and the customers can benefit from the large digital data portals that continuously record the state of the BSS. In this context, the current study developed generalized extreme value (GEV) count models that can predict hourly bike arrivals and departures at each station while accounting for time-of-day, weather, built environment, infrastructure, temporal, and spatial dependency factors. The proposed models were used to analyze the demand patterns in the Capital Bikeshare system and were found to predict the demand at both aggregate and disaggregate levels with reasonable accuracy. Specifically, the total demand in the entire system was predicted within 5% margin of error whereas 75% of the station-level arrival and departure predictions in the next one hour were within a margin of one from the observed counts. The proposed modeling system is useful (a) to BSS customers to better plan their travel based on expected bike and dock availability at the origin and destination ends of their BSS trips, and (b) to BSS operators to anticipate the future demand and optimize their rebalancing plans.