Modeling Real Demand in Dockless Bike-Sharing Systems: Integrating User Preferences and Behavioral Insights

Abstract

The usage data of dockless bike-sharing (DBS) systems are often used as real data for station location and rebalancing optimization research. However, when DBS users cannot find available bikes at their desired locations, their travel mode decisions may result in demand being recorded to other areas or not recorded because they abandon the use of bikes, resulting in a bias between usage data and real demand. However, the lack of understanding of users’ travel mode decisions in the case of unavailable bicycles causes the accuracy of the real demand estimation model to be limited. To address the above issue, this study designed a stated preference questionnaire to assess users’ travel choices when bikes are unavailable at their preferred pickup locations. The results indicate that users are inclined to seek alternative bikes along their intended path and typically do not walk more than 300 m to retrieve a bike. Then, we developed a travel mode decision model based on user preferences, which considered user walking endurance and migration area choice probability in the absence of bicycles available for borrowing. Finally, based on the Poisson model, a user preference-based demand truncation and migration Poisson model (DTMP) was developed. The results demonstrate that the user preference-based DTMP model enhances accuracy by approximately 8% compared to the baseline DTMP model. These results offer a more precise foundation for DBS demand forecasting and fleet size and distribution, improving service quality and user satisfaction.