A Smart Predict-then-Optimize method for dynamic green bike relocation in the free-floating system

Cities are facing numerous challenges such as road traffic congestion and environmental pollution. Bike-sharing, as an emission-free travel mode, aligns with the principle of green and environmental protection and serves mainly for short-distance trips in urban areas. Shared bikes pose significant issues to operators due to supply–demand imbalances across different time and space. This study proposes a Smart Predict-then-Optimize method for dynamic green bike relocation in the free-floating system, which aims to minimize the cost of fuel and carbon emissions from repositioning vehicles and the total unmet demand during the operating period. A multi-task deep neural network model is designed to predict regional inflow and outflow demand, where targeted modules are embedded to extract the spatio-temporal characteristics. Potential unusable shared bikes are discovered from the users’ travel behavior and the usage characteristics of shared bikes. Then, we build a data-driven optimization model for bike-sharing relocation and design an iterative decomposition algorithm that incorporates an adaptive large neighborhood search for relocation routes and vehicle speed optimization. The proposed method is tested on real-world bike-sharing trips in Shenzhen, China, and results show that the relocation distance and carbon emission cost can be reduced by 11.69% and 14.09% by relocating operational bikes and unusable bikes simultaneously. Additionally, route decision-making with speed optimization can decrease the total fuel and emission, where considering the collection of unusable bikes help improve the service level of the system.