Are you struggling to find available docks to return your bikes or frustrated with finding no charging ports in a charging station for your EVs? Our future mobility systems require better simulation and prediction to support systems design decisions, such as station locations and capacities. Our recent study, led by Yinshuang Xiao and collaborated with Dr. Faez Ahmed, using cutting-edge graph neural networks to improve demand prediction in shared mobility networks was featured by ASME Journal of Mechanical Design. Let us rev up shared mobility for a greener and smarter commute!

Emerging shared mobility systems are gaining popularity because of their cost-effectiveness and positive environmental impact. In designing such urban transportation networks, questions such as where a station should be constructed and how much capacity it should offer are essential because they influence system usage, adoption rate, and return on investment. In this study, we develop a cutting-edge method using graph neural networks to support the design decision-making of shared mobility systems, like bike-sharing programs. One uniqueness of this method is that it accounts for the neighborhood information of a station, such as shopping malls or schools, in the formulation of node embedding in training the graph neural networks. When predicting if there are sufficient demands between stations, our results indicate the proposed method has superior performance, thus offering practical guidance for optimizing station locations and capacities. Our model can be readily transformed into a simulation tool or integrated into a decision-support platform for the design of more efficient and sustainable shared mobility networks.