Sahnoon, Mohammad (University of Calgary), Manuel, Aaron (University of Calgary), Demissie, Merkebe Getachew (University of Calgary, Calgary, Canada), Medeiros de Souza, Roberto (University of Calgary)

UNET and UNETR Based Frameworks for Predicting the Short-Term Spatiotemporal Demand of E-Scooter Sharing Services

Scheduled for presentation during the Poster Session "Modeling and Optimization of Mobility and Transport Systems " (ThBT14), Thursday, September 26, 2024, 14:30−16:30, Foyer

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 3, 2024

Abstract

Several factors have contributed to the emergence of shared on-demand mobility services in recent years, including urbanization, technological advancements, environmental concerns, changing consumer behavior, regulatory changes, and cost savings. Among these services, shared electric micromobility services represent the latest entrant to the market. One significant challenge for shared mobility services is that demand for these services can vary significantly throughout the day and across different locations, leading to imbalances in vehicle availability. This can result in long wait times for users, negatively impacting user experience and discouraging future service usage. In this study, we propose the use of the state-of-the-art deep learning models, such as the UNET and UNETR, for short-term spatiotemporal micromobility demand prediction. Our study reveals that UNETR surpasses the baseline model in predicting demand for the entire region of interest. For the next-hour pick-up and drop-off demand prediction, UNETR achieves mean absolute errors of 0.0163 and 0.0158, respectively, while for the next 24-hour prediction, the errors are 0.0166 and 0.0158, respectively. Additionally, UNET outperforms the baseline model and UNETR at the non-zero demand level, with mean absolute errors of 1.4886 and 1.4430 for the next-hour prediction, and 1.5607 and 1.5339 for the next 24-hour prediction, respectively.