Fleet availability analysis and prediction for shared e-scooters: An energy perspective

E-scooters have become a prevalent mode of transportation in many cities. The availability of e-scooters is a crucial indicator of service quality but has not been sufficiently investigated. We propose a two-stage method for fleet availability analysis and prediction, considering stochastic demand and a new energy perspective. First, we developed a SpatioTemporalAttentionNet (STAN) model to predict trip OD. Second, we propose a Monte Carlo-based algorithm to match demand with existing e-scooters across spatiotemporal and energy dimensions. We conduct case studies using real-world data from Gothenburg, Sweden. The results indicate an average unavailability rate of 6.71%, nearly doubling that of the benchmark group, which uses a 20% SoC threshold for determining availability. This rate is significant considering the large fleet size and highlights the need to incorporate battery levels into fleet management. We further investigate the multifaceted impacts of land use and walking distance on availability dynamics.