T-STAR: A Context-Aware Transformer Framework for Short-Term Probabilistic Demand Forecasting in Dock-Based Shared Micro-Mobility

TL;DR

This paper tackles the challenge of high-resolution, uncertainty-aware short-term demand forecasting for dock-based bike-sharing. It proposes T-STAR, a two-stage spatial-temporal adaptive contextual representation built on a Time Series Transformer, which first models hourly demand patterns and then refines predictions with fine-grained signals including metro deviations, yielding probabilistic forecasts via a Negative Binomial distribution. Across a Washington, D.C. case study with 235 stations, T-STAR demonstrates superior deterministic and probabilistic accuracy, robustness across stations and time, and strong zero-shot generalization to unseen service areas. The framework’s hierarchical design and rich contextual integration enable accurate, scalable, and uncertainty-aware forecasts that can support proactive fleet rebalancing and multimodal trip planning in MaaS ecosystems.

Abstract

Reliable short-term demand forecasting is essential for managing shared micro-mobility services and ensuring responsive, user-centered operations. This study introduces T-STAR (Two-stage Spatial and Temporal Adaptive contextual Representation), a novel transformer-based probabilistic framework designed to forecast station-level bike-sharing demand at a 15-minute resolution. T-STAR addresses key challenges in high-resolution forecasting by disentangling consistent demand patterns from short-term fluctuations through a hierarchical two-stage structure. The first stage captures coarse-grained hourly demand patterns, while the second stage improves prediction accuracy by incorporating high-frequency, localized inputs, including recent fluctuations and real-time demand variations in connected metro services, to account for temporal shifts in short-term demand. Time series transformer models are employed in both stages to generate probabilistic predictions. Extensive experiments using Washington D.C.'s Capital Bikeshare data demonstrate that T-STAR outperforms existing methods in both deterministic and probabilistic accuracy. The model exhibits strong spatial and temporal robustness across stations and time periods. A zero-shot forecasting experiment further highlights T-STAR's ability to transfer to previously unseen service areas without retraining. These results underscore the framework's potential to deliver granular, reliable, and uncertainty-aware short-term demand forecasts, which enable seamless integration to support multimodal trip planning for travelers and enhance real-time operations in shared micro-mobility services.

Figures (11)

• Figure 1: Pickup and drop-off demand forecasting pipeline of Two-stage Spatial and Temporal Adaptive contextual Representation (T-STAR).
• Figure 2: Illustration of the generation process of intermediate demand variational signals.
• Figure 3: Architecture of a Transformer-based probabilistic forecasting model. Note: For one-step forecasting ($H=1$), the decoder branch is inactive, and the encoder output is passed directly to the distribution head.
• Figure 4: Hourly average bike pickup and drop-off demand (Washington D.C., Oct–Dec 2022).
• Figure 5: Distribution of zero-demand percentages in the historical 15-minute intervals across 235 bike-sharing stations in Washington D.C case study.