A digital twin-based multi-horizon passenger demand forecasting framework for public transits

Accurate and reliable passenger demand prediction is crucial for optimizing operations and enhancing the performance of intelligent transportation systems. However, multi-horizon demand forecasting remains challenging as ridership patterns exhibit strong temporal variability and complex spatial interdependencies across routes. Existing forecasting approaches often focus on isolated time series models or short-term horizons, limiting their ability to capture network-wide demand propagation and support proactive operational decision-making. To address these challenges, this study proposes an integrated digital twin framework for public transit built upon the FIWARE ecosystem and next-generation service interfaces-linked data (NGSI-LD) context management standards. The framework consolidates various data sources into a unified, interoperable environment, enabling real-time monitoring and analytics. Within the digital twin, a diffusion convolutional recurrent neural network (DCRNN) is deployed to learn spatio-temporal dependencies and predict passenger demand across multiple horizons (3, 12, and 24 h ahead) for every route in the network. By explicitly capturing spatial correlations between routes that share common stops or segments, the proposed approach captures demand propagation patterns that conventional forecasting methods fail to represent. Experimental results demonstrate that the proposed approach outperforms a standard DCRNN implementation by achieving a lower mean absolute percentage error and explaining up to 94.1% of the variance in passenger demand at the 3-hour horizon. Beyond forecasting accuracy, the integration with FIWARE NGSI-LD-based context management enables operational monitoring, proactive alerts, and what-if service scenario analysis within a live operational environment. Overall, this research demonstrates how spatio-temporal graph-based forecasting with digital twin technology can support data-driven decision-making in public transport operations, bridging the gap between predictive analytics and real-world transit management.