Predicting passenger flows in rail transport systems plays an important role for traffic management centers to make fast decisions during service disruptions. This paper presents an innovative cross-disciplinary approach based on digital twins, deep learning, and traffic simulation to predict the total number of passengers in each train stations and evaluate the impact of service disruptions across stations. First, we present a four-layer system architecture for building a digital twin which ingests real-time data streams, including train movements and timetable scheduling. Second, we deploy several deep learning models to predict the total number of passengers in each station using mobile data. The results showcase significant accuracy for recurrent versus non-recurrent traffic conditions even under severe large disruptions such as the COVID-19 travel restrictions. Our case study of the Sydney rail network demonstrates that the proposed digital twin powered by deep learning can provide more granular real-time insights into the impact on passengers, allowing rail operation centers to better mitigate service disruptions.

This research showcases the innovative integration of Large Language Models into machine learning workflows for traffic incident management, focusing on the classification of incident severity using accident reports. By leveraging features generated by modern language models alongside conventional data extracted from incident reports, our research demonstrates improvements in the accuracy of severity classification across several machine learning algorithms. Our contributions are threefold. First, we present an extensive comparison of various machine learning models paired with multiple large language models for feature extraction, aiming to identify the optimal combinations for accurate incident severity classification. Second, we contrast traditional feature engineering pipelines with those enhanced by language models, showcasing the superiority of language-based feature engineering in processing unstructured text. Third, our study illustrates how merging baseline features from accident reports with language-based features can improve the severity classification accuracy. This comprehensive approach not only advances the field of incident management but also highlights the cross-domain application potential of our methodology, particularly in contexts requiring the prediction of event outcomes from unstructured textual data or features translated into textual representation. Specifically, our novel methodology was applied to three distinct datasets originating from the United States, the United Kingdom, and Queensland, Australia. This cross-continental application underlines the robustness of our approach, suggesting its potential for widespread adoption in improving incident management processes globally.