Research and/or engineering questions / objective This work addresses the pressing need of the local public transport organization in Genoa (Italy) to tackle overcrowding on the metro line and improve service levels. A multi-method simulation approach has been developed to forecast crowding levels and train frequencies in different scenarios, while Machine Learning algorithms are also exploited to enhance the accuracy of predictions. The primary goal is to create a valuable tool that can optimize the metro service and enhance the passenger experience. To achieve this, real-time data are utilized to provide accurate forecasts and support the public operator to make informed decisions to improve service efficiency. Methodology The work has developed an advanced agent-based simulator based on the AnyLogic framework. The simulator employs a discrete-event approach to schedule events and ensure their accuracy, which is validated and calibrated using historical data. To predict future crowding levels, the simulation results are used to train Machine Learning algorithms that can identify patterns. Real-time data collection is enabled through a data collector that transmits sensor data to the model, allowing for real-time simulation and monitoring of the station. The model's accuracy in simulating system behavior and predicting future crowding levels is assessed to evaluate its performance. In summary, the methodology involved developing a complex agent-based simulator, integrating a discrete-event approach for scheduled events, training machine learning algorithms for prediction, and connecting the model to a data collector for real-time monitoring and simulation. Results The outcome of the work is a cloud-based simulation model that is connected to a mobility platform sharing real-time data from sensors and cameras in the metro of Genoa, Italy. This tool can provide improved awareness of demand levels and offerings, enabling transport operators to optimize service levels and make informed decisions. The simulation model also supports the prediction of crowding levels and train frequencies under different scenarios, with improved accuracy through the use of Machine Learning algorithms. Moreover, the simulation results can influence passenger behavior, such as suggesting alternative modes of transportation when metro stations or trains are overcrowded, thus enhancing passenger experience and reducing overcrowding. This capability of accurate prediction and influencing passenger choices can significantly enhance the public transport service in Genoa, providing a valuable tool for transport operators to optimize service efficiency and improve passenger experience. Limitations of this study The limitations of the work are well known and relate to the practical application of the findings for the public transport operator. An ideal solution would be to modify train frequency in real-time based on demand, but this is unfeasible due to technical reasons such as driver availability and training time. Another limitation is the challenge of influencing passengers’ behavior. One possible solution would be to send notifications via the mobile app to alert passengers about current crowding levels, although implementing such measures would require a significant shift in passenger behavior. What does the paper offer that is new in the field in comparison to other works of the author? The work reported in this paper offers a novel approach to modelling and predicting crowd behavior and train frequency in a station using a multi-method simulation approach and Machine Learning algorithms. The integration of real-time data from sensors and cameras in the station allows for accurate and up-to-date predictions, which can be leveraged to optimize the metro service and enhance passenger experience. This approach may be unique in comparison to other studies in the field that may focus on different aspects of metro transportation, such as network optimization or passenger flow analysis. Conclusion This work presents a novel approach that utilizes simulation, real-time data from sensors and cameras to predict crowding levels and train frequencies, offering a valuable decision-making tool for public transport operators. The project showcases the potential of simulation and Machine Learning to optimize transportation services and enhance passenger experience. The limitations identified can potentially be addressed by implementing real-time train scheduling. In summary, this work offers a practical solution for managing passengers in transportation hubs and improving the efficiency of transportation services.
Mr. Pietro De Vito, Project Manager, STAM