Adaptive Metro Passenger Flow Prediction

The rapid expansion of metropolitan regions has intensified reliance on metro systems as the backbone of urban mobility. Accurate passenger flow prediction is essential for efficient scheduling, congestion management, and resource allocation. However, traditional statistical and survey-based methods fail to capture the complex, nonlinear, and dynamic patterns of metro ridership influenced by temporal, spatial, and external contextual factors. This paper presents a Metro Passenger Flow Prediction System using Adaptive Feature Fusion Networks (AFFNs), designed to integrate heterogeneous data sources such as historical travel records, weather conditions, holidays, and special events. The framework employs an Enhanced Multi-Graph Convolution with Gated Recurrent Units (EMGC-GRU) to model spatial dependencies across stations and temporal passenger flow trends. Additionally, a multi-task AFFN jointly predicts both Origin-Destination (OD) and Inflow-Outflow (IO) flows, mitigating sparsity in OD matrices and improving accuracy. The system is implemented with Python, Django-ORM, MySQL, and a web-based interface for real-time visualization and analytics. Experimental evaluation demonstrates significant improvements in forecasting accuracy and robustness, highlighting the model’s potential to enhance operational efficiency, passenger experience, and the development of smart city transportation solutions.