Ultra-Flexible, Explainable, and Scalable Traffic Prediction with Dynamic Future Routes

Traffic forecasting is essential for intelligent transportation systems, aiming to predict future traffic dynamics such as speed and travel time through the analysis of past observations. However, mainstream deep learning frameworks, which rely heavily on historical data, often struggle in realworld applications due to their inadaptability to dynamic future changes, neglect of future traffic flow as the root cause of traffic conditions, and the complexity of model structures for city-scale road networks. To solve these limitations, we propose a Route Data Management System (RouteSys) that integrates a macroscopic simulation module with lightweight traffic prediction models to estimate the future traffic conditions on individual road segments by accurately and efficiently simulating vehicle travel sequences and traffic states in advance. Additionally, we integrate the microscopic traffic simulation tool SUMO with the custom route planning logic to generate synthetic route data, supporting model training and application evaluation. RouteSys has been validated on real-world road networks in various scenarios, showing substantial improvements in prediction accuracy, efficiency, and scalability compared to the mainstream structures.