Adaptive Traffic Congestion Prediction Using LSTM Networks and Physics-Based Models

Abstract

Accurately modelling and predicting traffic congestion is essential for effective traffic management in dynamic urban environments. This study introduces a hybrid model that combines a deep neural network architecture, designed to capture long-term temporal dependencies, with physics-based approaches to traffic modelling. The deep learning component identifies complex temporal patterns and non-linear behaviours in traffic evolution, while the physics-based component incorporates dynamic constraints that enhance the model’s robustness and ability to generalise. This integration leverages the predictive power of machine learning while maintaining consistency with the fundamental principles governing vehicle flow. Experiments conducted using real-world traffic data demonstrate strong predictive performance, establishing a solid foundation for the development of intelligent transportation management systems and providing advanced tools to support adaptive mobility optimisation in congested urban settings.