Filter for Traffic Congestion Prediction: Leveraging Traffic Control Signal Actions for Dynamic State Estimation

The field of intelligent transportation systems is rapidly evolving, with increasing focus on addressing traffic congestion, a pervasive problem in urban environments. This study contributes to this domain by enhancing traffic prediction models. Traditional traffic models often fall short in accurately predicting traffic flow, particularly in dynamic urban settings. This limitation necessitates the development of more adaptive and accurate predictive models to manage traffic congestion effectively. In urban environments, traffic estimation at multi-intersection nodes and connecting roadis approached using statistical filters (Kalman or particle filters). The challenge lies in accurately predicting the vehicular traffic state at intersections, considering the dynamic nature of urban traffic and potential communication failures. This study introduces a novel approach in traffic modeling using statistical filters. The method involves the use of dynamically interchangeable state transition matrices, aligned with specific traffic signal control actions. This allows for more precise predictions under varying traffic conditions and control scenarios. The effectiveness of the proposed models is validated across multiple intersections in urban settings. The evaluation focuses on the models’ scalability, adaptability to complex traffic networks, and robustness in communication failure scenarios. Our proposed Kalman filter showcased superior performance in traffic prediction with an average RMSE of 12.15975, demonstrating a significant improvement over the average measurement RMSE of 13.996875 and over traditional smooth Kalman filter and traditional particle filter.