Traffic state estimation and prediction based on Bayesian approach in urban road networks using AVI and floating vehicle data

Accurate estimation and prediction of traffic state is crucial for the development of intelligent transportation systems. However, existing studies focus on traffic state at the links or intersections, and under-saturated scenarios, limiting their applicability at the network. This study proposes a framework for network-scale traffic state estimation and prediction by integrating trajectory and AVI data. The framework includes queue length estimation for different links equipped with AVI systems and state estimation and prediction for unobserved links. Validation uses large-scale real-world and simulation datasets. Results show that, with real-world data, the MAE for queue length and travel time estimation are less than 0.69 vehicles and 1.35 s, respectively, with prediction MAE around 1 vehicle. In simulations, the proposed method outperforms benchmarks under various demands, achieving queue length MAE of 2.43 vehicles under high demand. These findings indicate high accuracy in both estimation and prediction, suitable for under-saturated and over-saturated conditions.