Applicable filtering framework for online multiclass freeway network estimation

Real-time traffic flow estimation is important for online traffic control and management. The traffic state estimator optimally matches traffic measurements from detectors with traffic flow predictions from a dynamic traffic model under a certain control strategy. The current and widely used estimator is based on the Extended Kalman Filter algorithm (EKF). Basically, EKF is developed from the recursive Bayesian estimation technique for Gaussian random distribution of the state. This approximation may result in large errors in the estimation and even lead to divergence of the filter in highly non-linear dynamic system such as heterogeneous traffic flow operations. The aims of this paper are therefore twofold. On the one hand, we present a generalized stochastic macroscopic traffic model for multiclass freeway networks. The model is developed in the form that can be applied by filtering methods. On the other hand, we implement an accurate probabilistic framework to the real-time multiclass freeway network estimation. The framework uses a variation of Kalman Filter, namely Unscented Kalman Filter, and a different filter that is based on a sequential Monte Carlo method, namely Unscented Particle Filter. We investigate the performance of the proposed framework with respect to accuracy and computational effort using real-life data collected in a freeway network in England. We expect that the developed tool is useful for traffic operators and planners in controlling large-scale multiclass freeway networks.