Efficient Traffic State Estimation and Prediction Based on the Ensemble Kalman Filter with a Fast Implementation and Localized Deterministic Scheme

Traffic state estimation and forecasting are central components in dynamic traffic management and information applications. This paper proposes a traffic state estimation approach based on an improved formulation of the traditional Ensemble Kalman filter (EnKF), including a fast implementation and a localized deterministic scheme. A reformulation of the EnKF equations leads to efficient computation. The deterministic scheme implies that we use the same observations for each of the ensembles instead of randomized observations. The use of a deterministic algorithm can reduce the impact of coincidental sampling and associated sampling errors. Localization is in contrast with a global method. In the global method, both the evolution of system states and the incorporation of observations are considered as an entity (within a global matrix). Here, the inclusion of localization has several potential advantages for large-scale applications: blocking spurious correlations, decreasing computation time due to smaller matrix inversions, increasing the accuracy by increasing the effective ensemble size. The proposed implementation of the EnKF for traffic state estimation and prediction is tested and validated in a realistic Dutch freeway network. The experiment studies deliver promising results for large-scale practical applications.