The Ensemble Kalman Filter (EnKF) is a Monte Carlo-based sequential data assimilation method that approximates the Kalman filter by representing the probability distribution of a system's state through an ensemble of sample trajectories, updating each ensemble member as new observations become available. In the context of traffic state estimation and agent-based pedestrian modeling, EnKF is particularly valuable because it can handle high-dimensional, nonlinear dynamical systems without requiring explicit computation of covariance matrices, making it computationally tractable for large-scale simulations where traditional Kalman filtering is infeasible. Key variants include the Ensemble Kalman Smoother, which incorporates future observations for retrospective state correction, and ensemble square-root filters, which avoid perturbing observations to reduce sampling error, with the method frequently appearing as a component within physics-informed and hybrid deep learning frameworks to fuse model-based priors with sparse sensor measurements.
Dictionary term