Ensemble unscented Kalman filter for state inference in continuous–discrete systems

The authors consider non‐linear state filtering problem in continuous–discrete systems, where the system dynamics is modelled by a stochastic differential equation, and noisy measurements of the system are obtained at discrete time instances. A novel particle method is proposed based on sequential importance sampling. This approach uses a bank of the continuous–discrete unscented Kalman filters (CDUKFs) to obtain the importance proposal distribution, retaining the advantage of the CDUKF in continuous–discrete systems as well as the accuracy of particle filter in highly non‐linear systems. Simulation results show that the algorithm outperforms some other benchmarks substantially in estimation accuracy.