Predict chaotic time-series using unscented Kalman filter

Although the extended Kalman filter is a widely used estimator for nonlinear systems, it has two drawbacks: linearization can produce unstable filters and it is hard to implement the derivation of the Jacobian matrices. This work presents a new method of predicting Mackey-Glass equation based on unscented Kalman filter. The principle of unscented transform is analyzed and the algorithm of UKF is discussed And then EKF and UKF methods are used to estimate the noisy chaotic time-series, and the estimation errors between two different algorithms are compared. Simulation results show this filter can predict chaotic time-series more effectively and accurately than extended Kalman filter.