Study of nonlinear parameter identification using UKF and Maximum Likelihood method

The nonlinear parameter identification is studied using UKF and Maximun Likelihood (ML) method. The proposed scheme consists of two sequential stages. The first stage conducts the state estimation using UKF, where the estimated state is a function of unknown parameters. A likelihood function is constructed in the second stage based on the estimated state. Thereby, the parameter identification problem becomes an optimization of the parameterized likelihood function. The proposed method is further compared with EKF based approach. Several case studies show a clear benefit using UKF instead of EKF based approach for a class of nonlinear identification in terms of precision and fast convergence.