A simplified functional link net architecture for dynamic system identification with a UKF algorithm

This paper presents uncented Kalman filter technique for identification of non linear dynamic systems. A novel unscented Kalman filter (UKF) has been proposed that has the advantage over EKF since it does not use linearization for state prediction and covariance´s and costly calculations of derivatives. This leads to an accurate computation of Kalman gain and error covariance matrices which ultimately leads to an accurate identification of the system. The approach is shown to exhibit robustness characteristics and fast convergence property. A simulation example dealing with applications of the proposed algorithm is given.