A modified adaptive filtering algorithm for identification of a dynamic system

This paper presents a modified adaptive Unscented Kalman filter for parameter estimation of nonlinear dynamic systems by analysis of input and output signals. A wide variety of artificial neural network (ANN) methods are used for this purpose. But the selection of appropriate model, and number of neural nodes made it difficult. This paper proposes a model based approach for dynamic system identification, which utilizes Unscented kalman filter (UKF) having multi fold advantages over other adaptive filters. To achieve better accuracy in estimation, the model and measurement error covariance of the UKF are made adaptive. Hence a self tuning approach for updating error covariance is proposed, which utilizes innovation error present in the model with a learning factor inversely proportional to an average value of the covariance. Various simulations have been carried out to show the superiority of the proposed algorithm. The mean square error obtained using this approach is found to be significantly less than other conventional methods.