Identification of Non-Linear Dynamic systems applying Unscented H-Infinity Adaptive Filtering approach

This study proposes an adaptive filtering approach for nonlinear system identification employing a functional link artificial neural network (FLANN) including the extended H infinity filter (EHF) and the unscented H infinity filter (UHF). By identifying the system and state model based on the FLANN model's structure and estimating its parameters, the nonlinear system is identified. The desired output of the system is compared with original and yields the mean square error (MSE) using EHF and UHF. The performance is compared with extended Kalman filter (EKF), unscented Kalman filter (UKF) and many state of art methods and it is found that UHF is a better performer in terms of MSE. By incorporating the unscented transform principle into the EHF technique, the linear statistical error propagation strategy may be involved to perform the unscented H-infinity filtering. The suggested strategy is resistant against model uncertainty and can produce better accurate identification results with less sophisticated processing. Its efficiency is demonstrated by three simulated cases.