Online black-box model identification and output prediction for sampled-data systems

Abstract

In this work, black-box model identification and output prediction for unknown sampled-data minimum phase system has been achieved. Feedforward neural network (multilayer perceptron) is used for system identification. Unscented Kalman Filter (UKF) online determine weights of neural network and predicts output in open-loop sampled-data configuration. Magnetic levitation and DC motor model has been identified in computer simulations using the presented black-box identification and prediction scheme.