Extended and unscented kalman filters for artificial neural network modelling of a nonlinear dynamical system

摘要: Recently, artificial neural networks, especially feedforward have been widely used for the identification and control of nonlinear dynamical systems. However, determination a suitable set structural learning parameter value feed-forward networks still remains difficult task. This paper is concerned with use extended Kalman filter unscented based training algorithms. The comparisons performances both algorithms are discussed illustrated using simulated example. simulation results show that in terms mean squared errors, algorithm superior to back-propagation since there improvements between 2.45–21.48% (for training) 8.35–29.15% testing). indicates could be good alternative network models applications