Appendix C : Matlab code for linearization

A general approach for system identification with artificial neural networks is presented. A dynamic process can be represented as a function of the evolution of the inputs/outputs of the system. The standard approach to system identification is to describe the system by a set of equations that take into account the underlying mechanisms of the studied plant. If the equations of the system are of complex non-linear nature the system can be approximated by a set of linear equations with unknown parameters. Then, the unknown parameters are estimated. Due to the non-linearity of the system and data quality, in many cases this approach might fail. An alternative approach that does not require a priori information about the system is to use artificial neural networks (NN) [2]. The traditional feedforward neural network can be regarded as a universal approximator. In this work, a close-loop neural network architecture based on multilayer feed-forward network is presented as an alternative approach to standard system identification. The adaptations of the NN weights were done by an improved error backpropagation algorithm. The algorithm automatically selects the number of neurons in the hidden layer and provides an educated guess of the initial weight values. A comparison between the NN identification approach and the linear Autoregressive Moving Average with exogenous input (ARMAX) technique will be presented. The identified model was simulated with the SIMULINK package. The NN identification approach can be easily extended to multi-input multi-output (MIMO) dynamic system.