ABSTRACT
In non linear systems identification or modeling there is always the problem of finding the best test signal to be used in order to make the identification in a correct way. It’s obvious that in linear systems a signal rich in frequencies is good enough but this is not the case for non linear systems. We can state that the best test signal will be the one that is able to make the system evolve through the whole working region of the state space. A method is presented that using artificial neural networks to identify non linear systems is able to optimize its own test signal to be used for system modeling and identification. As the definition of the test signal can not be done without a knowledge of the system behavior, an iterative process uses the partially learned system dynamics to improve the test signal. The same neural network that identifies the system will be used to generate the test signal for the next training phase. At every iteration the state space region already identified grows till it fills the predefined working area.
