Identification structures using rational wavelets: Examples of application Original Research Article

In this paper we present three examples that show the applications of a black-box identification structure already defined. This structure can be described as a concatenation of a mapping from observed data to a finite set of linear filters realized using rational wavelets, and a nonlinear mapping from the output of the linear dynamic part to the system output represented by a hidden layer perceptron neural network, or a basis (that might be orthonormal) of high level canonical piecewise linear functions. The wavelets used for identifying the linear dynamic part are selected taking into account the linear dynamics of the system and consequently they can be considered as semiphysical regressors. Also, this structure allows to approximate the dynamic evolution of any nonlinear, causal, time-invariant system with fading memory.