Identification of Nonlinear Systems Using Neural Networks and Polynomial Models: A Block-Oriented Approach