Batch-mode identification of black-box models using feedforward neural networks

Feedforward neural networks are used for the purpose of black-box modeling. The optimization of the network parameters (i.e., the weights) is accomplished using a recursive batch-mode algorithm that is based on the minimization of a cost function. The cost is the summation of two quadratic contributions: a fitting penalty term and a term related to changes in the parameters, which can be suitably emphasized or, on the contrary, de-emphasized by choosing a proper scalar. Simulation results are reported to confirm the effectiveness of the algorithm.