Conrprop: an algorithm for nonlinear optimization with constraints

Resilent Backpropagation is a gradient-based powerful optimization technique commonly used for training artificial neural networks, which is based on the use of a velocity for each parameter in the model. However, although this technique is able to solve unrestricted multivariate nonlinear optimization problems there are not references in the operations research literature. In this paper, we propose a modification of Resilent Backpropagation that allows us to solve nonlinear optimization problems subject to general nonlinear restrictions. The proposed algorithm is tested using six common used benchmark problems; for all cases, the constrained resilent backpropagation algorithm found the optimal solution and for some cases it found a better optimal point that the reported in the literature.