Optimización De Pórticos Sismorresistentes Utilizando Redes Neuronales Y Algoritmo Sin Cálculo De Gradientes

Abstract

An optimization process of earthquake-resistant structures is presented, which means to
meet values of a set of design parameters that satisfy prescribed probabilities of annual excedance or
reliability indices, and to minimize an objective function like the total weight or cost. The major
computational effort is the structural non linear dynamic analysis to obtain the response parameters,
taking into account the uncertainties, for a set of earthquake records likely to occur at a site. The
reliability evaluation is done by direct simulation of Monte Carlo, efficiently due to the approximation
of the structural response by a “substitute”, in this work neural networks that show more flexible and
efficient representation. The optimization process is carried out by a search algorithm, gradient free.
The presented examples show the power and advantages of these numerical techniques.