Utilización de redes neuronales para la determinación de respuestas de sitio a partir de ondas coda: Aplicación para Armenia, Colombia.

Triaxial acelerografic data related to replies of Armenia earthquake in Colombia (25-01-99, Mw=6.2) have been used to determine the transfer function of Soft Formation / Hard Rock in two sites of Armenia city by using a neuronal network. This network was trained with spectral data of HARD ROCK as input parameter and spectral data of Soft Formation as output parameter for N-S and Z components. The error of the neuronal network respect to the training data was of 0,01%. It allowed to simulate spectral data for the E-W component of registries with a quadratic mean error minor to 8%. It was possible to observe that with these few records used, the neuronal networks facilitate the process of deduction of transference functions linear and not-linear oriented to the dynamic ground evaluation. [ABSTRACT FROM AUTHOR]