Aplicación de modelos de redes neuronales artificiales a la determinación de movimientos en una presa bóveda.

Abstract

The complexity of the dam-foundation makes difficult to interpret the records of auscultation. It is not easy to make a prognosis of the behavior of the dam in accordance with the different situations that may occur over the life of the dam. N or is it easy to interpret the deviations observed from the values estimated by numerical or statistical modeling. Neural networks offer the possibility of treating the whole dam foundation as a complex system which rules of behavior are established solely on the basis of the observed behavior without making simplifying assumptions. It has shaped the radial movement in different seasons of the pendulums available in a dam vault, which has been used as test case. We analyzed the response of the dam complex models growing. T he simplest considered only as pa r ame t e rs determining the behavior reservoir level and a moving average of ambient temperatures (simple model). In order to model applicable to any pendulum of any dam, has raised the consideration of various means mobile temperature, so that is the model itself that determines the weight of each moving average (model general). Have also been raised as da ta models that incorporate the values of the movements measured immediately prior (short-term models). The comparison of results obtained through each of the previous models of neural networks with statistical and numerical models commonly used that allows the use of neural network models useful for the interpretation of auscultation. The submission details the results obtained and the accuracy of the prognosis of each of the models studied. Models are expected to nalyze incorporate additional time variable (dynamic models), to c apture the effect of drift. This research is pa rt of the R&D financed by the Ministerio de Ciencia e Innovación: estudio de la seguridad de presas e identificación de escenarios de riesgo mediante sistemas inteligentes, reference number 048/RN08/04.5.