Artificial Neural Networks in Evaluating Piezometric Levels at the Foundation of Itaipu Dam

Abstract

Itaipu Dam is an engineering work of high importance. Located at the border between Brazil and Paraguay in the Paraná River and with approximated geographical coordinates 25°24′29″S, 54°35′21″W, it feeds these two countries with electrical energy and has to be constantly monitored in order to maintain its levels of quality and security. Over two thousand instruments, including more than 650 piezometers, have been installed for the monitoring of the dam and they provide continuous data about several characteristics of its foundation and structure. The evaluation of piezometric levels in dams is important for it reflects the values of the uplift pressure that acts on the structure of the dam. The utilization of new methods in such an analysis can provide agility to decisions-taking by the security team of the dam. Depending on the method applied, a better comprehension of the phenomenon in time and space may be achieved. This study employs Artificial Neural Networks (ANN) to simulate the behavior of the piezometers installed in a geological discontinuity in the foundation of Itaipu Dam. It considers different types of entry data in a Multilayer Neural Network and determines the best ANN architecture that is closest to the real situation. Some parameters have a higher weight in the variation of the piezometric levels, whereas some others do not affect it considerably. A geological geotechnical model of the foundation rock fractures would be helpful to improve the entry data and achieve better results.