| dc.description.abstract | The techniques of geophysical logging and core descriptions, used on lithology identification, represent a high financial cost and involve a considerable amount of time from a specialist. On that direction, the main objective of this research is to propose an alternative method of lithological classification, through Artificial Neural Networks (ANNs), to assist the process of geophysical data interpretation. The study area is located in Leão coal field, where a major part of its territory is inside the municipalities of Rio Pardo, Minas do Leão and Butiá (RS). The set of ANN training and validation contain information of eight boreholes coming from Palermo and Rio Bonito formations. The input variables include depth data and geophysical information of gamma-ray profiles, spontaneous potential, resistance and resistivity. For all experiments, the lithologies to be classified were: sandstone, silt and coal. The neural network model utilized was feedforward multilayer perceptron (MPL). Networks were trained by Levenberg-Marquardt and Resilient backpropagation algorithms. A success rate of approximately 80% was obtained on classification. | |
