Abstract

LOPEZ-AGUIRRE, Daniel; GARCIA-BENITEZ, Silvia Raquel; NICOLAS-LOPEZ, Rubén  and  COCONI-MORALES, Enrique. Petrophysical parameterization of sand-clay siliciclastic sequences with neural networks. Bol. Soc. Geol. Mex [online]. 2023, vol.75, n.3, e150823.  Epub May 28, 2024. ISSN 1405-3322.  https://doi.org/10.18268/bsgm2023v75n3a150823.

In this work neural networks are used as an advantageous tool to estimate petrophysical parameters of the stratigraphic column traversed by several wells. The parameters porosity, mineral volumes, and water and hydrocarbon saturation are obtained from basic geophysical well logging (gamma rays, deep resistivity, volumetric density and transit time) and are inferred for other sites, in the same geological area, where they are not measured, so this information matrix is not available. This analysis was performed on sand-clay siliciclastic sequences traversed by several wells drilled to reach a low-permeability hydrocarbon reservoir. Estimates with empirical models are presented to compare them with those obtained with neural networks in order to qualify the performance of the intelligent alternative. The laws that govern the dynamics of the parameters as well as the details of the geological context are immersed in the weights of the network and the phenomenological consistency is defined through the congruence of the inputs to achieve the chosen outputs. The way in which the neural model enables the reliable propagation of property values is shown and becomes an advantageous auxiliary in the study of very complex or poorly parameterized geological contexts in which the conditions for the application of correlations and empirical methods as well as how the time invested in the processes of adjustment and contextualization of records, decreases the quality and quantity of knowledge obtained about the environment.

Keywords : petrophysical models; well logs; siliciclastic sequences; neural networks.