engleski

Application of artificial neural networks for lithofacies determination in absence of sufficient well data

Lithofacies definition in the subsurface is an important factor in modelling, regardless of the scale being on the reservoir or basin level. In areas with low exploration level, modelling of lithofacies distribution presents a complicated task as very few inputs are available. For this purpose, a case study in Požega Valley vas selected with only one well and several seismic profiles on an area covering roughly 850 km2. A well to seismic correlation of lithofacies was performed with three different neural networks - multi-layer perceptron, radial-basis function and probabilistic neural network. Variables for the analysis were lithofacies determined from well logs, cores and cuttings, and twelve seismic attribute data. Lithofacies as the output variable was adequately presented in numerical values as a categorical variable and predictions on it were tested through a process of artificial neural network training in analytics software StatSoft Statistica. All of the neural networks were successful in the training process. However, the probabilistic neural network showed better results than other two. By extrapolating the lithofacies data onto the seismic profile, a greater input dataset for facies modelling was obtained. In this way, an application of neural networks in early phase of exploration has been confirmed for better definition of lithofacies distribution in the subsurface.

artificial neural networks, Croatia, facies prediction, Pannonian Basin, Požega Valley

nije evidentirano