Lithofacies prediction in non-cored wells from the Sif Fatima oil field (Berkine basin, southern Algeria): A comparative study of multilayer perceptron neural network and cluster analysis-based approaches

Abstract

The purpose of this study is to investigate the possibility of applying multilayer perceptron neural network (MLPNN) and cluster analysis approaches for rebuilding non-cored lithofacies. These techniques are carried out to predict missing lithofacies intervals from the reservoir of Trias Argileux Grèseux Inférieur in the Sif Fatima oil field (Berkine basin- Southern Algeria). The performances of the suggested models were evaluated using root mean square error (RMSE), mean absolute error (MAE) and correlation coefficient (R). The MLPNN model was developed using four input variables of nuclear well logging data including: Gamma-Ray, Density, Potassium and Thorium. MLPNN model shows lower RMSE and MAE with 0.39 and 0.23 respectively, together with strong R values (training = 0.87; validation = 0.78; test = 0.92). The cluster analysis model displays lower performances (R = 0.68, RMSE = 1.04 and MAE = 0.54). This quantitative comparison between real and predicted electrofacies using the two methods indicates that MLPNN model is more recommended in rebuilding non-cored lithofacies than cluster analysis. The MLPNN model makes possible to estimate lithofacies in 333 m of non-cored lithofacies allowing an important economic benefit.