Discriminating Shale Layers by Pseudo CGR Logs Created Using Artificial Intelligence

The discrimination of shale vs. non-shale layers significantly influences the quality of reservoir geological model. In this study, a novel approach was implemented to enhance the model by creating Pseudo Corrected Gamma Ray (CGR) logs using Artificial Intelligence methods to identify the thin shale beds within the reservoir. The lithology of the carbonate reservoir understudy is mostly composed of dolomite and limestone rock with minor amounts of anhydrite and thin shale layers. The identification of shale layers is challenging because of the nature of such reservoirs. The high organic content of the shales and the presence of dolomites, particularly the floatstones and rudstones, can adversely affect the log quality and interpretation and may result in inaccurate log correlations, overestimating/ underestimating Original Oil In Place (OOIP) and reservoir net pays. In such cases, Corrected Gamma Ray (CGR) curves are typically used to identify shale layers. The CGR curve response is due to the combination of thorium and potassium that is associated with the clay content. The difference between the total GR and the CGR is essentially the amount of uranium-associated organic matter. Because of the very limited number of CGR logs in this reservoir, Artificial Intelligence (AI) approach was used to identify shale volume across the entire reservoir. Synthetic CGR curves were generated for the wells lacking CGR logs using AI methods. Resistivity, Density, Neutron and total GR logs were used as inputs while CGR was set as the target. Five wells that have CGR logs were used to train the model. The created pseudo logs were then used to identify shale layers and could also be used to correct effective porosity logs. After statistical analysis of the data, two different Artificial Intelligence Techniques were tested to predict CGR logs; Adaptive Neuro-Fuzzy Inference System (ANFIS) and Artificial Neural Network (ANN). A Sugeno-type FIS structure using subtractive clustering demonstrated the best prediction with correlation coefficient of 0.96 and mean absolute percentage error (MAPE) of 20%. The resulting synthetic CGR curves helped identify shale layers that do not extend over the entire reservoir area and ultimately correct the effective porosity logs in the reservoir model. Porosity was primarily obtained from the neutron-density logs which results in very high porosity measurements across the shale layers. This study shows a new workflow to predict shale layers in Carbonate reservoirs. The created pseudo CGR logs would help predict shale and is an added-value data that could be incorporated into the Earth model.