Integration of Rock Typing and Neural Network Techniques for Accurate Permeability Prediction in Heterogeneous Carbonate Reservoirs: A Case Study from Abu Dhabi Offshore Field

Permeability prediction or calculation in heterogeneous carbonate formations is a challenging task due to the complexity of the rock properties and pore systems that are difficult to characterize accurately. In this study an innovative and efficient approach developed and used to overcome this challenge by combining rock typing and machine learning neural network (MLNN) techniques to accurately predict the permeability of heterogeneous carbonate formations. The supervised machine learning approach based on a neural network algorithm trained on a large dataset of offset wells data. The input data for the algorithm included various rock properties from core and well logs analysis data. The output of the algorithm was a prediction of rock types and their corresponding permeability values based on the porosity and permeability equations. The MLNN model was trained using a backpropagation algorithm and validated using an independent dataset to ensure the accuracy and reliability of the results. The trained model was then used to predict rock types and permeability values for new wells. The results of the study showed that the approach of using rock typing and machine learning neural network outperforms other traditional methods in predicting permeability in heterogeneous carbonate formations. The predicted permeability values were validated against actual measurements from core data and formation testing mobility, and the results showed a good correlation between predicted and measured values and demonstrating the model reliability. The use of rock typing provides a more accurate characterization of the reservoir and helps to improve the prediction of permeability. The study also revealed that the rock types and permeability values varied significantly across the carbonate formation, and the neural network model was able to capture this heterogeneity accurately by learn the complex relationships between the rock types and petrophysical properties, which resulting in improved permeability predictions. The predicted permeability values were used to generate permeability maps that helped identify areas with higher permeability values, which can be targeted for well placement to improve hydrocarbon recovery. This approach lies in the integration of rock typing and machine learning neural network to predict permeability in heterogeneous carbonate formations. This method provides an innovative solution to the challenges associated with traditional methods, which often fail due to the complex nature of carbonate reservoirs. The approach is applicable to a wide range of carbonate formations, and has the potential to significantly improve reservoir characterization and production optimization.