Perpustakaan Digital ITB

Critical property data of the reservoir, including formation data, facies, porosity, permeability, and water saturation, are analyzed as a reference for reservoir characterization through data recorded by wireline logs and core samples. However, with the limited core data, both the routine core analysis (RCAL) method to obtain porosity and permeability as well as special core analysis (SCAL) data and empirical interpretation of water saturation from the simulation results, then it becomes a problem and becomes the background for this study by learning the nature of the data and then making predictions for depths that are not there are data. To obtain the prediction of the missing data, machine learning applications are needed to help the project run more effectively and efficiently by studying the algorithm model using the supervised learning method as the target has been labeled to understand the trend of the characteristics of the labeled data to be searched, more explicitly using the classification method in determining the formation and multiclass regression on porosity, permeability, and water saturation by including previously predicted data. This study was applied to 14 wells in the South Sumatra Basin area, which produced solid learning results, and the accuracy was quite precise. Namely, the formation classification gave an F1 score of 0.997, then porosity prediction by adding formation features resulted in an R2 score of 0.646 and an MAE score of 0.042, then like porosity. The permeability prediction resulted in an R2 score of 0.856 and an MAE score of 0.415. Finally, in the water saturation prediction, an R2 score of 0.915 and an MAE score of 0.013 was obtained, which is robust enough. In addition, to get more accurate results in the machine learning process, it is necessary to use a more sophisticated learning algorithm with better methods, such as deep learning with the recurrent neural network algorithm or performing further hyperparameter tuning.