PERMEABILITY PREDICTION STUDY USING HYDRAULIC FLOW UNIT AND MACHINE LEARNING METHOD IN

• Main Author: Samudera Hafwandi, Babas
• Format: Theses
• Language: Indonesia
• Subjects: Teknologi minyak, lemak, lilin, gas industri
• Online Access: https://digilib.itb.ac.id/gdl/view/71501
• Institution: Institut Teknologi Bandung

Permeability is part of the input used to determine the rate of hydrocarbon production, recovery esrimation, optimal well placement, downhole pressure, fluid contact evaluation, etc. More accurate prediction of permeability in uncored depth remains a challenge in the petroleum industry The goal of this study is to predict the permeability value using the machine learning method and then compare it to the permeability prediction results using the hydraulic flow unit (HFU) method. There are six machine learning models used in permeability prediction: Decision Tree, Gradient Boosting, AdaBoost, Random Forest, Support Vector Machines, and K-Nearest Neighbor. These models are used in the scientific domain of petroleum engineering and involve four prediction scenarios, with the best scenario determining permeability is scenario four, which includes all well log data as well as depth and well coordinates. The Random Forest model is the best for predicting permeability after hyperparameter tuning with prediction score of MAE of 0.4842, MSE of 0.4026, and RMSE of 0.6345 Permeability predictions using the HFU method cannot be performed in the "BSH" field at depths without core data due to a lack of facies data that must be validated with rock types that match the depth intervals without core data. However, using machine learning methods, predictions of permeability values at uncored depths can be made by combining existing log data and predicted porosity data.