#TITLE_ALTERNATIVE#

• Main Author: PRASETYA (NIM : 12214002), NOVRISAL
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/29742
• Institution: Institut Teknologi Bandung

Proxy model to predict recovery factor is essential as a preliminary screening of steamflooding EOR design to find potential scenarios before it is simulated in reservoir simulator. There are three common mathematical models (analytical, numerical, and statistical) that has been published by several investigators to predict steamflood performance, but the model does not represent a wide range of reservoir properties. This study aims to develop a proxy model that are able to predict recovery factor of a wide range of various steamflood field parameters by using Deep Neural Network (DNN). <br /> <br /> Generality of the model is the main challenge in this study. The key ideas are to properly analyze the bias-variance trade-off, and to train the proxy model with a wide range of data. In this study, the bias-variance trade-off is systematically analyzed by application of dataset splitting (train-val-test split), and several diagnostic curve (Learning Curve and Validation Curve), while the dataset is generated by using Latin Hypercube sampling towards steamflood screening criteria and several steamflood fields properties to represent worldwide steamflood project. The proxy model performance is evaluated using Root Mean Squared Error (RMSE) and Coefficient of Determination ( ). <br /> <br /> The DNN proxy model has 2 hidden layers with 50 neurons each. It successfully handled the bias-variance problem concluded from the resulting performance metrics. It gives RMSE of 3.32; 4.80; 5.37 and of 0.90; <br /> <br /> 0.80; 0.72, respectively for training; validation; and test. set. Overall, it has a low RMSE (3.93) and high (0.86), no overfit, and is a relatively general model. <br /> <br /> The novelty of this study is the implementation of machine learning technology to develop a steamflood proxy model, that is, the Deep Neural Network to predict recovery factor. It can be applied for different reservoir flow properties, representative towards worldwide steamflood EOR project, and also provide a very fast prediction.