DEVELOPING A PREDICTIVE MODEL OF CO2 FLOODING PROJECT

• Main Author: Marusaha Matthew Pandi, D'Aqnan
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/40060
• Institution: Institut Teknologi Bandung

Economic analysis that perceive both technical and economical parameters plays crucial role in identifying the feasibility of EOR application. This step usually be done before further reservoir investigation being conducted. A model that can fulfill the analysis will greatly help the project feasibility study. This study focused on creating a predictive model to accurately predict the reservoir performance for 10 years CO2 flooding project. The model itself constructed by 25 parameters, which affecting both technically and economically in which its value distributed in three ways, discrete real, continuous real, and using a formula and trained in experiments using CMG-CMOST. Experiments that passed the data quality control through several constraints then used as model training and verification data. Net Present Value (NPV) is then used as project's economic objective for the predictive model as it represents the viability of EOR application. Several methods, both regression and neural network was done to predict the chosen objective function, NPV. 6089 experiments generated by CMG-CMOST used as the proxy material to generate the model. Mainly considering the proxy cumulative error and error distribution, the study showed that multilayer artificial neural network with 20-9-6-1 structured neurons gave the best fitted model, where fitted more than 97% with training validated with verification data, followed by CMG-CMOST generated regression, CMG-CMOST generated one layered radial basis function neural network, and self-approached regression. The predictive model that chosen was expected to generate the project’s NPV with confidence level around 80% based on P50 value of proxy verification data.