![](https://digilib.itb.ac.id/assets/icon/pdf.gif)
2024 TA PP ZALBI CHOHID HUSYEN 1-ABSTRAK
Terbatas  Suharsiyah
» Gedung UPT Perpustakaan
Terbatas  Suharsiyah
» Gedung UPT Perpustakaan
Carbon capture, utilization, and storage (CCUS) technologies play a crucial role in mitigating greenhouse gas emissions by capturing CO2 from industrial sources and injecting it underground for storage or enhanced oil recovery (EOR). CCUS can significantly reduce CO2 emissions while simultaneously increasing oil recovery rates by up to 30% (Smith et al., 2019). However, despite its importance, CCUS faces many challenges that require a deeper understanding. One of the primary challenges is optimizing the operational and reservoir parameters to maximize economic viability. Therefore, this study aimed to determine the optimal reservoir parameter criteria for implementing the inverted 5-spot pattern CO2 injection and to develop a predictive model to forecast the economic viability of the project. A literature study approach was used to select parameters for optimization processes, focusing on the inverted 5-spot case with a homogeneous box model. Using CMG-CMost, 500 scenarios were simulated by varying 7 reservoir and operational parameters to construct a predictive model employing the artificial neural network method. The artificial neural network with a 7-7 architecture achieved R² values of 0.989 and 0.945 for training and verification data, respectively. It indicates the reliability of this model for practical applications is acceptable. The optimization of this model, conducted with over 500 experiments in a 20-year forecast, identified the best scenario with an NPV of 3,082 MMUSD, which is an increase of 272% from the base case NPV. The high porosity and thickness supported substantial CO2 storage. Based on the study, it can be concluded that neural networks with a 7-7 architecture have proven to be a successful approach in predicting the most optimal operational and reservoir parameters for inverted 5-spot CO2 injection. The results showed R² values of 0.989 for the training set and 0.945 for the verification set. These high R² values indicate that the predictive model is highly reliable for application in other cases. This model is considered successful because, from the NPV calculation using the proxy model on the best scenario, approximately 2,942 MMUSD was obtained, which has an error of only 4.6% from the simulation calculation.