Prediction/Assessment of CO2 EOR and Storage Efficiency in Residual Oil Zones Using Machine Learning Techniques

Residual oil zones (ROZ) arise due to a wide range of geologic conditions and are located under the oilwater contact of main pay zones. These ROZs have historically been deemed economically unviable for development using conventional primary recovery methods due to the presence of immobile oil. Yet, they represent substantial subsurface volume suitable for CO2 sequestration and storage. However, there is a deficiency of effective techniques for assessing the performance of CO2-EOR in coupled with CCUS in ROZs. This study introduces the use of Machine Learning techniques to assess/predict the potential of oil recovery and CO2 storage capacity in ROZs. Our framework was built upon the concept of supplying the machine learning model with data obtained from several simulation runs involving CO2 injection in ROZs. This dataset includes key geological and operational attributes as inputs (Thickness, Permeability/Kh, Porosity, Sorw, Sorg, Producer BHP, Injection rate, formation water salinity). The objective is to forecast CO2 storage capacity and oil recovery potential, eliminating the necessity for time-consuming and costly reservoir simulations. We have tested this method in both synthetic and field-scale cases. The study results demonstrated a significant positive correlation between the cum-oil production with sorw, CO2 injection rate, reservoir permeability. In contrast, producer BHP and the vertical permeability to horizontal permeability ratio showed negative correlation. Conversely, the cumulative CO2 storage in ROZs exhibited a positive relation with producer BHP, reservoir thickness, and CO2 injection rate, while showing a negative correlation with reservoir permeability. The utilization of our proposed ANN models has proven highly effective accuracy in predicting CO2-EOR and storage performance. Notably, the tested R2 values for Cumulative oil production and CO2 Storage models were in range of 0.9 to 0.98 with low average absolute percentage error less than 10%. Furthermore, these models serve as a valuable tool for improved reservoir management by optimizing operational parameters, such as producer BHP and CO2 injection rates. These findings have been rigorously validated through real field data, affirming a high level of agreement between the model's predictions and actual outcomes. The developed model can be applied as a fast technical and cost-effective tool for evaluating CO2-EOR and storage in ROZs. Using real ROZs field data demonstrated an excellent agreement between the ANN’s forecasts and the actual data, making it wellsuited for field applications.