STEAM FLOOD RESERVOIR PERFORMANCE ESTIMATION USING MACHINE LEARNING-BASED PREDICTIVE MODEL

• Main Author: Fatih Imani, Kresno
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/48106
• Institution: Institut Teknologi Bandung

Steam injection, known to be capable of reducing oil viscosity is one of the most popular methods for enhanced heavy oil recovery. Although effective, the method’s underlying implementation issue is usually the cost. Tackling cost issues for thermal recovery projects requires a sufficient amount of cumulative produced oil. With an accurate execution strategy, steam flooding can lead to greater recovery than cyclic methods the due to reservoir thermal maturity. Even though attractive, the thermal recovery methods exhibit a considerable amount of uncertainties, especially those associated with reservoir properties. This makes performance prediction through reservoir simulation take upon great importance. A problem commonly encountered in thermal simulations is its lengthy running time due to the complex integration of heat transfer and fluid flow in porous media. This paper provides a practical approach on steam flood reservoir performance prediction which considers the challenges identified so far. A predictive model is completed through coupling experimental reservoir simulations with machine learning. The development involves the acknowledgement of comprehensive field data to ensure a realistic collection of simulation cases. To produce the reservoir performance outcomes, the model accounts for several key parameters including oil gravity, depth, and operating conditions. After comparing with the reservoir simulation results, the developed model showed acceptable accuracy and shorter calculation time. The model can be utilized further for feasibility study and project estimation.