Three-Stage Machine Learning Pipeline Ensemble with Gradient Boosting and SHAP Analysis – Evaluating Flow Properties of Heavy Crude Under Thermal Conditions

In this study we investigated the enhanced recovery of Agbabu bitumen in Ondo State, southwestern Nigeria, via a thermal injection approach via machine learning architecture using ensemble model 3-stage SK learn pipeline with Gradient Boosting and SHAP. This thermal  method involves the use of a furnace coupled with rheometry to measure the flow properties of the bitumen. The flow properties measured were the plastic viscosity (PV) and yield point (YP) of the bitumen in its natural and thermal states (true boiling point of 977 0F). The plastic viscosity of the bitumen at the thermal state was 0.1433 cP, and it decreases as temperature increases, contrary to when it was in its natural state at no thermal condition. The highest plastic viscosity at an index of 3.894 cP was recorded in the natural state of the bitumen. This shows that, in its natural state, the bitumen has the highest resistance to flow or deformation under shear stress or gravitational force in boreholes, whereas it has the lowest resistance to flow or deformation at the true boiling point. Agbabu bitumen will flow easily or deform under shear stress or gravitational force in boreholes at thermal state.  While measuring the yield point, it was observed that the minimum stress required to initiate flow in the heavy oil at no thermal state is 38.25 lb/100 ft² at a shear stress of 525 MPa.s and a shear rate of 125 s⁻¹. At the thermal state, the minimum stress required to start the fluid flow is 224.57 lb/100 ft² at a shear stress of 225 MPa.s and a shear rate of 5 s⁻¹. The knowledge of the thermal properties of Agbabu bitumen is important to predict its behavior under heat or load and the safe temperature for enhancing its recovery.

Keywords: Heavy oil, Stress, Thermal, Plastic Viscosity and True Boiling temperature.