| Summary: | With diminishing light crude oil reserves, the focus shifts to heavy and extra-heavy crude oil, posing challenges with high viscosity impeding flow. Water-lubricated technology addresses this issue in oil transmission lines. This study introduces a novel method integrating response surface methodology (RSM), computational fluid dynamics (CFD), and optimized machine learning (ML) models to analyze pipeline pressure gradients (PG) in oil–water two-phase flows downstream of T-junctions. The present study uses the D-optimal technique for simulation design to optimize CFD computational demands efficiently. This study breaks new ground by proposing a framework that leverages support vector machines (SVMs). The proposed framework incorporates metaheuristic optimization algorithms (genetic algorithm (GA) and particle swarm optimization (PSO)) to achieve superior PG prediction accuracy. The optimized ML models outperformed RSM models for predicting PG. Results indicated that oil-to-water viscosity ratio and oil inlet velocity significantly affect PG, followed by water inlet velocity and surface tension between phases. In contrast, the oil-to-water density ratio, oil entry angle at the T-junction, and wall contact angle have minimal impact. Furthermore, statistical metrics and visual comparison tools identified the PSO-optimized SVM model based on linear kernel function as the most effective (MAPE = 13.2 % and R = 0.9949). The hybrid methodology presented in this research holds significant promise for optimizing heavy oil transfer efficiency in applications involving water-lubricated technologies. © 2024 Elsevier Ltd
|
|---|