Hydrate phase equilibria in oil-containing fluid: Experimental investigation and machine learning prediction based stacked ensemble

During oil and gas development and transportation, the presence of oil and complex component in natural gas significantly impacts the hydrate phase equilibrium. However, the specific characteristics of hydrate phase equilibrium in natural gas systems containing oil have not been clearly defined. To address this gap, this study systematically investigated the hydrate phase equilibrium of two natural gas components with oil contents of 40 vol%, 50 vol%, and 60 vol% under pressures ranging from 2 to 8 MPa, based on the actual components from the South China Sea oil and gas fields. Experimental results showed that the oil phase shifted the hydrate phase equilibrium curve towards lower temperatures and higher pressures, with the most significant suppression at 50 vol% oil content. Furthermore, the oil phase had a stronger suppressive effect on hydrates in natural gas containing heavier hydrocarbons (e.g., butane and pentane), with these effects becoming more pronounced at higher pressures. The study indicated that neglecting the oil phase or using simplified natural gas components for hydrate phase equilibrium predictions and inhibitor dosage calculations could have led to overuse of inhibitors, increasing production costs. Based on the experimental data, this study proposed a stacked ensemble model (SELM-DCC) combining multiple POD-RBF base learners and decision coefficients for predicting hydrate phase equilibrium in oil-containing systems. Compared to six other prediction methods, the SELM-DCC model demonstrated lower maximum relative error, enhanced stability, and superior denoising performance. The findings provide practical insights for the study of hydrate phase equilibrium in oil-containing systems and offer a solid framework for model development. Additionally, the proposed model serves as an accurate and versatile predictive tool for the intelligent regulation of inhibitors, offering significant engineering application value.