An AI-Based Model for Optimizing Pressure, °API, and GOR in Multistage Separator Trains

Abstract This study presents an AI-based model for optimizing pressure, °API, and produced gas-oil ratio (GOR) in a two-phase multistage separator train in oilfields. The optimum second-stage separator pressure for a three-stage separation process is determined using an optimized equation of state in conjunction with flash calculations. A comprehensive database comprising 115,000 data points was generated from 1,349 fluid compositions combined with variations in the first stage separator pressure (ranging from 200 to 1500 psia) and ambient temperatures (60 °F to 90 °F). A total of 14 machine learning regression algorithms were assessed to capture the complex nonlinear relationships between fluid composition, operating conditions, and the resulting optimum performance parameters: separator pressure, °API, and GOR. Linear models were first evaluated, including Linear, Ridge, Lasso, and Elastic-Net regression, followed by nonlinear techniques such as decision tree-based algorithms (Decision Tree, Random Forest, Gradient Boosting, AdaBoost, Extra Trees, XGBoost, and CatBoost), support vector regression, K-Nearest Neighbor, and a multi-layer perceptron model. The results showed that among these algorithms, the CatBoost model achieved superior performance with a mean absolute error (MAE) of 0.65 for the optimum separator pressure (R2 = 0.999), an MAE of 54 for optimum GOR (R2 = 0.999), and an MAE of 0.16 for optimum °API (R2 = 0.999). Moreover, the machine learning approach achieved an increase in liquid yield, with a maximum additional gain of 2% for high °API fluids. The study findings further confirmed that fluid composition and wellhead pressure significantly influence the separation process and that integrating conventional flash calculation methods with machine learning techniques can enhance overall performance. In turn, even marginal improvements in liquid recovery translate into considerable economic benefits for large-scale production facilities.