Developing a cost-effective tool for choke flow rate prediction in sub-critical oil wells using wellhead data.

Accurate prediction of oil production rates through wellhead chokes is critical for optimizing crude oil production and operational efficiency in the petroleum industry. The central thrust of this investigation involves the systematic creation of machine learning (ML) paradigms for the robust prediction of choke flow performance. This endeavor is rigorously informed by comprehensive data acquired from an operational petroleum production facility in the Middle East. Within the dataset, produced gas-oil ratio (GOR), choke size, basic sediment and water (BS&W), wellhead pressure (THP), and crude oil API stand out as key parameters. Each plays a vital role in forecasting the oil production rate. To ensure reliability, robust data preprocessing was conducted using the Monte Carlo outlier detection (MCOD) method to recognize and manage data outliers. The models were trained using 198 data points, employing K-fold cross-validation (five folds) to ensure generalization. Gradient boosting machine (GBM) models were optimized using advanced algorithms like self-adaptive differential evolution (SADE), evolution strategy (ES), Bayesian probability improvement (BPI), and Batch Bayesian optimization (BBO). Among these, SADE demonstrated superior performance based on metrics such as average absolute relative error (AARE%), R