Machine learning-based prediction of well performance parameters for wellhead choke flow optimization

Accurate measurement and forecast of fluid flow rates in production wells are important to the estimation of hydrocarbon recovery, attainment of stable and controllable flow regimes, and optimization of production plans. Wellhead chokes, or pressure control valves, find widespread use in the hydrocarbon industry for two major reasons: provision of a stable downstream pressure and creation of the necessary backpressure for balancing gas well productivity and controlling in-well pressure drops. Over the past fifty years, numerous multiphase flow models and empirical correlations have been developed to estimate flow rates under a wide range of fluid properties, flow regimes, and pressure drop conditions. None of these models is deemed to be globally applicable to every region because each has inherent measurement errors that limit the accuracy of predictions for well performance parameters. In this study, three machine learning algorithms-Convolutional Neural Network (CNN), Multilayer Perceptron (MLP), and Radial Basis Function Network (RBFN)-were employed to predict well performance parameters. The dataset consisted of 182 samples for each of the five input parameters-liquid production rate (QL), wellhead pressure (Pwh), choke size (D64), basic sediment and water content (BS&W), and gas-liquid ratio (GLR)-resulting in a total of 910 data points. Among the tested models, MLP demonstrated the highest predictive performance, achieving R² values of 0.9985 (training), 0.9856 (validation), and 0.9936 (testing). Four error metrics-Root Mean Square Error (RMSE), Mean Square Error (MSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE)-were used for evaluation. For the MLP model, RMSE values of 0.0024 (training) and 0.0057 (testing) were obtained. The dataset was split into training and testing sets with a ratio of 70:30.