Prediction of pressure drop in gas-water-foam three-phase flow using a data-driven approach: foam multiphase flow experimental analysis

Foam lift is a widely employed technique to address liquid loading in gas wells., but accurately predicting pressure distribution remains challenging. However, the injection of foaming agents introduces significant complexity and variability to wellbore pressure, while traditional empirical models involve numerous parameters and are often overly complex and redundant. To address above, an innovative Black Kite Algorithm (BKA)-optimized BP neural network is proposed to predict pressure drops in gas-water-foam three-phase flow. Using experimental data foam multiphase flow tests under varying gas volume flowrates (0.5~150 m3/h), liquid volume flowrates (0.1~3.0 m3/h), foaming agent concentrations (0~0.3%), pipe inclinations (30°~90°), and liquid holdup conditions, 6678 data points were collected for model training, and 450 literature-based data points for validation. The BKA-BP model achieved high predictive accuracy, with R2 values of 0.837–0.905 and RMSE of 0.00632–0.0075, effectively capturing nonlinear pressure variations, including reductions at moderate gas volume flowrates (5–50 m3/h) and increases at high volume flowrates (80–150 m3/h) due to foam viscosity and interfacial friction. Inspired by black kite hunting behavior, the BKA algorithm combines global search and local optimization for rapid convergence. Compared to RF, SVM, and CNN models, BKA-BP outperformed in accuracy, robustness, and efficiency, providing a reliable data-driven solution for optimizing gas well production and extending operational lifespans.