Prediction and Experimental Study of Gas-Water-Foam Three-Phase Flow Patterns in Vertical Wells Using Bka-Bp Neural Network

Abstract

Liquid loading poses a major challenge in late-stage gas well production, requiring effective deliquification to sustain output. Foam-assisted gas lift, achieved by injecting a foaming agent into the wellbore, uses gas-liquid agitation to generate foam, reducing liquid density and surface tension while modifying the flow regime. This study focus on the Sulige Gas Field, where production data under field conditions were scaled using similarity principles. Utilizing the multiphase flow simulation apparatus, gas-water-foam three-phase flow experiments were conducted under varying gas volume flowrate, liquid volume flowrate, and foam drainage agent concentrations at normal temperature and pressure, resulting in the collection of 5,572 experimental data sets. And the three-phase flow pattern of gas-water-foam under different experimental conditions was established. Additionally Based on typical gas-water-foam three-phase flow patterns in vertical wells, we identified the flow pattern under different experimental conditions and mapped them for vertical well. The BKA-BP neural network was optimized through the Black-winged Kite Algorithm (BKA) to improve flow pattern prediction accuracy. The optimization process involved adjusting the BP neural network's weights and biases, enhancing its learning efficiency and convergence rate. BKA, a metaheuristic algorithm inspired by the hunting behavior of the Black-winged Kite, searched for the global optimum by balancing exploration and exploitation in the solution space. This optimization resulted in a 97.45% accuracy for recognizing flow patterns, outperforming other neural network models.

Keywords: Vertical well, Gas-water-foam three-phase flow experiment, BKA-BP neural network, Flow pattern prediction, Multiclass algorithm comparison