Sine-PSO-RF-SHAP prediction gas-liquid two-phase flow patterns in pipelines

Abstract

The identification of gas-liquid two-phase flow patterns in pipelines is of critical significance for industrial process safety and efficiency optimization. However, traditional methods face challenges of insufficient model generalization ability and lack of interpretability. This paper proposes a Sine-PSO-RF-SHAP ensemble model integrating two-stage Sine chaotic mapping, Particle Swarm Optimization (PSO), Random Forest (RF), and SHAP interpretability analysis. Through multi-scale feature analysis and chaos-enhanced parameter optimization strategy, it achieves high-precision prediction of flow pattern classification and mechanistic analysis. Validated by datasets containing gas-liquid combinations of different media and experimental data from complex pipelines such as U-shaped, Z-shaped, and vertical pipes, the model achieves a flow pattern identification accuracy of 94.4 % on the test set, with a single inference time as low as 35.81 ms, significantly outperforming traditional machine learning and deep learning methods. SHAP analysis quantitatively reveals the dominant influence of features such as Surface Gas Velocity and pipe Inclination Angle on flow pattern transition, providing interpretable decision support for industrial scenarios. This study provides a technical paradigm combining efficiency and interpretability for intelligent monitoring of gas-liquid two-phase flows.