Application of sparrow search algorithm and long-short term memory network in three phase flow pattern prediction

Accurate prediction of flow regimes in vertical gas-liquid slug flow is vital for understanding the flow characteristics of gas-liquid bubble three-phase systems and enhancing oil and gas production efficiency. This study presents a KPCA-ISSA-BiLSTM model for classifying and predicting gas-liquid slug flow regimes. The model considers factors like gas and liquid flow rates, foaming agent concentration, and pressure, using experimental or real production data. Feature extraction and preprocessing are applied before training the model. The results show that the KPCA-ISSA-BiLSTM model achieves 99.69% accuracy on the training set and 98.33% on the test set, with the highest accuracy for foam slug flow prediction. In contrast, BP, CNN, ELM, and LSTM models yield accuracies between 84% and 90%. The proposed model outperforms these alternatives, offering an effective tool for predicting flow regimes and providing valuable support for optimizing gas-liquid three-phase flow applications in engineering.