Research on Liquid holdup Prediction of Multi Undulating Pipeline Based on CNN-LSTM and Residual Attention

A multi undulating pipeline holdup prediction model (CNN-LSTM-RA) that integrates convolutional neural network (CNN), long short-term memory network (LSTM), and residual attention (RA) mechanism is proposed in this paper. The key features related to pipeline liquid holdup are effectively extracted by the CNN layer, and the dynamic changes of the associated data are analyzed in depth by the LSTM layer. The key features in pipeline data are enhanced by the RA mechanism for recognition. Using simulated and experimental data from literature as samples, the prediction accuracy and effectiveness of the CNN-LSTM-RA model are evaluated and compared with CNN, LSTM, CNN-LSTM, and LSTM-RA models. The results showed that the R2, MAE and RMSE of the CNN-LSTM-RA model are 0.9601, 0.1215, 0.2234, respectively, and indicating high prediction accuracy. The CNN-LSTM-RA model leverages the feature extraction capabilities of CNN, the relevant data analysis advantages of LSTM, and the RA mechanism to enhance the recognition of key features, making it a new method suitable for studying pipeline holdup.