Void fraction measurement using modal decomposition and ensemble learning in vertical annular flow

• Published in: Chemical engineering science 2022-01, Vol.247, p.116929, Article 116929
• Main Authors: Li, Chaofan, Liu, Miaomiao, Zhao, Ning, Wang, Fan, Zhao, Zhiyue, Guo, Suna, Fang, Lide, Li, Xiaoting
• Format: Article
• Language: English
• Subjects: Annular flow; Empirical modal decomposition; Extreme gradient boosting; Kernel Ridge Regression; Void fraction
• ISSN: 0009-2509; 1873-4405
• DOI: 10.1016/j.ces.2021.116929

•A new void fraction prediction model is proposed based on ensemble learning.•Near-infrared signal decomposition using the Empirical Modal Decomposition method to obtain the energy features of the Intrinsic Mode Function.•Analysis of annular flow characteristics from energy perspective. Determine input variables using energy decay analysis.•Five prediction models were developed and compared with other void fraction models. The void fraction is a key parameter for calculating the average density and pressure gradient and analyzing the flow conditions in gas–liquid two-phase flow. However, due to the complexity and variability of gas–liquid two-phase annular flow, the void fraction measurement has been an unsolved scientific problem in scientific research and industrial applications. In this study, a new high-precision real-time void fraction prediction model is proposed by combining the energy feature extraction from the empirical modal decomposition (EMD) method, the anomaly filtering from the kernel ridge regression (KRR), and ensemble learning from the extreme gradient boosting (XGBoost). To further validate the prediction performance of the model, it is compared with the lasso regression model (LASSO) based on the EMD decomposition method and a single XGBoost model. The results show that the prediction accuracy can be guaranteed in the case of anomalous energy eigenvalues.