Identification of Two-phase Flow Pattern Based on Ensemble Learning and Electrical Resistance Tomography

In order to identify the flow pattern of two-phase flow more accurately and efficiently, a flow pattern identification method using ensemble learning was proposed. The data of four flow pattern of gas-liquid two-phase flow in the vertical pipeline were collected by electrical resistance tomography (ERT) system. Firstly, the data were averaged by selecting a certain number of frames, and the extreme gradient boosting (XGBoost) algorithm was constructed by using multiple classification and regression trees (CART) as the weak classifier, which was pre-trained using feature gain as an indicator for feature selection to achieve data dimensionality reduction. Then, five deep neural network (DNN) models and adaptive boosting (AdaBoost) algorithm were combined to construct the DNN-AdaBoost algorithm for flow pattern identification of gas-liquid two-phase flow. Finally, the DNN-AdaBoost algorithm was compared with other flow pattern identification algorithms. Results show that the identification accuracy of the DNN-AdaBoost algorithm is higher than other algorithms, and its average identification accuracy can reach 98.25%.