MLFENet: Multilevel Feature Extraction Network for Water Cut Prediction Auxiliary by Flow Pattern Classification in Oil–Water Two-Phase Flow

The existing multitask learning (MTL)-based parameter prediction works mainly focus on the stagewise feature extraction of information sharing for oil–water two-phase flow, resulting in the inadequate interaction between multiple tasks. To this end, this article proposes a multilevel feature extraction network (MLFENet) that enables the progressive information sharing across multiple scales. MLFENet consists of two components: main task enhancement module (MTEM) and auxiliary expert module (AEM). In MTEM, the flow pattern classification is used as an auxiliary task, and a hybrid attention module is used to extract features from the flow pattern classification task (auxiliary task) to the water cut prediction task (main task). In AEM, the expert-gating mechanism is introduced to integrate the features of the auxiliary task into the main task further. Experimental results on the flow parameter prediction dataset demonstrate the superiority and effectiveness of the developed MLFENet on both water cut prediction and flow pattern classification tasks of oil–water two-phase flow.