Flow velocity calculation using slow feature and CNN

Objectives Electrical Resistance Tomography (ERT) is an advanced detection technology, but its current accuracy in pipeline flow velocity measurement is limited. Methods To address the significant errors in flow velocity calculation methods based on ERT, a flow velocity prediction method based on Slow Feature Analysis (SFA) and Convolutional Neural Network (CNN) is proposed. This method first employs the fundamental approach of SFA to extract slow features from historical data, which are then used as key features in the CNN module to predict flow velocity. Results Compared to direct methods such as using CNN for prediction alone, the proposed approach demonstrates higher accuracy and stability, and the average prediction error has decreased by approximately 12.1%. Conclusions By constructing a solid-liquid two-phase flow experimental platform and conducting comparative simulation experiments under various working conditions, the effectiveness and accuracy of the proposed method in predicting flow velocity are validated.