Flow pattern identification of oil-water two-phase flow based on multi-feature convolutional neural network

In order to extract the multivariate characteristics of oil-water two-phase flow accurately and improve the accuracy of flow pattern identification, in this paper, we combine complex network and multivariate fluid feature to propose a multi-feature convolutional neural network. Firstly, we construct a 1D-CNN model for network input, and discuss the influence of each parameter setting on classification performance. Then the original signal (X), the energy signal (ES), the power spectrum signal (P) and the fused signal (M) are selected as the input of the 1D-CNN model respectively. The experimental results show that when we select fusion feature information as network input, the performance of two-phase flow pattern classification is the best. Finally, through multiple sets of experiments to verity the recognition performance of fusion feature information, it can reach 97.6% accuracy on the existing test set. Therefore, the 1D-CNN method based on deep learning theory provides a new idea for flow pattern identification.