Imaging of flow pattern of gas-oil flows with convolutional neural network

Abstract

Gas volume fraction (GVF) is an important parameter for the measurement of oil-gas two-phase flow. Online measurement of the GVF of oil and gas two-phase flow is of great significance for the safety monitoring and measurement of oilfield production processes. So it is an urgent problem to quickly and accurately detect the real-time GVF according to the non-destructive testing of oilfield field devices. In this paper, a method based on different flow rates and oil-gas ratios is proposed. The method of convolutional neural network (CNN) is used to predict the gas and oil flow rate in oil-gas two-phase flows. Compared with traditional algorithms, CNN algorithm solves the problem of the relationship between high-dimensional data (streaming image pixels) and low-dimensional data (GVF values and traffic) that cannot be solved by traditional algorithms. The data of different flow and oil-gas ratios of oil and gas two-phase flow were collected by experiment. The data collected by electrical capacitance tomography (ECT) was reconstructed using linear projection algorithm (LBP) to obtain the flow pattern. The reconstructed flow graphs are predicted by Binarized image measurement algorithm, SVM algorithm and convolutional neural network algorithm for oil flow rate, gas flow rate, and GVF. The average relative error of GVF prediction is 43% for Binarized image measurement algorithm, 8% for SVM algorithm, and 5% for the CNN algorithm. The CNN algorithm effectively avoids the possible over-fitting problem. Its loss function uses ElasticNet regression instead of least squares regression. The inception V3 model used is decomposed into small convolutions, which can reduce the amount of parameters, reduce over-fitting, and enhance the nonlinear expression of the network. The final model has an allowable error range of 5% and the accuracy can reach more than 90%.