Research On Data Processing Of Two-phase Flow With Low Liquid Fraction Measured By Double Differential Pressure Method

Gas-liquid two-phase flow exists widely in modern process industry. But the flow pattern identification and the flux measurement remain an international challenge and being studied and explored all around the world. Automation Department of China University of Petroleum (East) now is developing a new type and low cost wet gas flowmeter. Its hardware has been completed. On the basis of experiment data acquired from indoor and outdoor experiments, this paper analyzes different feature vectors and uses them as the input of a neuron network system developed to predict the flow pattern and the flux of wet gas.Firstly the maximum modular value of a wavelet is used to denoise so as to prepare for extracting different flow vectors in different operating conditions. Then the chaos theory is used to analyze correlation dimension, Lyapunov index and Hurst index. Comparisons of different changing relations between these parameters, flow patterns and flux of the two-phase flow are also performed. Meanwhile statistics theory is used to analyze parameters including multi-scale information entropy, PSD parameters, PDF parameters, mean, root mean square, standard deviation, skewness, kurtosis and mean absolute error of the differential pressure signals. The analysis of different relations among flow vectors, flow pattern and flux is the key part of this paper. Finally combining of these flow vectors and using them as the inputs of a neuron network system and comparing the performance of a BP network and a BP-GA network are performed.The results show that the flow vectors are effective on flow pattern identification and flux prediction. And the performance achieved by a BP-GA network is more excellent than that of a BP neuron network system. Because of the limitation of training data scale, the validity of this measurement algorithm needs further study.