Flow Rate Measurement Of Gas-Liquid Two-Phase Flow Based On EMD And Data Fusion

Gas-liquid two-phase flows exist widely on petroleum,chemical,medicine and power industrial areas,the void fraction,mass quality,flow rate measurements of which has great research values on both academic research and industrial application.But because of the inherent complexity of gas-liquid two-phase flow,the parameter estimation of which has been a difficult problem.In horizontal pipe,considering the significant influence on the phase distribution in gas-liquid two-phase flow caused by the gravity and buoyancy,pure measurement of partial differential pressure from a single direction cannot reflect the flow characteristics comprehensively and accurately.So venturi differential pressure signals which measured from upward and downward directions have been analyzed,and a measurement method of gas-liquid two-phase flow rate which based on empirical mode decomposition(EMD)and neural network data fusion is put forward.Empirical mode decomposition technique is used to extract both the stable components R and fluctuant components D of differential pressure signals.The superiority of EMD worked on the frequency domain analysis is used to reduce the noise arising in actual working conditions.Also energy threshold method is used to deal with the pseudo components since EMD could produce low-frequency pseudo components.Then characteristics of differential pressure signals from different directions are analyzed,followed the feature of which are extracted based on different denoising standards and different energy thresholds.According to the certain nonlinear relationships found from the analysis between feature vectors and flow parameters,the void fraction from different pressure measured directions is predicted by neural network,which has a nonlinear mapping ability.Results indicate that the void fraction prediction effect of bubbly flow and slug flow from upward-get differential pressure data is better,but that of plug flow is worse than downward-get differential pressure data.Clearly,the pressure data from two different direction have obvious complementary information.A neural network data fusion algorithm is utilized to predict the void fraction which is a result based on the fusion data of upward-get differential pressure data and downward-get differential pressure data.Results indicate that data fusion improved the precision significantly.Based on the void fraction prediction values,the nonlinear relationship between void fraction and mass quality was analyzed as well as the nonlinear relationship of void fractionstable component-total mass flow rate.Then the measurements of liquid and gas phase mass flow rate were achieved by using neural network.The average relative error of liquid mass flow rate is 1.71%,and that of gas mass flow rate is 14.97%.Based on feature vectors,we found that the mass quality and the total mass flow rate both have nonlinear relationships with the stable and fluctuant components.Neural network prediction results show that the average relative error of liquid mass flow rate is 1.58%,and that of gas mass flow rate is 8.44%.Therefore,the predictions based on feature vectors have less intermediate process,and have higher prediction precision.