Development Of A Novel Method Of Multiphase Flow Metering In Coriolis Flow Meter

Coriolis flowmeter is widely used in industrial metering because of its certainty and the ability to measure both fluid density and mass flow at the same time.The single-phase measurement error is as low as 0.1%,but there are still obvious measurement errors in multi-phase flow measurement applications such as petroleum industry.In order to solve this problem,this paper designed and built a three-phase flow measurement platform to study the influence of gas volume fraction,water cut,liquid velocity and other factors on the measurement error under different fluid components,and also to propose a Coriolis flowmeter multi-phase flow measurement correction method combined with BP neural network and genetic algorithm proposed a Coriolis flowmeter multi-phase flow measurement correction method.In order to study the performance of Coriolis flowmeter,a total of 152 data sets of oil,water and gas multiphase mixed fluid were collected by using an experimental matrix including liquid volume flow rate,gas volume flow rate,gas volume fraction and water cut.The results show that the water cut,gas volume fraction and liquid volume flow rate of multiphase fluid all have a certain influence on the measurement error,but the gas volume fraction has the greatest influence on the measurement error.When the gas volume fraction is low(<10%),the measurement error is small and the growth trend is slow with the increase of gas volume fraction.When the gas volume fraction gradually increased(<55%),the error increases sharply.When the gas volume fraction exceeds 55%,the density measurement function stops working due to too large error.In view of the influence of gas volume fraction on measurement error,the measurement error of gas-liquid two-phase flow including oil-gas two-phase flow and gas-water two-phase flow is analyzed,and two main reasons for measurement error of gas-liquid flow are found out:phase separation and gas compressibility.The variation of mass flow measurement error is mainly caused by multiphase separation and gas compressibility,and the measurement error turns from positive to negative with the increase of gas holdup.Density measurement is mainly affected by gas compressibility,and its measurement error remains negative and increases with the increase of gas holdup.The effects of water cut and liquid volume flow rate were studied experimentally.For the liquid volume flow rate,the measurement errors of the instrument when the liquid volume flow rate is 5,10,15,20 and 25 m~3/h are analyzed respectively with the control of the gas volume fraction and water cut unchanged.The results show that the measurement errors increase relatively with the increase of the liquid volume flow rate,but generally not more than 2%.For moisture content,the mixed fluid without gas was selected,the liquid volume flow was controlled unchanged,and the measurement error when the moisture content was respectively 0%,30%,70%and 100%was explored.Through experimental analysis and conclusion,when the moisture content is 30%,the measurement error is relatively obvious.Based on experimental study on the measurement error law of multiphase flow,a deep learning correction framework is built to solve the problem of large measurement error of multiphase flow.On the premise of repetitive experiments provide a theoretical support,has established two correction model:(1)the error model was built based on BP neural network,provide the correction for multiphase flow measurement,correction for the first time,according to the results of mass flow prediction model of correction error within 0.94%,density prediction model of correction error within 1.27%;(2)On this basis,the second revision model was established by increasing the training set capacity and optimizing the genetic algorithm,etc.,and the mass flow error was successfully corrected to within-0.21%and the density error to within0.38%.In addition,this paper also predicts and compares the four different data distribution modes of 80%?20%,50%?50%,40%?60%,and 20%?80%,in order to find out the minimum proportion of training data set that is economical and necessary.The results show that 50%training data-50%verification data allocation method is the best scheme.