Dual-parameter measurement method for gas-liquid two-phase flow based on ensemble learning and conductance sensors

Abstract

A dual-parameter measurement method for gas-liquid two-phase flow based on ensemble learning and conductivity sensors belongs to the technical field of multiphase flow detection. A dual-plane eight-electrode rotating electric field type conductivity sensor is used to collect the flow information of the gas-liquid two-phase flow and upload it to the host computer for storage; the normalized conductivity average value of the mixed fluid is defined, and the normalized conductivity average values of the eight channels are averaged to obtain the normalized conductivity; the cross-correlation calculation is performed on the voltage signals of two pairs of electrodes at the same position on different planes to obtain the delay time of the two signals, and the delay times at four positions are averaged to obtain the average delay time; the normalized conductivity, the average delay time, and their ratio are used as characteristic variables and input into the stacking ensemble learning model for the prediction of the gas holdup and the liquid-phase volume flow rate. The present invention can obtain the flow information of the gas-liquid two-phase flow in space and time and avoid the mutual coupling and nonlinearity of the two parameters during measurement, and is a method with relatively high measurement accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of multiphase flow detection, and particularly relates to a method for measuring two parameters of gas holdup and liquid-phase volume flow rate in gas-liquid two-phase flow. Background Art

[0002] Gas-liquid two-phase flow is a common multiphase flow, which widely exists in fields such as petrochemical industry, aerospace, and power generation. The accurate measurement of the flow parameters of gas-liquid two-phase flow is of great significance in the industrial production process. However, the morphology of gas-liquid two-phase flow is complex and changeable, and the flow structure shows complexity, randomness, and instability. The flow velocity and concentration distributions are uneven, and there is also an obvious slip effect between the gas phase and the liquid phase. All these make the parameter measurement of gas-liquid two-phase flow very difficult. Cross-sectional gas holdup and flow velocity are two important flow parameters, which can be used to determine fluid density, analyze flow behavior, identify flow states, etc. So far, many sensors based on different principles have been used to measure the cross-sectional gas holdup and flow velocity in gas-liquid two-phase flow, such as differential pressure sensors, conductivity sensors, capacitance sensors, ultrasonic sensors, fiber optic sensors, infrared sensors, thermal sensors, etc.

[0003] Among them, the conductance sensor has been widely used due to its advantages such as low price, high measurement accuracy, and fast response speed. It mainly measures the conductivity of the multiphase flow mixture between two electrodes to calculate the cross-sectional gas holdup. Therefore, it requires the continuous phase of the fluid to be measured to be conductive. For applications in different fields, many different types of conductance sensors have been developed for monitoring flow, identifying flow patterns, calculating gas holdup, and measuring flow velocity. Wang et al. designed and optimized three non-invasive conductance sensors, namely four-electrode, six-electrode, and eight-electrode rotating electric field conductance sensors in "Development of a rotating electric field conductance sensor for measurement of water holdup in vertical oil-gas-water flows" (Measurement Science and Technology, 2018, Vol 29, No 7, Art. no. 075301), and proved that the performance of the eight-electrode rotating electric field conductance sensor is superior to that of the four-electrode and six-electrode rotating electric field conductance sensors. The invention patent with the publication number CN104897737A also discloses a method for measuring gas holdup of an eight-electrode rotating electric field conductance sensor, which mainly discloses a single-plane eight-electrode rotating electric field conductance sensor, and the measurement method adopted is also a traditional direct measurement method. However, the cross-sectional gas holdup and flow velocity have strong coupling and nonlinearity, and there are certain limitations in measuring them based on the traditional empirical formula method, and the conductance sensor of a single plane cannot obtain information in time.

[0004] Therefore, the existing technology urgently needs a new technical solution to solve the above problems. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for measuring two parameters of gas-liquid two-phase flow based on ensemble learning and a conductance sensor, which can obtain the flow information of gas-liquid two-phase flow in space and time, and avoid the mutual coupling and nonlinearity of the two parameters during measurement, and is a method with relatively high measurement accuracy.

[0006] The method for measuring two parameters of gas-liquid two-phase flow based on ensemble learning and a conductance sensor includes the following steps, and the following steps are carried out in sequence:

[0007] Step 1: Adopt a double-plane eight-electrode rotating electric field conductance sensor to collect the flow information of gas-liquid two-phase flow under various working conditions and upload it to the host computer for storage;

[0008] Step 2: Define the normalized conductance average value of the mixed fluid as the ratio of the average voltage of the mixed fluid to the average voltage under the full-water condition, and average the normalized conductance average values of the eight channels to obtain the normalized conductance under the corresponding working condition;

[0009] Step 3: Perform cross-correlation calculation on the voltage signals of two pairs of electrodes at the same position on different planes, obtain the delay time of the two signals, and average the delay times of the four positions to obtain the average delay time under the corresponding working condition;

[0010] Step 4: Use the normalized conductance, average delay time, and their ratios as characteristic variables, input them into the stacking ensemble learning model, and perform predictions on the gas holdup and liquid-phase volume flow rate.

[0011] The eight pairs of electrodes of the double-plane eight-electrode rotating electric field type conductance sensor in the above Step 1 are divided into two groups and embedded in the plexiglass tube, and the four pairs of electrodes on the same plane are arranged at equal intervals.

[0012] In the above Step 4, the stacking ensemble learning model uses the gradient boosting regression tree GBRT, extreme gradient boosting tree XGBoost, and light gradient boosting machine LightGBM as the base learners of the stacking model, and uses ridge regression with cross-validation RidgeCV as the meta-learner of the stacking model.

[0013] Through the above design scheme, the present invention can bring the following beneficial effects: A dual-parameter measurement method for gas-liquid two-phase flow based on ensemble learning and conductance sensor, which uses a double-plane eight-electrode rotating electric field type conductance sensor and a stacking ensemble learning model to predict the gas holdup and liquid-phase volume flow rate of gas-liquid two-phase flow. It can obtain the flow information of gas-liquid two-phase flow in space and time, and avoid the mutual coupling and nonlinearity of the two parameters during measurement, which is a method with relatively high measurement accuracy. Brief Description of the Drawings

[0014] The following further describes the present invention in conjunction with the drawings and specific embodiments:

[0015] Figure 1 Schematic diagram of the double-plane eight-electrode rotating electric field type conductance sensor adopted by the dual-parameter measurement method for gas-liquid two-phase flow based on ensemble learning and conductance sensor of the present invention: (a) Three-dimensional view; (b) Plan view; (c) Electrode three-dimensional view.

[0016] Figure 2Voltage signal diagrams of the dual-plane eight-electrode rotating electric field type conductivity sensor under three flow patterns for the dual-parameter measurement method of gas-liquid two-phase flow based on ensemble learning and conductivity sensors of the present invention: (a) Four pairs of electrodes on the same plane in bubbly flow; (b) Two pairs of electrodes at the same position on different planes in bubbly flow; (c) Four pairs of electrodes on the same plane in slug flow; (d) Two pairs of electrodes at the same position on different planes in slug flow; (e) Four pairs of electrodes on the same plane in churn flow; (f) Two pairs of electrodes at the same position on different planes in churn flow.

[0017] Figure 3 Schematic diagram of the normalized conductivity average values of eight pairs of electrodes under different liquid superficial velocities and gas superficial velocities for the dual-parameter measurement method of gas-liquid two-phase flow based on ensemble learning and conductivity sensors of the present invention.

[0018] Figure 4 Schematic diagram of the average delay times of two planes under different liquid superficial velocities and gas superficial velocities for the dual-parameter measurement method of gas-liquid two-phase flow based on ensemble learning and conductivity sensors of the present invention.

[0019] Figure 5 Schematic diagram of the gas holdup prediction results of gas-liquid two-phase flow based on the stacking ensemble learning model for the dual-parameter measurement method of gas-liquid two-phase flow based on ensemble learning and conductivity sensors of the present invention.

[0020] Figure 6 Schematic diagram of the liquid volume flow rate prediction results of gas-liquid two-phase flow based on the stacking ensemble learning model for the dual-parameter measurement method of gas-liquid two-phase flow based on ensemble learning and conductivity sensors of the present invention. Specific implementation manners

[0021] For the dual-parameter measurement method of gas-liquid two-phase flow based on ensemble learning and conductivity sensors, a dual-plane eight-electrode rotating electric field type conductivity sensor and a stacking ensemble learning model are used to predict the gas holdup and liquid volume flow rate of gas-liquid two-phase flow.

[0022] Specifically, the dual-plane eight-electrode rotating electric field type conductivity sensor is as Figure 1 shown. The eight pairs of electrodes are divided into two groups and are respectively embedded in an organic glass tube with an inner diameter of 40 mm. The distance between the upper and lower groups of electrodes is 40 mm and they are flush in the vertical direction. The four pairs of electrodes on the same plane are equally spaced, and each pair of electrodes is arranged at the positions of Figure 1 (b), which are respectively at the four positions of 0° and 180°, 45° and 225°, 90° and 270°, 135° and 315°. The electrode opening angle of each electrode is 22.5°, the electrode height is 8 mm, and the electrode thickness is 2 mm.

[0023] The designed double-plane eight-electrode rotating electric field type conductivity sensor is installed on a vertical upward gas-liquid two-phase flow pipeline with an inner diameter of 40 mm. By changing the superficial gas velocity and the superficial liquid velocity, different working conditions are generated. When the two-phase fluid flows through the conductivity sensor and tends to be stable, the output voltage signal of the sensor is collected.

[0024] Figure 2 are the voltage signal diagrams of the double-plane eight-electrode rotating electric field type conductivity sensor under three flow patterns. It can be seen from the figure that the measurement signals of the four pairs of electrodes on the same plane are approximately the same, which is due to the non-uniform distribution of the fluid in the pipeline. The measurement signals of the two pairs of electrodes at the same position on different planes are also nearly the same, but there is a time delay, because it takes a certain amount of time for the fluid with the same structure to flow from the upstream electrode to the downstream electrode. By comparing the voltage signal diagrams under the three flow patterns, it can be seen that the voltage waveforms under different flow patterns are also different, which also reflects the ability of the sensor to distinguish flow patterns. In order to better capture the flow characteristics of the gas-liquid two-phase flow, the voltage signals of the eight pairs of electrodes are averaged to obtain the normalized conductivity average value. And the cross-correlation calculation is carried out on the two pairs of electrodes at the same position on different planes, and the calculation results at the four positions are averaged to obtain the average delay time. Figure 3 and Figure 4 respectively show the changes of the normalized conductivity average value and the average delay time under different working conditions. It can be seen that the sensor has satisfactory sensitivity and resolution.

[0025] Specifically, the method for obtaining the normalized conductivity average value is as follows:

[0026] Define the normalized conductivity value G of a single channel e as the ratio of the mixed-phase voltage value V m to the full-water-phase voltage value V w , and the expression is:

[0027]

[0028] Define the normalized conductivity average value G of the double-plane eight-electrode rotating electric field type conductivity sensor e * as the average value of the normalized conductivities of the eight channels, and the expression is:

[0029]

[0030] where G e 1 , G e 2 , G e 3 , G e 4 , G e 5 , Ge 6 , G e 7 , G e 8 are the normalized conductance values of eight channels respectively.

[0031] Specifically, the method for obtaining the average delay time is as follows:

[0032] Perform cross-correlation calculation on the voltage signals of two channels at the same position on the upstream plane and the downstream plane of the double-plane rotating electric field type conductance sensor through equation (3).

[0033]

[0034] Among them, x(k) and y(k+m) are the discrete forms of the upstream signal and the downstream signal, m is the number of delayed samples, N is the length of the samples, and R xy (m) is the cross-correlation function. When R xy (m) takes the maximum value, the corresponding m is the delayed sample point, and the delay time can be calculated based on the sampling frequency.

[0035] Gas-liquid two-phase flow two-parameter prediction based on ensemble learning

[0036] Use the stacking ensemble learning model to predict the gas holdup and liquid volume flow rate of gas-liquid two-phase flow.

[0037] First, take the extracted normalized conductance values, delay time, and their ratio as the inputs of the stacking ensemble learning model. The calculation process of the ratio is as follows:

[0038]

[0039] Among them, and are the normalized conductance values of two pairs of electrodes at the same position on different planes respectively, and τ i is the delay time between the two signals.

[0040] Then preprocess the data. First, shuffle the dataset. 80% of the data is used as the training set, 20% of the data is used as the test set, and then normalize the training set and the test set.

[0041] Secondly, build an ensemble learning model based on stacking. The stacking ensemble learning model uses Gradient Boosting Regression Tree (GBRT), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) as the base learners of the stacking model, and uses Ridge Regression with Cross-Validation (RidgeCV) as the meta-learner of the stacking model. Each of the base learners is an ensemble learning model with decision trees as the basic units, and its basic boosting tree model can be expressed as:

[0042]

[0043] where h m (x; θ m ) represents the m-th decision tree, and θ m represents the parameters of the m-th decision tree, and M is the number of decision trees. The boosting tree algorithm first determines the initial boosting tree, and the expression is:

[0044] f 0 (x) = 0 (6)

[0045] The expression of each subsequent decision tree is:

[0046] f m (x) = f m-1 (x) + h m (x; θ m ) (7)

[0047] GBRT determines the parameters θ m of the m-th decision tree by minimizing the empirical risk, and the expression is:

[0048]

[0049] For the loss function L, GBRT mainly uses the value of the negative gradient of the loss function at the current model as an approximation of the residual, that is, performs a first-order Taylor expansion on the loss function to fit a regression tree. Among them, is the true value.

[0050] XGBoost determines the parameters θ m of the next decision tree by minimizing the structural risk, and the expression is:

[0051]

[0052] where Ω(h m (x)) is the regularization term of the m-th decision tree, and on the basis of GBRT, XGBoost also performs a second-order Taylor expansion on the loss function.

[0053] The described LightGBM mainly uses Gradient-based One-Side Sampling (GOSS) and Exclusive Feature Bundling (EFB) to accelerate the training of the model without affecting the model accuracy.

[0054] The described RidgeCV combines the characteristics of Ridge regression and Cross Validation to improve the performance of the model.

[0055] After adjusting the parameters of the above model, a stacking ensemble learning model suitable for the two-parameter prediction of gas-liquid two-phase flow is obtained. By training the model and denormalizing the training results, the final prediction results are obtained as shown in Figure 5 、 Figure 6 The prediction results shown. Through statistics, the average absolute error of the measurement results of the gas holdup of gas-liquid two-phase flow by the described measurement method is 0.0095, and the average absolute error of the measurement results of the liquid volume flow rate is 0.1539.

Claims

1. A dual-parameter measurement method for gas-liquid two-phase flow based on ensemble learning and conductivity sensors, characterized in that: It includes the following steps, and the following steps are carried out sequentially: Step 1: Use a dual-plane eight-electrode rotating electric field conductivity sensor to collect the flow information of gas-liquid two-phase flow under various working conditions and upload it to the host computer for storage; the eight pairs of electrodes of the dual-plane eight-electrode rotating electric field conductivity sensor are divided into two groups and embedded in the plexiglass tube, and the four pairs of electrodes on the same plane are arranged at equal intervals; each pair of electrodes is respectively at four positions of 0° and 180°, 45° and 225°, 90° and 270°, 135° and 315°; Step 2: Define the normalized conductivity value of a single channel of the mixed fluid as the ratio of the voltage value of the mixed fluid to the voltage value under the full-water condition; Step 3: Perform cross-correlation calculation on the voltage signals of two pairs of electrodes at the same position on different planes to obtain the delay time of the two signals; Step 4: Take the normalized conductivity value, delay time, and their ratio as characteristic variables and input them into the stacking ensemble learning model for predicting the gas holdup and liquid-phase volume flow rate; the calculation process of the ratio is: wherein, and are the normalized conductance values of two pairs of electrodes at the same position on different planes, and τ i is the delay time between the two signals.

2. The dual-parameter measurement method for gas-liquid two-phase flow based on ensemble learning and conductivity sensors according to claim 1, characterized in that: In step 4, the stacking ensemble learning model uses the gradient boosting regression tree GBRT, extreme gradient boosting tree XGBoost, and light gradient boosting machine LightGBM as the base learners of the stacking model, and uses ridge regression with cross-validation RidgeCV as the meta-learner of the stacking model.

Images