Gas-liquid two-phase flow phase fraction prediction method based on cylindrical resonant cavity technology

Abstract

The invention relates to the technical field of phase fraction prediction, in particular to a gas-liquid two-phase flow phase fraction prediction method based on a cylindrical resonant cavity. The prediction method comprises the following steps: when a gas-liquid two-phase flow passes through the resonant cavity sensor from the direct-current pipeline, obtaining an actually measured resonant frequency in a resonant cavity body of the resonant cavity sensor to determine a relative frequency difference; acquiring a Froude number Frg representing the gas phase velocity of the gas-liquid two-phase flow; training a void ratio model by taking the relative frequency difference and the Froude number Frg as input and the void ratio as output; and training a volume gas-containing rate model by taking the relative frequency difference, the Froude number Frg and the dynamic viscosity ratio and density ratio of the gas-liquid two-phase flow as well as the volume gas-containing rate as output. According to the invention, the relative frequency difference generated by the change of the resonant frequency is matched with the deep learning model to predict the phase fraction, and the prediction precision of the phase fraction prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of phase fraction prediction, and particularly to a method for predicting the phase fraction of gas-liquid two-phase flow based on a cylindrical resonator. Background Art

[0002] Currently, the techniques for measuring the phase fraction include the gamma-ray method, the capacitance method, the wire mesh method, etc. The gamma-ray method obtains the phase fraction by measuring the attenuation of the ray after passing through the medium, but due to its radioactivity, there are limitations in applications; the measurement accuracy of the capacitance method is greatly affected by the flow pattern, the sensor design and installation are complex, the signal processing and analysis are complex, the system maintenance and calibration are difficult, and the cost is relatively high; the wire mesh method is an intrusive measurement method, which will interfere with the flow pattern during measurement and affect the cross-sectional water content. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for predicting the phase distribution of gas-liquid two-phase flow based on a cylindrical microwave resonator, which overcomes the problems of relying on data quality and overfitting caused by noise in the measurement of gas-liquid two-phase flow, can effectively realize the on-line detection of the phase fraction of gas-liquid two-phase flow, and achieves the purpose of improving the measurement accuracy.

[0004] The present invention provides a method for predicting the phase fraction of gas-liquid two-phase flow based on cylindrical resonator technology, which is carried out based on a resonator sensor pre-installed on a DC pipeline to be detected. The prediction method includes:[[]]

[0005] When the gas-liquid two-phase flow passes through the resonator sensor in the DC pipeline, obtaining the measured resonance frequency in the resonator cavity of the resonator sensor;

[0006] Determining the relative frequency difference according to the measured resonance frequency and the static cavity resonance frequency of the pre-measured resonator;

[0007] Obtaining the Froude number Frg characterizing the gas phase velocity of the gas-liquid two-phase flow;

[0008] Taking the relative frequency difference and the Froude number Frg as inputs and the void fraction as the output, training a pre-constructed void fraction model, and using the trained void fraction model to predict the void fraction in the phase fraction;

[0009] Taking the relative frequency difference, the Froude number Frg and the pre-obtained dynamic viscosity to density ratio of the gas-liquid two-phase flow, and taking the volume gas content as the output, training a pre-trained volume gas content model per unit volume, and using the trained volume gas content model per unit volume to predict the volume gas content per unit volume in the phase fraction.

[0010] In a preferred embodiment, the resonator sensor includes:[[]]

[0011] A wave-transmitting tube, which is embedded in the DC pipeline;

[0012] A resonant cavity is coaxially arranged outside the wave-transmitting tube. An annular cavity is formed between the inner cavity wall of the resonant cavity and the outer tube wall of the wave-transmitting tube.

[0013] Radially-coupled antennas arranged oppositely are both embedded in the cavity along the radial direction of the resonant cavity. One of them is used to couple microwave signals, and the other is used to receive microwave signals.

[0014] When the gas-liquid two-phase flow passes through the wave-transmitting tube of the resonant cavity sensor in the DC pipeline, the internal dielectric constant of the resonant cavity changes, resulting in changes in the magnitude and structure of the magnetic field in the resonant cavity, so that the measured resonant frequency in the resonant cavity changes.

[0015] In a preferred embodiment, the included angle between the oppositely arranged radially-coupled antennas is 180°.

[0016] In a preferred embodiment, both the void fraction model and the gas holdup per unit volume model are XGBoost models.

[0017] In a preferred embodiment, the training method of the XGBoost model includes:

[0018] Preprocess the acquired data;

[0019] Construct a data set according to the preprocessed data. Each sample of the data set includes input features and corresponding output labels;

[0020] Divide the data set into a training set and a test set according to a preset ratio;

[0021] Normalize and transpose the input features and output labels in the training set and the test set to meet the input requirements of the XGBoost model;

[0022] Train the XGBoost model using the training set;

[0023] Based on the test set, use the ant colony algorithm to optimize the parameter combination of the trained XGBoost model to obtain the optimal parameter combination of the XGBoost model when the ant colony algorithm converges;

[0024] Retrain the XGBoost model with the optimal parameter combination;

[0025] Use the retrained XGBoost model for prediction, and perform denormalization processing on the prediction results to obtain the output labels in the original scale.

[0026] In a preferred embodiment, the parameter combination includes a learning rate and a maximum depth; optimizing the parameter combination of the trained XGBoost model using the ant colony algorithm includes:

[0027] Setting the ranges of the learning rate and the maximum depth; setting the parameters of the ant colony algorithm;

[0028] At the initial stage of the algorithm, randomly initialize the positions of each ant, where the position of the ant corresponds to the parameter combination of the XGBoost model;

[0029] For the XGBoost models trained under each parameter combination, evaluate the performance of the model corresponding to each ant according to the root mean square error of the test set, and update the pheromone intensity;

[0030] Each ant selects a new parameter combination according to the updated pheromone intensity and heuristic information, retrains the model using the new parameter combination and updates the pheromone intensity;

[0031] If the performance of the model under the new parameter combination is better than the current optimal solution, update the optimal solution of the parameter combination.

[0032] In a preferred embodiment, the parameters of the ant colony algorithm include the number of ant colonies, the number of iterations, the pheromone importance parameter, the heuristic information importance parameter, and the pheromone evaporation rate.

[0033] In a preferred embodiment, the training set accounts for 70% of the data set.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention takes into account the natural sensitivity of microwave technology to water, and proposes a method for predicting the phase fraction of gas-liquid two-phase flow based on a cylindrical resonator. By using the change in the resonant frequency caused by the change in the dielectric constant of the gas-liquid two-phase flow passing through the microwave resonator, the relative frequency difference generated by the change in the resonant frequency is combined with a deep learning model to predict the phase fraction, improving the prediction accuracy of the phase fraction prediction and achieving the accurate measurement of the phase fraction. Description of the Drawings

[0036] Figure 1 is a schematic structural diagram of the resonator sensor in Embodiment 1;

[0037] Figure 2 is a flowchart of the method for predicting the phase distribution of gas-liquid two-phase flow based on cylindrical resonator technology in Embodiment 1;

[0038] Figure 3 is a prediction result diagram of the void fraction prediction model for predicting the training set and the test set;

[0039] Figure 4It is the prediction result graph of the volume gas holdup model for predicting the training set and the test set.

[0040] Figure 5 It is the performance comparison graph of multiple models for void fraction prediction;

[0041] Figure 6 It is the performance comparison graph of multiple models for volume gas holdup prediction. Specific implementation manners

[0042] The inventors of the present application considered the natural sensitivity of microwave technology to water, and proposed a method for predicting the phase fraction of gas-liquid two-phase flow based on a cylindrical resonator, that is, using the change in the resonant frequency caused by the change in the dielectric constant of the gas-liquid two-phase flow passing through the microwave resonator, plus set factors (pressure, flow rate) to establish a machine learning model, so as to achieve accurate measurement of the phase fraction.

[0043] In addition, machine learning for measuring the phase fraction has a powerful processing ability in nonlinear mapping problems, and at the same time has higher prediction accuracy and stronger generalization ability; it has high accuracy and reduces the risk of overfitting; it can process high-dimensional features and process large-scale data sets; it has flexibility and agility, and allows for partial error costs. When facing problems such as missing values and many variables, machine learning algorithms are very robust.

[0044] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and cannot be used to limit the protection scope of the present invention.

[0045] Combined with Figure 1 , this embodiment provides a method for predicting the phase distribution of gas-liquid two-phase flow based on cylindrical resonator technology, which is carried out based on a resonator sensor pre-installed on a DC pipeline to be detected. The resonator sensor includes a wave-transmitting tube, a resonator cavity body, and radially coupled antennas arranged oppositely; the wave-transmitting tube is embedded in the DC pipeline; in this embodiment, the embedded installation method of the wave-transmitting tube is: connecting two parts of the DC pipeline through a U-shaped groove, the inner wall of the wave-transmitting tube coincides with the inner wall of the DC pipeline, and the outer wall of the wave-transmitting tube coincides with the outer wall of the DC pipeline, so as to effectively prevent the mixed medium from entering the ring cavity. The resonator cavity body is coaxially arranged outside the wave-transmitting tube, and a ring cavity is formed between the inner cavity wall of the resonator cavity body and the outer tube wall of the wave-transmitting tube; during actual operation, the wave-transmitting tube, the ring cavity, and the mixed medium will form three concentric media, and the electromagnetic field has continuity at the boundary, and the equal standing equations with the same establishment of the electric field and magnetic field on both sides are used.

[0046] The relatively arranged radially coupled antennas are all embedded in the cavity along the radial direction of the cavity of the resonant cavity. One of them is used to couple microwave signals, and the other is used to receive microwave signals. The included angle between the relatively arranged radially coupled antennas is 180°. When the gas-liquid two-phase flow passes through the wave-transmitting tube of the resonant cavity sensor in the DC pipeline, the internal dielectric constant of the resonant cavity changes, resulting in changes in the magnitude and structure of the magnetic field in the resonant cavity, so that the measured resonant frequency in the resonant cavity changes. Therefore, the measured resonant frequency measured by the resonant cavity sensor can reflect the change of the phase fraction. In this embodiment, this principle is used to predict the phase fraction through the measured resonant frequency.

[0047] Specifically, the prediction method includes:

[0048] Step S1: Determine the relative frequency difference according to the measured resonant frequency and the static cavity resonant frequency of the resonant cavity measured in advance. In this embodiment, the result of obtaining the measured resonant frequency and the static cavity resonant frequency of the resonant cavity is the relative frequency difference of dimensionless parameters.

[0049] Step S2: Obtain the Froude number Frg characterizing the gas-phase velocity of the gas-liquid two-phase flow; the superficial gas velocity is positively correlated with Frg, so Frg is used to characterize the superficial gas velocity.

[0050] The calculation formula of the Froude number Frg is:

[0051]

[0052] Where, Usg represents the superficial gas velocity; m / s 2 , D represents the fluid pipe diameter; m, ρl represents the liquid-phase density, ρg represents the gas-phase density; kg / s 2 ; g is the acceleration due to gravity; m / s 2 .

[0053] Specifically, the gas-phase Froude number is an extended application of the Froude number in multiphase flow, which is used to describe the relative importance of inertial force and gravity in gas-phase flow. To ensure dimensional homogeneity, the superficial gas velocity information is introduced into the gas-phase Froude number Frg. It can be seen from the formula that the Froude number is positively correlated with the superficial gas velocity and can directly reflect the change of the superficial gas velocity. In the gas-liquid two-phase flow applied in this embodiment, the gas phase occupies a larger proportion compared with water, and the gas phase has a greater influence on the void fraction. Therefore, the Froude number Frg is calculated in this embodiment as an input feature for the subsequent model.

[0054] Step S3: Using the relative frequency difference and the Froude number Frg as inputs and the void fraction as the output, train a pre-constructed void fraction model, and use the trained void fraction model to predict the void fraction in the phase fraction; using the relative frequency difference, the Froude number Frg, and the pre-acquired dynamic viscosity ratio and dynamic density ratio of the gas-liquid two-phase flow (the dynamic density ratio refers to the ratio of the gas density to the liquid density, and the dynamic viscosity ratio refers to the ratio of the gas viscosity to the liquid viscosity), with the volume gas holdup as the output, train a pre-trained volume gas holdup model, and use the trained volume gas holdup model to predict the volume gas holdup in the phase fraction.

[0055] Specifically, the void fraction is the proportion of the gas phase in the total cross-sectional area of the flow area. The volume gas holdup refers to the proportion of the gas volume in the total mixed volume in static and dynamic fluids. In the actual operation process, the true values of the void fraction and the volume gas holdup are difficult to directly obtain through sensors. In this embodiment, the true values obtained from experiments using a fast-closing valve method microwave sensor void fraction calibration device are used as the true labels for model training.

[0056] Specifically, in this embodiment, four influencing factors (resonant frequency, gas phase velocity, pressure, density ratio) were initially considered when selecting the input features of the void fraction. However, it was found during the later fitting process that the addition of pressure and density ratio would cause overfitting. The expressions of the relative frequency difference and the Froude number Frg include the four factors, have a high degree of correlation with the void fraction, and have good actual prediction effects. Therefore, this embodiment finally determines the two parameters of the relative frequency difference and the Froude number Frg as the input features of the void fraction model.

[0057] In this embodiment, the prediction model of the phase fraction includes a void fraction prediction model and a volume gas holdup model. As an efficient gradient boosting algorithm, XGBoost performs well in processing large-scale data sets. Therefore, both the void fraction model and the volume gas holdup model in this embodiment are XGBoost models.

[0058] Since the void fraction in the gas-liquid two-phase flow is affected by multiple factors such as flow velocity, pipe diameter, and fluid physical properties through non-linear coupling, the XGBoost model can accurately capture the high-order interactions between variables through the ensemble learning of gradient boosting trees. The training process of the XGBoost model includes:

[0059] Step S31: Preprocess the acquired data (the data corresponding to the void fraction prediction model is the relative frequency difference, the Froude number Frg, and the void fraction; the data corresponding to the volume gas holdup model is the relative frequency difference, the Froude number Frg, and the pre-acquired dynamic viscosity ratio and density ratio of the gas-liquid two-phase flow). The preprocessing process in this embodiment includes randomly shuffling the data to eliminate the influence of data order on model training.

[0060] Step S32: Construct a data set based on the preprocessed data. Each sample in the data set includes input features and corresponding output labels.

[0061] Step S33: Divide the data set into a training set and a test set according to a preset ratio. In this embodiment, the training set accounts for 70% of the data set.

[0062] Step S34: Normalize the input features and output labels in the training set and the test set, map the data to the interval [0, 1], and transpose it to meet the input requirements of the XGBoost model.

[0063] Since the performance of the XGboost model highly depends on its parameter settings, for the parameter combination of the XGboost model, this embodiment uses the ant colony algorithm to determine the optimal solution. As a heuristic optimization algorithm, the ant colony algorithm optimizes the solution of the problem by simulating the behavior of ants releasing pheromones during the process of searching for food, and can effectively search the parameter space to find the optimal parameter combination.

[0064] Step S35: Use the training set to train the XGBoost model; use the ant colony algorithm based on the test set to optimize the parameter combination of the XGBoost model to obtain the optimal parameter combination of the XGBoost model when the ant colony algorithm converges.

[0065] Specifically, in this embodiment, the ant colony algorithm is used to optimize two key parameters of the XGBoost model: the learning rate (eta) and the maximum depth (max_depth). The parameter range of the XGBoost model is set as follows: the learning rate is between [0.01, 0.3], and the maximum depth is between [3, 10]. The parameter settings of the ant colony algorithm are as follows: the number of ant colonies is 5, the number of iterations is 50, the importance parameter α of pheromone is 1, the importance parameter β of heuristic information is 2, and the pheromone evaporation rate ρ is 0.1.

[0066] The steps of using the ant colony algorithm to optimize the parameter combination of the XGBoost model include: at the initial stage of the algorithm, randomly initialize the position of each ant. The position of the ant corresponds to the parameter combination of the XGBoost model, that is, randomly generate the XGBoost parameter combination corresponding to each ant; for the XGBoost model trained under each parameter combination, evaluate the performance of the model corresponding to each ant according to the root mean square error (RMSE) of the test set, and update the pheromone intensity; in the main loop, each ant selects a new parameter combination according to the pheromone intensity and heuristic information, retrains the model with the new parameter combination and updates the pheromone intensity. If the performance of the model under the new parameter combination is better than the current optimal solution, update the optimal solution of the parameter combination. After multiple iterations, the ant colony algorithm converges to obtain the optimal parameter combination (the best learning rate and maximum depth) of XGBoost.

[0067] Step S36: Retrain the XGBoost model with the optimal parameter combination; after the model training is completed, predict the training set and the test set respectively; and perform inverse normalization on the prediction results to obtain the output labels at the original scale.

[0068] After the above steps, the accuracy of the phase fraction predicted by the XGboost model in this embodiment can be controlled within ±1%, which is a significant increase compared to the prediction accuracy of the traditional empirical model. At the same time, we also compared the five models. By comparing the MAE and RMSE values, the XGBoost model has the best effect and the XGBoost model is finally determined. The model performance comparison diagram is as Figure 5 and Figure 6 shown.

[0069] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and deformations can be made, and these improvements and deformations should also be regarded as the protection scope of the present invention.

Claims

1. A method for predicting the phase fraction of gas-liquid two-phase flow based on cylindrical resonator technology, characterized in that, Based on a resonant cavity sensor pre-installed on the DC pipeline to be detected, the prediction method includes: When the gas-liquid two-phase flow passes through the resonant cavity sensor in the DC pipeline, obtain the measured resonant frequency in the resonant cavity of the resonant cavity sensor; Determine the relative frequency difference according to the measured resonant frequency and the static cavity resonant frequency of the resonant cavity measured in advance; Obtain the Froude number Frg characterizing the gas superficial velocity of the gas-liquid two-phase flow; Taking the relative frequency difference and the Froude number Frg as inputs and the void fraction as the output, train a pre-constructed void fraction model, and use the trained void fraction model to predict the void fraction in the phase fraction; Taking the relative frequency difference, the Froude number Frg, and the dynamic viscosity ratio and dynamic density ratio of the gas-liquid two-phase flow obtained in advance as inputs and the volume gas holdup as the output, train a pre-trained volume gas holdup model, and use the trained volume gas holdup model to predict the volume gas holdup in the phase fraction.

2. The gas-liquid two-phase flow phase fraction prediction method based on cylindrical resonator technology according to claim 1, wherein The resonant cavity sensor includes: A wave-transmitting tube embedded in the DC pipeline; A resonant cavity coaxially arranged outside the wave-transmitting tube, and an annular cavity is formed between the inner cavity wall of the resonant cavity and the outer tube wall of the wave-transmitting tube; Radially coupled antennas arranged oppositely, both embedded in the cavity along the radial direction of the resonant cavity, one of which is used to couple microwave signals and the other is used to receive microwave signals; When the gas-liquid two-phase flow passes through the wave-transmitting tube of the resonant cavity sensor in the DC pipeline, the internal dielectric constant of the resonant cavity changes, resulting in changes in the magnitude and structure of the magnetic field in the resonant cavity, so that the measured resonant frequency in the resonant cavity changes.

3. The gas-liquid two-phase flow phase fraction prediction method based on the cylindrical resonator technology according to claim 1, wherein The included angle between the oppositely arranged radially coupled antennas is 180°.

4. The gas-liquid two-phase flow phase fraction prediction method based on the cylindrical resonator technology according to claim 1, wherein Both the void fraction model and the volume gas holdup per unit volume model are XGBoost models.

5. The gas-liquid two-phase flow phase fraction prediction method based on the cylindrical resonator technology according to claim 4, characterized in that The training method of the XGBoost model includes: Preprocess the obtained data; Construct a data set according to the preprocessed data, and each sample of the data set includes input features and corresponding output labels; Divide the data set into a training set and a test set according to a preset ratio; Normalize the input features and output labels in the training set and the test set, and transpose them to meet the input requirements of the XGBoost model; Use the training set to train the XGBoost model; based on the test set, use the ant colony algorithm to optimize the parameter combination of the XGBoost model to obtain the optimal parameter combination of the XGBoost model when the ant colony algorithm converges; Retrain the XGBoost model with the optimal parameter combination; Use the retrained XGBoost model for prediction, and perform anti-normalization processing on the prediction result to obtain the output label in the original scale.

6. The gas-liquid two-phase flow phase fraction prediction method based on the cylindrical resonator technology according to claim 5, characterized in that The preprocessing of the obtained data includes randomly shuffling the data.

7. The gas-liquid two-phase flow phase fraction prediction method based on the cylindrical resonator technology according to claim 5, wherein The parameter combination includes the learning rate and the maximum depth; the optimization of the parameter combination of the XGBoost model using the ant colony algorithm based on the test set includes: Set the range of the learning rate and the maximum depth of the XGBoost model, and set the parameters of the ant colony algorithm; At the beginning of the algorithm, the positions of each ant are randomly initialized, and the positions of the ants correspond to the parameter combinations of the XGBoost model; For the XGBoost models trained under each parameter combination, evaluate the performance of the model corresponding to each ant according to the root mean square error of the test set, and update the pheromone intensity; Each ant selects a new parameter combination according to the updated pheromone intensity and heuristic information, retrains the model with the new parameter combination and updates the pheromone intensity; If the performance of the model under the new parameter combination is better than the current optimal solution, update the optimal solution of the parameter combination.

8. The method for predicting the phase fraction of gas-liquid two-phase flow based on the cylindrical resonator technology according to claim 7, wherein the parameters of the ant colony algorithm include the number of ant colonies, the number of iterations, the pheromone importance parameter α, the heuristic information importance parameter, and the pheromone evaporation rate.

Images