A method for identifying gas-liquid two-phase flow patterns based on a dual-frequency Coriolis flowmeter and a deep neural network

Abstract

This invention relates to the field of pipeline safety management technology, specifically to a method for identifying flow patterns in gas-liquid two-phase flow based on a dual-frequency Coriolis flowmeter and a deep neural network. The method includes experimental testing and parameter selection: using only a dual-frequency Coriolis flowmeter to measure various parameters under different pressures and flow patterns, obtaining dynamic parameter curves characterizing different flow patterns; converting parameter curves into curve images: obtaining a sample dataset of curve images; establishing a flow pattern identification model: using a deep neural network to establish a flow pattern identification model and training it with the sample dataset; and finally, flow pattern identification. This invention leverages the advantages of the dual-frequency Coriolis flowmeter's multi-parameter, high-precision measurement capabilities, directly acquiring the flowmeter's output parameters, converting characteristic parameter curves characterizing different flow patterns into parameter images, and establishing a flow pattern identification model based on a deep neural network. This approach can significantly improve the accuracy of flow pattern identification, meeting the needs of metering and safe production operations in the petroleum industry.

Description

Technical Field

[0001] This invention relates to the field of pipeline safety management technology, and more specifically, to a method for identifying gas-liquid two-phase flow patterns based on a dual-frequency Coriolis flowmeter and a deep neural network. Background Technology

[0002] In the oil extraction process, the produced fluid from oil wells is a mixture of water, petroleum, gas, and sand, characterized by large variations in gas-liquid flow rates and complex and variable flow patterns. Common gas-liquid two-phase flow patterns in horizontal pipes include bubbly flow, slug flow, laminar flow, and plug flow. These flow pattern variations significantly affect the flow and metering characteristics of the two-phase flow, making them a key technology for ensuring high-precision metering of gas-liquid two-phase flow and safe pipeline operation.

[0003] Currently, the main methods for flow pattern identification can be divided into two categories: direct identification methods and indirect identification methods. Direct identification methods determine the flow pattern directly based on the form of multiphase / two-phase flow images. Direct identification methods mainly include:

[0004] (1) Visual observation method: The flow state of multiphase flow inside the tube can be directly observed with the naked eye through the transparent tube. It requires manual observation and identification and is limited to transparent pipes and transparent fluids.

[0005] (2) High-speed photography method: High-speed cameras are used to record the flow state of fluids, and this method is limited to transparent pipes and transparent fluids.

[0006] Indirect identification methods involve using different testing instruments to measure and collect various characteristic parameters that reflect the changing characteristics of multiphase flow patterns. Different data processing methods are then applied to establish flow pattern identification models to indirectly identify flow patterns. This has become a major development direction in flow pattern identification; it mainly includes:

[0007] Flow pattern diagram method: This method requires measuring the relevant parameters and then manually identifying them by referring to flow pattern diagrams, resulting in low identification efficiency. However, due to varying experimental conditions, it is difficult to replicate the experimental conditions of flow pattern diagrams. Therefore, the boundaries of different flow patterns are drawn roughly, and the method has certain applicable conditions. Furthermore, parameters such as phase separation velocities are difficult to measure directly.

[0008] Direct identification methods, such as visual inspection, are highly subjective and difficult to quantify. High-speed photography, due to multiple reflections and refractions at the multiphase flow interface, reduces accuracy. Indirect identification methods, utilizing the close relationship between characteristic parameters and flow patterns, are easy to quantify and have broad development prospects.

[0009] Traditional indirect identification methods utilize various measuring instruments, such as gamma-ray densitometers, conductivity probes, capacitance probes, and differential pressure flowmeters, to collect different flow parameter signals of two-phase flows. However, traditional measuring instruments suffer from the drawback of having only one output parameter. Furthermore, traditional flow pattern identification modeling data processing methods extract time-domain or frequency-domain features characterizing flow pattern changes from the dynamic curves of the parameters output by the measuring instruments, using these features as input to the flow pattern identification model. This manual extraction of parameter features suffers from incomplete extraction of useful features. In addition, modern data processing and modeling methods have evolved from traditional statistical analysis to time-frequency domain analysis, wavelet transforms, neural networks, and other methods. Compared to traditional neural network classification and support vector machine models, deep neural networks can automatically identify the features of different flow patterns contained in the pixels of the parameter curve image, resulting in more comprehensive feature extraction and higher accuracy in classification models.

[0010] Furthermore, indirect identification methods require various measuring instruments, which have drawbacks such as high cost, applicability only to specific operating conditions, unsuitability for complex on-site conditions, and limited output parameters. The acquired parameter signals characterizing different flow patterns require manual feature extraction, resulting in a large number of required features or incomplete extraction of useful signals. Modeling based on photographic images also suffers from the adverse effects of strong image dynamics and randomness. Therefore, to improve the accuracy of two-phase flow pattern identification, we propose a gas-liquid two-phase flow pattern identification method based on a dual-frequency Coriolis flowmeter and a deep neural network. Summary of the Invention

[0011] The purpose of this invention is to provide a method for identifying gas-liquid two-phase flow patterns based on a dual-frequency Coriolis flowmeter and a deep neural network, so as to solve the problems mentioned in the background art.

[0012] To address the aforementioned technical problems, the present invention aims to provide a method for identifying gas-liquid two-phase flow patterns based on a dual-frequency Coriolis flowmeter and a deep neural network, comprising the following steps:

[0013] S1. Experimental Testing and Parameter Selection: Taking advantage of the multi-parameter measurement capabilities of the dual-frequency Coriolis flowmeter, and based on the designed two-phase flow experimental scheme, a single dual-frequency Coriolis flowmeter was used to measure various parameters output by the flowmeter under different pressures and flow patterns, from which dynamic curves of parameters that can characterize different flow patterns were obtained.

[0014] S2. Converting parametric curves into curve images: A processing method is adopted to convert dynamic parametric curves into curve images. The time-domain parametric curves containing different flow pattern characteristics are converted into curve images. The pixels in the image can fully reflect the characteristics of different flow patterns. The different parametric curves of each experimental data point representing the flow pattern change are normalized and mapped onto the same image, thereby obtaining a sample dataset of curve images.

[0015] S3. Establish a manifold recognition model: Use the curve image sample set obtained in step S2 as the model input, and use a deep neural network suitable for image processing to establish a manifold recognition model. After training with the sample dataset of curve images, it can accurately identify a variety of different manifolds, serving as the basis for measuring different manifolds.

[0016] S4. Manifold Recognition: New curve images are acquired in real time using steps S1-S2. These new curve images are directly input into a trained manifold recognition model based on a deep neural network for recognition, and the recognized manifold results are directly output.

[0017] As a further improvement to this technical solution, in step S1, the different flow patterns of the gas-liquid two-phase flow include at least bubbly flow, slug flow, laminar flow, and plug flow.

[0018] As a further improvement to this technical solution, in step S1, the parameters that the dual-frequency Coriolis flow meter can measure and output include at least the apparent mass flow rate, apparent density, first and third resonant frequencies, first and third damping, and amplitude.

[0019] As a further improvement to this technical solution, in step S1, when measuring gas-liquid two-phase flow using a dual-frequency Coriolis flowmeter, the responses of the two vibration modes are completely different; the third-order resonant frequency is approximately six times the first-order resonant frequency. The different properties of the third-order modes provide different characteristics for flow pattern identification. Because the gas holdup and rate of change differ among different flow patterns, the time-domain density characteristics of different flow patterns vary significantly. The first and third-order resonant frequencies of the dual-frequency Coriolis flowmeter also have the following relationship with density:

[0020]

[0021]

[0022] In the formula, c 11 c 12 and c 31 c 32 Both are called density measurement coefficients, ρ app1 and ρ app3 The apparent densities corresponding to the first and third resonant frequencies are f1 and f3, respectively. Since the density magnitudes and degrees of variation differ among different flow patterns, the first and third frequencies output by the dual-frequency Coriolis flowmeter can closely characterize the changes in different flow patterns. Furthermore, to characterize the influence of pressure on flow patterns under different operating conditions, the gas-liquid density ratio is used to characterize the effect of different pressures on different flow patterns.

[0023] As a further improvement to this technical solution, in step S2, when acquiring the sample dataset of the curve image, at least two characteristic parameters that can characterize different flow patterns of gas-liquid two-phase flow are selected, and the curves of different parameters are normalized and mapped onto the same image to form a flow pattern image.

[0024] As a further improvement to this technical solution, in step S3, when training the manifold recognition model, the sample dataset of the curve image needs to be divided into a training set and a test set according to a certain proportion using a random method in advance, so as to fully train and test the manifold recognition model.

[0025] As a further improvement to this technical solution, in step S3, when dividing the training set and the test set, the proportion of the test set to the training set is less than or equal to 3 / 7.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] 1. In this gas-liquid two-phase flow pattern identification method based on dual-frequency Coriolis flowmeter and deep neural network, compared with the direct identification method, this method belongs to the indirect identification method, which is not limited to transparent pipes and transparent fluids, is easy to model, and has high online identification accuracy;

[0028] 2. Compared to indirect identification methods, this gas-liquid two-phase flow pattern identification method based on a dual-frequency Coriolis flowmeter and deep neural network uses only one dual-frequency Coriolis flowmeter in the experimental setup. Compared with existing measuring instruments, it has the advantages of small size, simple structure, multi-parameter measurement, and high precision. It can obtain more selectable parameters that characterize the flow pattern, such as mass flow rate, density, resonant frequency, and damping. It can achieve flow pattern identification without adding any equipment while measuring flow rate with high precision. The flow pattern identification results can be used to further improve the measurement accuracy of the flowmeter.

[0029] 3. In this gas-liquid two-phase flow pattern identification method based on dual-frequency Coriolis flowmeter and deep neural network, compared with the traditional method of manually extracting multiphase flow parameter features, the dynamic curves of parameters that can characterize different flow patterns are converted into parameter curve images. The parameter curve images of different flow patterns have obvious features, which can better retain and apply all the useful features of the characteristic parameters of different flow patterns. Furthermore, a flow pattern identification model is established by using a deep neural network suitable for image processing. The output parameter image features of the Coriolis flowmeter under different flow patterns are obvious, eliminating the need for manual feature extraction and better retaining all the useful features of the parameters reflecting different flow patterns, resulting in higher identification accuracy.

[0030] 4. In this gas-liquid two-phase flow pattern identification method based on dual-frequency Coriolis flowmeter and deep neural network, the advantages of multi-parameter and high-precision measurement of dual-frequency Coriolis flowmeter are utilized to directly collect the output parameters of the flowmeter, transform the characteristic parameter curves that can characterize different flow patterns into parameter images, and establish a flow pattern identification model based on deep neural network, with an overall recognition rate of 99.88%. Transforming the dynamic parameter curves that can characterize different flow patterns output by dual-frequency Coriolis flowmeter into parameter images as input to the deep neural network model can better improve the accuracy of flow pattern identification and meet the needs of petroleum industry metering and safe production operation. Attached Figure Description

[0031] Figure 1 This is an exemplary overall method principle flowchart of the present invention;

[0032] Figure 2 This diagram illustrates a comparison of the principles of the conventional method and the method described in this invention.

[0033] Figure 3 This is an example of the distribution of experimental data points for different flow patterns in this invention;

[0034] Figure 4 These are sample images of different flow patterns exemplified in this invention;

[0035] Figure 5 This is an exemplary manifold recognition model structure diagram based on the AlexNet network in this invention;

[0036] Figure 6 This is a training parameter table diagram of an exemplary deep neural network in this invention;

[0037] Figure 7 The figures show the test results of different exemplary models in this invention.

[0038] Figure 8 This is a confusion matrix diagram of the test results of an exemplary AlexNet model in this invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0040] Example

[0041] like Figures 1-2As shown, this embodiment provides a method for identifying the flow pattern of gas-liquid two-phase flow based on a dual-frequency Coriolis flowmeter and a deep neural network. It fully utilizes the advantages of the dual-frequency Coriolis flowmeter—multi-parameter, high-precision measurement, and simultaneous dual-frequency measurement—to convert parameter curves into parameter images. A flow pattern identification model is then established using a deep neural network, which excels at image classification, thereby achieving accurate identification of the flow pattern of gas-liquid two-phase flow in a horizontal pipe. Specifically, the method includes the following steps:

[0042] S1. Experimental Testing and Parameter Selection: Taking advantage of the multi-parameter measurement capabilities of the dual-frequency Coriolis flowmeter, and based on the designed two-phase flow experimental scheme, a single dual-frequency Coriolis flowmeter was used to measure various parameters output by the flowmeter under different pressures and flow patterns, from which dynamic curves of parameters that can characterize different flow patterns were obtained.

[0043] In this step, the different flow patterns of the gas-liquid two-phase flow include at least bubbly flow, slug flow, laminar flow, and plug flow. The parameters that the dual-frequency Coriolis flowmeter can measure and output include at least the apparent mass flow rate, apparent density, first and third resonant frequencies, first and third damping, and amplitude.

[0044] Specifically, the dual-frequency Coriolis flowmeter employs a unique multi-frequency technology (MFT), especially the dual-frequency technology, which adds a third-order vibration mode on top of the first-order vibration mode. The first and third-order resonance modes exhibit different characteristics under different flow patterns, thus providing more information on the changes in the two-phase flow pattern.

[0045] Furthermore, when measuring gas-liquid two-phase flow using a dual-frequency Coriolis flowmeter, the responses of the two vibration modes are completely different; the third-order resonant frequency is approximately six times the first-order resonant frequency. The different properties of the third-order modes provide different characteristics for flow pattern identification. Because the magnitude and rate of change of gas holdup vary among different flow patterns, the time-domain density characteristics of different flow patterns differ significantly. The first and third-order resonant frequencies of the dual-frequency Coriolis flowmeter also have the following relationship with density:

[0046]

[0047]

[0048] In the formula, c 11 c 12 and c 31 c 32 Both are called density measurement coefficients, ρ app1 and ρ app3The apparent densities corresponding to the first and third resonant frequencies are f1 and f3, respectively. Since the density magnitudes and degrees of variation differ among different flow patterns, the first and third frequencies output by the dual-frequency Coriolis flowmeter can closely characterize the changes in different flow patterns. Furthermore, to characterize the influence of pressure on flow patterns under different operating conditions, the gas-liquid density ratio is used to characterize the effect of different pressures on different flow patterns.

[0049] S2. Converting parametric curves into curve images: A processing method is adopted to convert dynamic parametric curves into curve images. The time-domain parametric curves containing different flow pattern characteristics are converted into curve images. The pixels in the image can fully reflect the characteristics of different flow patterns. The different parametric curves of each experimental data point representing the flow pattern change are normalized and mapped onto the same image, thereby obtaining a sample dataset of curve images.

[0050] In this step, when acquiring the sample dataset of curve images, at least two characteristic parameters that can characterize different flow patterns of gas-liquid two-phase flow are selected, and the curves of different parameters are normalized and mapped onto the same image to form a flow pattern image.

[0051] S3. Establish a manifold recognition model: Use the curve image sample set obtained in step S2 as the model input, and use a deep neural network suitable for image processing to establish a manifold recognition model. After training with the sample dataset of curve images, it can accurately identify a variety of different manifolds, serving as the basis for measuring different manifolds.

[0052] In this step, when training the manifold recognition model, the sample dataset of curve images needs to be divided into training and test sets according to a certain ratio using a random method in advance, so as to fully train and test the manifold recognition model.

[0053] Specifically, when dividing the training set and the test set, the proportion of the test set to the training set is less than or equal to 3 / 7.

[0054] The ratio of training set to test set can be 7:3, 8:2, 9:1, etc.

[0055] S4. Manifold Recognition: New curve images are acquired in real time using steps S1-S2. These new curve images are directly input into a trained manifold recognition model based on a deep neural network for recognition, and the recognized manifold results are directly output.

[0056] like Figure 2 As shown, this scheme is most relevant to the indirect identification method in the prior art, especially the flow pattern identification technology based on deep neural networks and combined with signal processing and feature extraction methods. This scheme can be understood as an indirect identification method based on dual-frequency Coriolis flow meters (also known as dual-frequency Coriolis flow meters).

[0057] Compared to existing methods, the dual-frequency Coriolis flowmeter offers advantages such as smaller size and multi-parameter / high-precision measurement. Currently, this flowmeter is in mass production, significantly reducing costs. It also exhibits high vibration stability under multiphase flow conditions. Furthermore, the use of multi-frequency measurement technology provides more information on changes in the two-phase flow pattern.

[0058] Application Examples

[0059] like Figures 3-8 As shown, this embodiment provides a method for identifying gas-liquid two-phase flow patterns based on a dual-frequency Coriolis flowmeter and a deep neural network, including the following:

[0060] Step 1: Experimental Testing and Parameter Selection

[0061] The experiment collected various parameters of the dual-frequency Coriolis flow meter under four different flow patterns: bubbly flow, slug flow, laminar flow, and plug flow, including apparent mass flow rate, apparent density, first and third resonant frequencies, first and third damping, and amplitude.

[0062] Traditional flow pattern identification methods employ a variety of flow meters with limited output parameters. In contrast, this solution utilizes a dual-frequency Coriolis flow meter, offering the advantage of multi-parameter output. This provides more information on flow rate and flow pattern changes in two-phase flows, making it easier to select characteristic parameters representing these changes. For example... Figure 3 The figure shows the distribution of experimental data points for different flow patterns.

[0063] The dual-frequency Coriolis flowmeter employs a unique multi-frequency technology (MFT). When measuring gas-liquid two-phase flow with the dual-frequency Coriolis flowmeter, the two vibration modes respond completely differently; the third-order resonant frequency is approximately six times the first-order resonant frequency. The different properties of the third-order modes provide different characteristics for flow pattern identification. Because the magnitude and rate of change of gas holdup vary among different flow patterns, the time-domain density characteristics of different flow patterns differ significantly. The following relationship also exists between the first and third-order resonant frequencies of the dual-frequency Coriolis flowmeter and density:

[0064]

[0065]

[0066] In the formula, c 11 c 12 and c 31 c 32 Both are called density measurement coefficients, ρ app1 and ρ app3The apparent densities corresponding to the first and third resonant frequencies are f1 and f3, respectively. Since the density magnitudes and degrees of variation differ among different flow patterns, the first and third frequencies output by the dual-frequency Coriolis flowmeter can closely characterize the changes in different flow patterns. Furthermore, to characterize the influence of pressure on flow patterns under different operating conditions, the gas-liquid density ratio is used to characterize the effect of different pressures on different flow patterns.

[0067] Step 2: Convert the parametric curve into a curve image to facilitate manifold identification using image classification methods.

[0068] Based on the foregoing analysis, three characteristic parameters representing different flow patterns in two-phase flow were selected: first- and third-order resonance frequencies and gas-liquid density ratio. A method was employed to transform the dynamic curves of these parameters into dynamic curve images. The normalized images of the dynamic curves of the three characteristic parameters representing different flow patterns at each flow pattern experimental point were merged into a single image. Different colors were used in the image to represent different parameter curves, resulting in image sample datasets for different flow patterns, such as... Figure 4 As shown (where the density ratio is generally a horizontal straight line, which can be used as a reference line).

[0069] Traditional indirect flow pattern identification methods often extract feature vectors from parametric signals to establish flow identification models, offering advantages such as intuitive analysis, convenient processing, and wide applicability. However, traditional methods struggle to extract all appropriate features from parametric curves, and insufficient feature extraction fails to fully distinguish the details of various signal fluctuations. This proposed method, compared to traditional methods, fully utilizes the different features characterizing different flow patterns contained in dynamic curve images, thus improving the accuracy of flow pattern identification.

[0070] Step 3: Establish a manifold recognition model based on deep neural networks:

[0071] This scheme uses a deep neural network suitable for image processing to establish a manifold recognition model. The image sample datasets of different manifolds are used as the model input, and the image datasets are reasonably divided into training and testing sets. The manifold recognition model is fully trained and tested, and finally a manifold recognition model with high classification accuracy is obtained.

[0072] Compared to traditional neural network classification and support vector machine models, this deep neural network model does not require manual feature extraction and can automatically extract dynamic curve features of different manifolds from the pixels in the input image; deep neural networks can extract richer curve features and have better generalization ability.

[0073] like Figure 5As shown, this is the preferred deep neural network structure of AlexNet used in this embodiment. The model runs sequentially to extract image features. The image is first input from the input layer, and then deep and shallow feature information is extracted through continuously stacked convolutional layers. Since a large number of parameters are generated during convolution, pooling layers are used to reduce the data dimensionality and thus reduce the amount of data. Finally, the data flows to a fully connected layer to stretch the high-dimensional information into a one-dimensional vector and pass it to the output layer for classification and recognition of the extracted features.

[0074] Furthermore, 70% of the sample images for each manifold were used as the training set, and 30% were used as the test set for accuracy calculation. A manifold recognition model was built using the AlexNet deep neural network. The parameters of the selected deep neural network are as follows: Figure 6 Table 1 is shown.

[0075] like Figure 7 Table 2 shows the model test results, such as Figure 8 The image shows the confusion matrix of the AlexNet model test. The results indicate that the AlexNet manifold recognition network model has high recognition accuracy, with an overall recognition rate of 99.88%, and can be applied to manifold image recognition.

[0076] In summary, based on the measurement data of the dual-frequency Coriolis flowmeter, parameters that can characterize different flow patterns are selected and converted into parameter images. The dynamic features of the parameter images of different flow patterns are extracted using a deep neural network, thereby establishing a flow pattern identification model for gas-liquid two-phase flow in a horizontal pipe. The test results verify the effectiveness of the proposed method.

[0077] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0078] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying gas-liquid two-phase flow patterns based on a dual-frequency Coriolis flowmeter and a deep neural network, characterized in that, Includes the following steps: S1. Experimental testing and parameter selection: Using only one dual-frequency Coriolis flowmeter, various parameters output by the dual-frequency Coriolis flowmeter under different pressures and flow patterns were measured, and dynamic curves of parameters that can characterize different flow patterns were obtained. S2. Converting parametric curves into curve images: A processing method is adopted to convert dynamic parametric curves into curve images. The time-domain parametric curves containing different flow pattern characteristics are converted into curve images. The different parametric curves representing the flow pattern changes of each experimental data point are normalized and mapped onto the same image, thereby obtaining a sample dataset of curve images. S3. Establish a flow pattern recognition model: Use the curve image sample set obtained in step S2 as the model input, and use a deep neural network to establish a flow pattern recognition model. After training, it can accurately identify a variety of different flow patterns, which serves as the basis for measuring different flow patterns. S4. Manifold Recognition: New curve images are acquired in real time using steps S1-S2. These new curve images are directly input into a trained manifold recognition model based on a deep neural network for recognition, and the recognized manifold results are directly output.

2. The gas-liquid two-phase flow pattern identification method based on a dual-frequency Coriolis flowmeter and a deep neural network according to claim 1, characterized in that: In step S1, the different flow patterns of the gas-liquid two-phase flow include at least bubble flow, slug flow, laminar flow and plug flow.

3. The gas-liquid two-phase flow pattern identification method based on a dual-frequency Coriolis flowmeter and a deep neural network according to claim 2, characterized in that: In step S1, the parameters that the dual-frequency Coriolis flow meter can measure and output include at least the apparent mass flow rate, apparent density, first and third resonant frequencies, first and third damping, and amplitude.

4. The gas-liquid two-phase flow pattern identification method based on a dual-frequency Coriolis flowmeter and a deep neural network according to claim 3, characterized in that: In step S1, when measuring the gas-liquid two-phase flow using a dual-frequency Coriolis flowmeter, the responses of the two vibration modes are completely different. The different properties of the third-order modes provide different characteristics for flow pattern identification, and the time-domain characteristics of different flow patterns density vary significantly. Furthermore, the first and third-order resonant frequencies of the dual-frequency Coriolis flowmeter also have the following relationship with density: In the formula, c 11 c 12 and c 31 c 32 Both are called density measurement coefficients, ρ app1 and ρ app3 f1 and f3 are the apparent densities corresponding to the first and third order resonance frequencies, respectively.

5. The gas-liquid two-phase flow pattern identification method based on a dual-frequency Coriolis flowmeter and a deep neural network according to claim 1, characterized in that: In step S2, when acquiring the sample dataset of curve images, at least two characteristic parameters that can characterize different flow patterns of gas-liquid two-phase flow are selected, and the curves of different parameters are normalized and mapped onto the same image to form a flow pattern image.

6. The gas-liquid two-phase flow pattern identification method based on a dual-frequency Coriolis flowmeter and a deep neural network according to claim 1, characterized in that: In step S3, when training the manifold recognition model, the sample dataset of the curve image needs to be divided into a training set and a test set according to a certain ratio using a random method in advance, so as to fully train and test the manifold recognition model.

7. The gas-liquid two-phase flow pattern identification method based on a dual-frequency Coriolis flowmeter and a deep neural network according to claim 6, characterized in that: In step S3, when dividing the training set and the test set, the proportion of the test set to the training set is less than or equal to 3 / 7.

Images