A method for measuring the concentration of gas-liquid two-phase flow by integrating multiple sensors

Abstract

The application relates to a kind of integrated multi-sensor gas-liquid two-phase flow concentration measurement methods.Integrated multi-sensor, including double helix high-frequency microwave sensor and spin into vortex metering device, signal acquisition is carried out to gas-liquid two-phase flow under different working conditions, wherein based on the fluid fluctuation information measured by double helix high-frequency microwave sensor, based on the fluid flow and density information measured by spin into vortex metering device, double-flow fusion network is proposed for multi-sensor signal fusion and feature extraction, for realizing the measurement of gas-liquid two-phase flow gas concentration under different working conditions.

Description

Technical Field

[0001] This invention relates to a method for measuring the holdup of a gas-liquid two-phase flow, and more particularly to a method for measuring the holdup of a gas-liquid two-phase flow that integrates multiple sensors. Background Technology

[0002] Gas-liquid two-phase flow has wide applications in industrial settings, spanning multiple industries including petrochemicals, energy, pharmaceuticals, and food processing. Applications include gas-liquid separators, oil-water separators, and pipeline systems. In these contexts, accurate measurement of gas-liquid two-phase flow parameters is crucial for the stability, safety, and economic efficiency of industrial production.

[0003] In recent years, deep learning models have been widely used in sensor information fusion and parameter prediction due to their ability to extract elusive patterns and information from complex sensor data. Deep learning models can identify and utilize multi-level data features, thereby achieving more accurate parameter measurements than traditional methods. Deep learning-based methods for measuring complex parameters in gas-liquid two-phase flow possess multiple advantages, including high precision, real-time performance, data-driven decision-making, cost reduction, adaptability, and automation, and have broad application prospects. Summary of the Invention

[0004] The purpose of this invention is to provide an integrated multi-sensor method for measuring the gas-liquid two-phase flow holdup. Multiple sensors, including a double-helix high-frequency microwave sensor and a vortex metering device, are integrated to acquire signals from the gas-liquid two-phase flow under different operating conditions. Fluid fluctuation information is measured based on the double-helix high-frequency microwave sensor, and fluid flow rate and density are measured based on the vortex metering device. A dual-flow fusion network is proposed to fuse multi-sensor signals and extract features, enabling the measurement of the gas holdup of the gas-liquid two-phase flow under different operating conditions.

[0005] The technical solution adopted in this invention is: a method for measuring the holdup of a gas-liquid two-phase flow, particularly a method for measuring the holdup of a gas-liquid two-phase flow integrating multiple sensors. Its characteristics are as follows:

[0006] (1) Construct a double-helix high-frequency microwave sensor and a vortex metering device for measuring the water content fluctuation signal of gas-liquid two-phase flow.

[0007] The aforementioned double-helix high-frequency microwave sensor includes a measuring tube section, an excitation electrode, a receiving electrode, a protective electrode, and a shielding layer. The excitation electrode and the receiving electrode are rotated 360 degrees to the outside of the measuring tube in a wall-mounted structure. Two synchronous spiral protective electrodes are placed between the excitation electrode and the receiving electrode, and a shielding layer is installed outside the electrodes to prevent the electric field from spreading to the edge, thereby protecting the electric field and enhancing the electric field strength inside the tube.

[0008] The aforementioned vortex metering device includes a measuring pipe section, a throttling device, a differential pressure flow meter, and a vortex flow meter. The throttling device and the vortex flow meter are installed on the measuring pipe, and the throttling device is connected to the differential pressure flow meter.

[0009] (2) Conduct dynamic experiments to obtain multi-sensor signals. Specifically, a double-helix high-frequency microwave sensor is used to measure the gas-liquid two-phase flow fluctuation signal and convert it into a microwave difference frequency signal. At the same time, a vortex metering device is used to obtain flow rate and density signals.

[0010] When the gas-liquid two-phase flow passes through the measuring pipe section of the double-helix high-frequency microwave sensor, the sinusoidal excitation signal source generates a high-frequency excitation signal. This signal is sent to the excitation electrode of the helical high-frequency microwave sensor and to the monitor via a power divider. The excitation electrode emits high-frequency electromagnetic waves into the gas-liquid two-phase flow in the pipe. Due to the difference in the content of polar water molecules in the gas-liquid two-phase flow under different operating conditions, the absorbed wave energy is different. The sensing electrode transmits this different signal caused by the difference in the content of polar water molecules to the monitor. The two signals are mixed in the monitor and then processed by an adder to obtain a microwave difference frequency signal.

[0011] When the gas-liquid two-phase flow passes through the measuring pipe section of the vortex metering device, the total flow rate Q1 of the fluid in the measuring pipe is obtained based on the throttling device and the differential pressure flow meter.

[0012]

[0013] In the formula, u is the flow coefficient of the differential pressure flowmeter, S0 is the orifice area, ρ is the fluid density, and ΔP is the differential pressure value. Based on the precession frequency f of the pressure inside the pipe and the total flow rate Q2 of the fluid inside the pipe measured by the vortex flowmeter:

[0014]

[0015] In the formula K x The flow coefficient of the vortex flowmeter is given by the following formula: Since the flow rates Q1 and Q2 measured by the differential pressure flowmeter and the vortex flowmeter installed on the same pipeline are the same, a density signal ρ is obtained through calculation.

[0016]

[0017] (3) Constructing the dataset involves preprocessing the data, then using a sliding window to extract samples, adding corresponding labels to the samples, and using 80% of the samples as the training set and 20% as the test set. Specifically, this includes:

[0018] The microwave difference frequency signal and density signal are preprocessed separately using the following formulas:

[0019]

[0020] Among them, I o Representing the o-th data, I mean and I std These are the mean and standard deviation of the data, respectively. This is the o-th data point after preprocessing; the preprocessed microwave difference frequency signal and density signal are simultaneously segmented using a sliding window with a window length of H, and data acquisition is performed with a length of P. 100 samples, of which 100 samples 200 samples 300 samples 400 samples 500 samples 600 samples 700 samples 800 samples 900 samples 1 ... The value represents rounding down, and the actual gas content is used as the data label, resulting in N samples with labeled values. The N fluctuating samples are randomly divided into a dataset with a training set and a test set, with the specific ratio being [training set: test set] = [8:2].

[0021] (4) A dual-flow fusion network is proposed for multi-sensor signal fusion and feature extraction to measure the gas content of gas-liquid two-phase flow under different operating conditions. The dual-flow fusion network includes a parallel feature module, a feature fusion module, and a parameter prediction module.

[0022] The parallel feature module consists of two identical branches, with microwave difference frequency signal and density signal as inputs, respectively, and features are extracted from each signal separately. Each branch of the parallel feature module is a two-stream network structure, where one stream uses a larger convolutional kernel and the other uses a smaller convolutional kernel, comprehensively extracting features from different scales or dimensions.

[0023] Each stream of the two-stream network structure consists of three one-dimensional convolutional modules, two max-pooling layers, and one dropout layer. Each one-dimensional convolutional module consists of a convolutional layer, a batch normalization layer, and an activation function layer.

[0024] The convolutional layer is used to extract temporal features of the signal, and the convolutional kernel used in the convolutional layer is J. k There are three convolutional modules, where k = 1, 2, and 3, corresponding to three one-dimensional convolutional modules. The formula for the convolutional layer is as follows:

[0025]

[0026]

[0027] Where X is the input of the convolutional layer, Y is the output feature of the convolutional layer, f is the activation function of the convolutional layer, u is the index of the midpoint of the convolutional kernel, G is the feature dimension of the input of the convolutional layer, g is the feature index of the input of the temporal convolutional layer, and ω p，g，k and b p L represents the p-th convolutional kernel and bias in the convolutional layer, respectively. kS is the size of the convolution kernel, S is the stride of the temporal convolution kernel, in is the abbreviation of the subscript of the temporal convolution input, and m and n are the position indices of the extracted features.

[0028] The Batch Normalization layer standardizes the features to avoid numerical instability in neural networks, making the distribution of features in the same batch similar and enhancing the network's trainability; the activation function layer uses the ReLU activation function.

[0029] The max pooling layer is used to select the maximum value within the pooling region as a representative value to achieve feature dimensionality reduction and reduce computational burden. The dropout layer randomly discards a portion of neurons and their connections according to probability, forcing the neural network model to learn more global features.

[0030] The feature fusion module calculation formula is as follows:

[0031] fe = F 1 ||F 2

[0032] fe′=f D (fe*w D +b D )

[0033] Where F 1 F 2 f is the input feature to be fused, fe′ is the output feature, and f D For the activation function, w D b D These represent the weights and biases of the fully connected layer, respectively; the parameter prediction module uses the Sigmoid activation function to achieve numerical measurement of water content.

[0034] The hyperparameters of the dual-stream fusion network are optimized using the AMSGrad optimization algorithm. The weight parameters of the dual-stream fusion network are updated based on the training set. The mean squared error (MSE) is used as the loss function to measure the error between the output of the dual-stream fusion network and the true content. The network parameters are optimized with minimizing the error as the criterion.

[0035] The present invention has the following advantages due to the adoption of the above technical solutions:

[0036] This invention proposes a multi-sensor method for measuring the holdup of gas-liquid two-phase flow. It integrates multiple sensors, including a double-helix high-frequency microwave sensor and a vortex metering device, to collect signals from the gas-liquid two-phase flow under different operating conditions. Based on the vortex metering device, information such as flow rate and density can be directly measured. By fusing the characteristic information of different sensors, and capturing flow information from multiple angles based on microwave and vortex, the accuracy, reliability and real-time performance of the measurement are improved.

[0037] This invention proposes a dual-stream fusion method based on multi-sensor signals for signal fusion and analysis. It uses convolution kernels of different sizes to comprehensively extract features from different scales or dimensions, thereby improving measurement accuracy. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of a double-helix high-frequency microwave sensor and a vortex metering device.

[0039] Figure 2 This is a diagram of a dual-stream fusion network structure. Detailed Implementation

[0040] The present invention will now be described in detail with reference to the embodiments and accompanying drawings.

[0041] This invention discloses an integrated multi-sensor method for measuring the gas-liquid two-phase flow holdup. The method integrates multiple sensors, including a double-helix high-frequency microwave sensor and a vortex metering device, to collect signals from the gas-liquid two-phase flow under different operating conditions. The double-helix high-frequency microwave sensor measures fluid fluctuation information, while the vortex metering device measures fluid flow rate and density. A dual-flow fusion network is constructed to fuse multi-sensor signals and extract features, enabling the measurement of the gas holdup in the gas-liquid two-phase flow under different operating conditions.

[0042] The present invention provides a method for measuring the holdup of a gas-liquid two-phase flow using an integrated multi-sensor system, comprising the following steps:

[0043] (1) Construct a double-helix high-frequency microwave sensor and a vortex metering device for measuring the water content fluctuation signal of gas-liquid two-phase flow.

[0044] The aforementioned double-helix high-frequency microwave sensor includes a measuring tube section, an excitation electrode, a receiving electrode, a protective electrode, and a shielding layer. The excitation electrode and the receiving electrode are rotated 360 degrees to the outside of the measuring tube in a wall-mounted structure. Two synchronous spiral protective electrodes are placed between the excitation electrode and the receiving electrode, and a shielding layer is installed outside the electrodes to prevent the electric field from spreading to the edge, thereby protecting the electric field and enhancing the electric field strength inside the tube.

[0045] The aforementioned vortex metering device includes a measuring pipe section, a throttling device, a differential pressure flow meter, and a vortex flow meter. The throttling device and the vortex flow meter are installed on the measuring pipe, and the throttling device is connected to the differential pressure flow meter.

[0046] (2) Conduct dynamic experiments to obtain multi-sensor signals. Specifically, a double-helix high-frequency microwave sensor is used to measure the gas-liquid two-phase flow fluctuation signal and convert it into a microwave difference frequency signal. At the same time, a vortex metering device is used to obtain flow rate and density signals.

[0047] When the gas-liquid two-phase flow passes through the measuring pipe section of the double-helix high-frequency microwave sensor, the sinusoidal excitation signal source generates a high-frequency excitation signal. This signal is sent to the excitation electrode of the helical high-frequency microwave sensor and to the monitor via a power divider. The excitation electrode emits high-frequency electromagnetic waves into the gas-liquid two-phase flow in the pipe. Due to the difference in the content of polar water molecules in the gas-liquid two-phase flow under different operating conditions, the absorbed wave energy is different. The sensing electrode transmits this different signal caused by the difference in the content of polar water molecules to the monitor. The two signals are mixed in the monitor and then processed by an adder to obtain a microwave difference frequency signal.

[0048] When the gas-liquid two-phase flow passes through the measuring pipe section of the vortex metering device, the total flow rate Q1 of the fluid in the measuring pipe is obtained based on the throttling device and the differential pressure flow meter.

[0049]

[0050] In the formula, u is the flow coefficient of the differential pressure flowmeter, S0 is the orifice area, ρ is the fluid density, and ΔP is the differential pressure value. Based on the precession frequency f of the pressure inside the pipe and the total flow rate Q2 of the fluid inside the pipe measured by the vortex flowmeter:

[0051]

[0052] In the formula K x The flow coefficient of the vortex flowmeter is given by the following formula: Since the flow rates Q1 and Q2 measured by the differential pressure flowmeter and the vortex flowmeter installed on the same pipeline are the same, a density signal ρ is obtained through calculation.

[0053]

[0054] (3) Constructing the dataset involves preprocessing the data, using a sliding window to extract samples, adding corresponding labels to the samples, and then using 80% of the samples as the training set and 20% as the test set. This includes:

[0055] The microwave difference frequency signal and density signal are preprocessed separately using the following formulas:

[0056]

[0057] Among them, I o Representing the o-th data, I mean and I std These are the mean and standard deviation of the data, respectively. This is the o-th data point after preprocessing; the preprocessed microwave difference frequency signal and density signal are simultaneously segmented using a sliding window with a window length of H, and data acquisition is performed with a length of P. 100 samples, of which 100 samples 200 samples 300 samples 400 samples 500 samples 600 samples 700 samples 800 samples 900 samples 1 ... The value represents rounding down, and the actual gas content is used as the data label, resulting in N samples with labeled values. The N fluctuating samples are randomly divided into a dataset with a training set and a test set, with the specific ratio being [training set: test set] = [8:2].

[0058] (4) A dual-flow fusion network is proposed for multi-sensor signal fusion and feature extraction to measure the gas content of gas-liquid two-phase flow under different operating conditions. The dual-flow fusion network includes a parallel feature module, a feature fusion module, and a parameter prediction module.

[0059] The parallel feature module consists of two identical branches, taking microwave difference frequency signal and density signal as inputs, and extracting features from each signal separately. Each branch of the parallel feature module is a two-stream network structure, with one stream using a larger convolutional kernel and the other using a smaller convolutional kernel, comprehensively extracting features from different scales or dimensions.

[0060] Each stream of the two-stream network structure consists of three one-dimensional convolutional modules, two max-pooling layers, and one dropout layer. Each one-dimensional convolutional module consists of a convolutional layer, a batch normalization layer, and an activation function layer.

[0061] The convolutional layer is used to extract temporal features of the signal, and the convolutional kernel used in the convolutional layer is J. k There are three convolutional modules, where k = 1, 2, and 3, corresponding to three one-dimensional convolutional modules. The formula for the convolutional layer is as follows:

[0062]

[0063]

[0064] Where x is the input of the convolutional layer, Y is the output feature of the convolutional layer, f is the activation function of the convolutional layer, u is the index of the midpoint of the convolutional kernel, G is the feature dimension of the input of the convolutional layer, g is the feature index of the input of the temporal convolutional layer, and ω p，g，k and b p L represents the p-th convolutional kernel and bias in the convolutional layer, respectively. k S is the size of the convolution kernel, S is the stride of the temporal convolution kernel, in is the abbreviation of the subscript of the temporal convolution input, and m and n are the position indices of the extracted features.

[0065] The Batch Normalization layer standardizes the features to avoid numerical instability in neural networks, making the distribution of features in the same batch similar and enhancing the network's trainability; the activation function layer uses the ReLU activation function.

[0066] The max pooling layer is used to select the maximum value within the pooling region as a representative value to achieve feature dimensionality reduction and reduce computational burden. The dropout layer randomly discards a portion of neurons and their connections according to probability, forcing the neural network model to learn more global features.

[0067] The calculation formula for the feature fusion module is as follows:

[0068] fe = F 1 ||F 2

[0069] fe′=f D (fe*w D +b D )

[0070] Where F 1 F 2 f is the input feature to be fused, fe′ is the output feature, and f D For the activation function, w D b D These represent the weights and biases of the fully connected layer, respectively; the parameter prediction module uses the Sigmoid activation function to achieve numerical measurement of water content.

[0071] The hyperparameters of the dual-stream fusion network are optimized using the AMSGrad optimization algorithm. The weight parameters of the dual-stream fusion network are updated based on the training set. The mean squared error (MSE) is used as the loss function to measure the error between the output of the dual-stream fusion network and the true content. The network parameters are optimized with minimizing the error as the criterion.

Claims

1. A method for measuring the holdup of a gas-liquid two-phase flow integrating multiple sensors, characterized in that, Includes the following steps: (1) Construct a double-helix high-frequency microwave sensor and a vortex metering device for measuring the water content fluctuation signal of gas-liquid two-phase flow; (2) Dynamic experiments were conducted to obtain multi-sensor signals. Specifically, a double-helix high-frequency microwave sensor was used to measure the gas-liquid two-phase flow fluctuation signal and convert it into a microwave difference frequency signal. At the same time, a vortex metering device was used to obtain the flow rate and density signals. When the gas-liquid two-phase flow passed through the pipe section measured by the double-helix high-frequency microwave sensor, a sinusoidal excitation signal source generated a high-frequency excitation signal. One path was sent to the excitation electrode of the spiral high-frequency microwave sensor through a power divider, and the other path was sent to the monitor. The excitation electrode emitted high-frequency electromagnetic waves into the gas-liquid two-phase flow in the pipe. Due to the difference in the content of polar water molecules in the gas-liquid two-phase flow under different working conditions, the absorbed wave energy was different. The sensing electrode transmitted this different signal caused by the difference in the content of polar water molecules to the monitor. The two signals were mixed in the monitor and then processed by an adder to obtain a microwave difference frequency signal. When a gas-liquid two-phase flow passes through the measuring pipe section of the vortex metering device, the total flow rate of the fluid in the measuring pipe is obtained based on the throttling device and the differential pressure flow meter. : In the formula The flow coefficient of the differential pressure flow meter. The area of ​​the throttling orifice. For fluid density, It is a differential pressure value, based on the precession frequency of the pressure inside the pipe measured by a vortex flowmeter. And measuring the total flow rate of fluid in the pipe : In the formula The flow coefficient of the vortex flowmeter is determined by the difference in flow rate measured by the differential pressure flowmeter and the vortex flowmeter installed on the same pipeline. and Similarly, a density signal is obtained through calculation. The formula is as follows: ； (3) Constructing a dataset involves preprocessing the data, using a sliding window to extract samples from the data, adding corresponding labels to the samples, and then using 80% of the samples as the training set and 20% of the samples as the test set. (4) A dual-stream fusion network is proposed to perform multi-sensor signal fusion and feature extraction, which is used to measure the gas content of gas-liquid two-phase flow under different working conditions. The dual-stream fusion network includes a parallel feature module, a feature fusion module and a parameter prediction module. The parallel feature module consists of two identical branches, with microwave difference frequency signal and density signal as inputs. Feature extraction is performed on the two signals respectively. Each branch of the parallel feature module is a dual-stream network structure, in which one stream uses a larger convolution kernel and the other stream uses a smaller convolution kernel, comprehensively extracting features from different scales or dimensions.

2. The method for measuring the holdup of a gas-liquid two-phase flow integrating multiple sensors according to claim 1, characterized in that, Step (1) includes: The aforementioned double-helix high-frequency microwave sensor includes a measuring tube section, an excitation electrode, a receiving electrode, a protective electrode, and a shielding layer. The excitation electrode and the receiving electrode maintain a wall-mounted structure that rotates 360 degrees outside the measuring tube. Two synchronously spiral protective electrodes are placed between the excitation electrode and the receiving electrode, and a shielding layer is installed outside the electrodes to prevent the electric field from spreading to the edge, thereby protecting the electric field and enhancing the electric field strength inside the tube. The aforementioned vortex metering device includes a measuring pipe section, a throttling device, a differential pressure flow meter, and a vortex flow meter. The throttling device and the vortex flow meter are installed on the measuring pipe, and the throttling device is connected to the differential pressure flow meter.

3. The method for measuring the holdup of a gas-liquid two-phase flow integrating multiple sensors according to claim 1, characterized in that, Step (3) includes: The microwave difference frequency signal and density signal are preprocessed separately using the following formulas: in, Represents the o-th data. and These are the mean and standard deviation of the data, respectively. This is the o-th data point after preprocessing; the preprocessed microwave difference frequency signal and density signal are simultaneously segmented using a sliding window with a window length of H, and data acquisition is performed with a length of P. 100 samples, of which 100 samples 200 samples 300 samples 400 samples 500 samples 600 samples 700 samples 800 samples 900 samples 1 ... The value represents rounding down, and the actual gas content is used as the data label, resulting in N samples with labeled values. The N fluctuating samples are randomly divided into a dataset with a training set and a test set, with the specific ratio being [training set: test set] = [8:2].

4. The method for measuring the holdup of a gas-liquid two-phase flow integrating multiple sensors according to claim 1, characterized in that, Each stream of the dual-stream network structure described in step (4) consists of three one-dimensional convolutional modules, two max-pooling layers, and one dropout layer. Each one-dimensional convolutional module consists of a convolutional layer, a batch normalization layer, and an activation function layer. The convolutional layers are used to extract temporal features of the signal, and the convolutional kernels used in the convolutional layers are... One, of which These correspond to three one-dimensional convolutional modules, and the formulas for the convolutional layers are as follows: in, This is the input to the convolutional layer. The output features of the convolutional layer, The activation function for the convolutional layer. This is the index of the midpoint of the convolution kernel. The feature dimension input to the convolutional layer, The feature index is the input to the temporal convolutional layer. and These are the convolutional layers of the th Each convolutional kernel and bias, The size of the convolution kernel. The stride of the convolutional kernel in the temporal convolutional layer. It is an abbreviation for the subscript of the input of the temporal convolutional layer. The BatchNormalization layer is used to standardize features, avoiding numerical instability in neural networks and ensuring that features in the same batch are similarly distributed, thus enhancing the network's trainability. The activation function layer uses the ReLU activation function. The max pooling layer is used to select the maximum value within the pooling region as a representative value to achieve feature dimensionality reduction and reduce computational burden. The dropout layer randomly discards a portion of neurons and their connections according to probability, forcing the neural network model to learn more global features. The feature fusion module calculation formula is as follows: in The input features to be fused are... For output features, For activation function, These are the weights and biases of the fully connected layer, respectively. The parameter prediction module uses the Sigmoid activation function to achieve numerical measurement of water content.

5. The method for measuring the holdup of a gas-liquid two-phase flow using an integrated multi-sensor system according to claim 1, characterized in that, The hyperparameters of the dual-stream fusion network are optimized using the AMSGrad optimization algorithm. The weight parameters of the dual-stream fusion network are updated based on the training set. The mean squared error (MSE) is used as the loss function to measure the error between the output of the dual-stream fusion network and the true content. The network parameters are optimized with minimizing the error as the criterion.

Images