Multi-electrode two-phase flow parameter measurement method based on time channel fusion network

By using multi-electrode high-frequency capacitance sensors and time channels to fuse convolutional neural networks in two-phase flow measurements, the characteristics of two-phase flow signals are extracted, and the problem of low real-time and accuracy of two-phase flow parameter measurements in the prior art is solved. , realizes accurate measurement of gas-solid two-phase flow flow rate and oil-water two-phase flow moisture content.

CN115032244BActive Publication Date: 2024-11-26TIANJIN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210468333.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-29
Publication Date
2024-11-26
Estimated Expiration
2042-04-29

AI Technical Summary

Technical Problem

The existing two-phase flow parameter measurement methods have problems such as poor real-time performance and low metering accuracy, making it difficult to accurately measure the solid flow rate of gas-solid two-phase flow and the water content of oil-water two-phase flow.

Method used

The multi-electrode two-phase flow parameter measurement method based on the time channel fusion network is adopted. By constructing a multi-electrode high-frequency capacitance sensor, it can be converted into microwave differential frequency signals, and the signal can be extracted by using the time channel fusion convolutional neural network. The temporal and spatial characteristics of the system are characterized by fusion to achieve flow prediction and moisture content measurement.

Benefits of technology

It realizes rapid and accurate measurement of the solid flow rate of gas-solid two-phase flow and the water content of oil-water two-phase flow, and has real-time monitoring capabilities, which reduces the uncertainty of measurement and the subjectivity of human operations, and improves the accuracy and application range of measurement. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115032244B_ABST
    Figure CN115032244B_ABST
Patent Text Reader

Abstract

A multi-electrode two-phase flow parameter measurement method based on a time channel fusion network: construct a multi-electrode high-frequency capacitance sensor for measuring gas-solid or oil-water two-phase flow fluctuation signals; use a multi-electrode high-frequency capacitance sensor to measure the gas-solid or oil-water two-phase flow fluctuation signals and convert them into microwave difference frequency signals; construct a data set, specifically pre-process the microwave difference frequency signals, then use a sliding window to obtain samples from the data, add corresponding labels to the samples, and then use 80% of the samples as training sets, 10% of the samples as validation sets, and the remaining 10% of the samples as test sets; construct a time channel fusion convolutional neural network, extract features of the microwave difference frequency signals of the two-phase flow, including time features and spatial features, and fuse them to achieve the prediction of the flow rate of the gas-solid two-phase flow and the measurement of the water content of the oil-water two-phase flow. The present invention can be used for solid flow measurement of gas-solid two-phase flow and water content measurement of oil-water two-phase flow, and has a wide range of applications.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a two-phase flow parameter measurement method, and in particular to a multi-electrode two-phase flow parameter measurement method based on a time channel fusion network. Background Art

[0002] With the continuous development of modern science and technology, many industries have higher and higher requirements for the measurement accuracy and real-time performance of two-phase flow parameters, such as chemical, petroleum and many other industries. If the measurement accuracy and real-time performance of two-phase flow parameters can reach a certain level, it will have important guiding significance for the operation of these industries. At present, most of the measurement methods of two-phase flow parameters have the problems of poor real-time performance and low measurement accuracy, so accurate and rapid measurement of two-phase flow parameters is a current technical problem. Existing two-phase flow measurement methods such as optical method, tomography method, thermodynamic method, nuclear magnetic resonance, etc. are difficult to achieve real-time and accurate measurement of two-phase flow parameters.

[0003] In recent years, soft sensor models have also developed rapidly in the field of multiphase flow, especially the application of deep learning methods and artificial intelligence technology has broadened the application scope of multiphase flow soft sensor models. Deep learning technology is a new theory that has emerged in recent years. It extracts the features of the measured object layer by layer through unsupervised or supervised methods. Its features are highly objective and can accurately and comprehensively reflect the essence of the measured object. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a multi-electrode two-phase flow parameter measurement method based on a time channel fusion network that can be used for solid flow measurement of gas-solid two-phase flow and water content measurement of oil-water two-phase flow in order to overcome the shortcomings of the prior art.

[0005] The technical solution adopted by the present invention is: a multi-electrode two-phase flow parameter measurement method based on a time channel fusion network, comprising the following steps:

[0006] 1) Construct a multi-electrode high-frequency capacitance sensor for measuring gas-solid or oil-water two-phase flow fluctuation signals;

[0007] 2) Use a multi-electrode high-frequency capacitance sensor to measure the fluctuation signal of gas-solid or oil-water two-phase flow and convert it into a microwave difference frequency signal;

[0008] 3) Construct a data set, specifically preprocess the microwave difference frequency signal, then use a sliding window to obtain samples from the data, add corresponding labels to the samples, and then use 80% of the samples as a training set, 10% of the samples as a validation set, and the remaining 10% of the samples as a test set;

[0009] 4) Construct a time channel fusion convolutional neural network to extract the features of the microwave difference frequency signal of the two-phase flow, including time features and spatial features, and fuse them to predict the flow rate of the gas-solid two-phase flow and measure the water content of the oil-water two-phase flow.

[0010] The multi-electrode high-frequency capacitive sensor described in step 1) includes four electromagnetic excitation positive electrodes and four electromagnetic excitation negative electrodes installed at equal intervals on the same circumferential surface of the outer wall of the measuring pipe section of the non-conductive pipeline, and one electromagnetic excitation positive electrode is arranged adjacent to one electromagnetic excitation negative electrode, wherein one electromagnetic excitation positive electrode and one electromagnetic excitation negative electrode constitute a group of measuring electrode pairs, with a total of four groups of measuring electrode pairs, and each group of measuring electrode pairs is located on the same diameter on the radial cross-section of the measuring pipe section.

[0011] Step 2) includes:

[0012] When gas-solid or oil-water two-phase flow enters the measuring pipe section from the pipeline, the multi-electrode high-frequency capacitance sensor installed on the measuring pipe section measures the two-phase flow fluctuation signal; during measurement, the excitation signal is generated by the sinusoidal excitation signal source, one of which is sent to the detection module through the power divider, and the other is sent to the electromagnetic excitation positive electrode of each group of measuring electrode pairs of the multi-electrode high-frequency capacitance sensor to excite the gas-solid or oil-water two-phase flow, and the electromagnetic excitation negative electrode of each group of measuring electrode pairs sends the measured fluctuation signal under different flow rate conditions in the pipeline to the detection module, the excitation signal and the fluctuation signal enter the mixer for signal mixing, and the mixed signal is passed through the adder to obtain a microwave difference frequency signal; the four groups of measuring electrodes of the multi-electrode high-frequency capacitance sensor obtain a total of 4 microwave difference frequency signals.

[0013] Step 3) includes:

[0014] (3.1) Preprocess the microwave difference frequency signal, the formula is as follows:

[0015]

[0016] Among them, I i Represents the i-th data, I mean and I std are the mean and standard deviation of the data, is the i-th data after preprocessing;

[0017] The four microwave difference frequency signals are segmented by non-overlapping windowing, and the microwave difference frequency signals with window lengths of H and L are obtained. samples, among which, represents round down;

[0018] For gas-solid two-phase flow, the actual flow value is used as the label, and for oil-water two-phase flow, the actual water content value is used as the label, and a total of N fluctuation samples with label values ​​are obtained;

[0019] (3.2) Randomly divide N fluctuation samples with labeled values ​​into training set, validation set and test set to form a data set. The specific ratio is [training set: validation set: test set] = [8:1:1].

[0020] The time channel fusion convolutional neural network constructed in step 4) includes 16 network layers for feature extraction and parameter prediction, where:

[0021] The 1st to 13th layers are the feature extraction and fusion part, which includes four time-channel convolution modules and one attention layer, which are arranged in sequence to extract time-domain and space-domain features. The four time-channel convolution modules have the same structure, and the convolution kernel size decreases in the time domain direction to avoid excessive expansion of the receptive field.

[0022] Each time-channel convolution module has 3 layers, the first of which is the time convolution layer, which uses N T Convolution kernels, the formula of the temporal convolution layer is as follows:

[0023]

[0024] in is the output of the temporal convolutional layer, is the input of the temporal convolutional layer, f T is the activation function of the temporal convolutional layer, B T is the size of the convolution kernel of the temporal convolution layer, Z T is the dimension of the temporal convolution layer input, u is the index of the midpoint of the convolution kernel of the temporal convolution layer, v is the index of the temporal convolution layer input, C T is the index of the convolution kernel, P T and Q T is the position index of the temporal convolutional layer output, SE T is the step size of the temporal convolutional layer, and is the Cth temporal convolutional layer T The weights and biases of the convolution kernels. The temporal convolution layer is used to extract the time domain features of the signal, and the convolution kernel is calculated in the time dimension;

[0025] The second layer is the channel convolution layer, which uses N C Convolution kernel, the formula of channel convolution layer is as follows:

[0026]

[0027] in is the output of the channel convolutional layer, is the input of the channel convolution layer, f C is the activation function of the channel convolution layer, BC is the size of the convolution kernel of the channel convolution layer, Z C is the dimension of the channel convolution layer input, m is the index of the convolution kernel midpoint of the channel convolution layer, n is the index of the temporal convolution layer input, C C is the index of the convolution kernel, P C and Q C is the position index of the channel convolution layer output, SE C is the stride of the channel convolution layer, and is the Cth channel convolution layer C The weights and biases of the convolution kernels. The channel convolution layer is used to extract the spatial features of the signal, and the convolution kernel is calculated on the channel degree;

[0028] The third layer is the pooling layer, and the formula is described as follows:

[0029]

[0030] in, is the feature output by the pooling layer, is the input of the pooling layer, K P is the size of the pooling kernel of the pooling layer, C P ,P P ,Q P is the index of the pooling layer feature, O P is the index of the midpoint of the pooling kernel; this layer compresses the features extracted by the convolution layer, and uses average pooling to integrate local features to obtain higher-level features;

[0031] The attention layer assigns different attention weights to the output features of the pooling layer according to their different importance. The formula is described as follows:

[0032] y at =V at tanh(w at *X at +b at )

[0033] Where tanh is a nonlinear activation function, V at is the weight vector after activation by the activation function, w at represents the weight parameter matrix, b at is the bias term;

[0034] The 14th to 16th layers are the output part, including a flattening layer, a dense layer and an output layer in sequence. The flattening layer expands the attention-weighted features into one-dimensional features; the dense layer is provided with 50 neurons, and Relu is used as an activation function for fusion analysis of the features expanded into one dimension. The Sigmoid activation function is set in the output layer to realize the numerical measurement of the flow parameters.

[0035] In the time channel fusion convolutional neural network,

[0036] Setting hyperparameters, wherein the hyperparameters are obtained by using the AMSGrad optimization algorithm to update the weights of the time channel fusion convolutional neural network through gradient back-up based on the training set and the validation set;

[0037] The mean square error, mean absolute error and mean absolute percentage error are used as loss functions to measure the gap between the output of the time channel fusion convolutional neural network and the true label. Minimizing the gap is used as the criterion to guide network training and optimization in the right direction.

[0038] The multi-electrode two-phase flow parameter measurement method based on the time channel fusion network of the present invention has the advantages of:

[0039] 1. The multi-electrode high-frequency capacitance sensor used has a small time constant and fast response speed. It can perform real-time monitoring and dynamic processing, and can quickly and accurately obtain sequence fluctuation signals. The multi-electrode groups are symmetrically distributed on the pipe wall, and the fluctuation information of different cross-sectional positions of the pipeline is measured at different angles, which can accurately reflect the frequency changes of the material during the flow process.

[0040] 2. The present invention uses the above method to measure four groups of fluctuation signals measured at different angles at different positions through a multi-electrode high-frequency capacitance sensor, takes the channel as a node, considers both the relationship between the nodes and the characteristics of the nodes themselves, and inputs the time channel fusion network for learning, which can capture the basic characteristics and laws of the changes in fluctuation information.

[0041] 3. The time channel fusion network fully exploits the temporal and spatial characteristics of the two-phase flow signal, and performs effective fusion to effectively extract the characteristics of the two-phase flow signal.

[0042] 4. The deep neural network used in the present invention is objective in parameter prediction, eliminating the uncertainty in the test and the subjectivity of human operation. After being trained with a large amount of data and iterations, the time channel fusion network can have strong and objective prediction performance for this type of data.

[0043] 5. The present invention can be used to measure the solid flow rate of gas-solid two-phase flow and the water content of oil-water two-phase flow, and has a wide range of applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the structure of the multi-electrode high-frequency capacitance sensor in the present invention;

[0045] Figure 2 It is a structural schematic diagram of the time channel fusion convolutional neural network in the present invention. DETAILED DESCRIPTION

[0046] The multi-electrode two-phase flow parameter measurement method based on the time channel fusion network of the present invention is described in detail below in conjunction with the embodiments and drawings.

[0047] The multi-electrode two-phase flow parameter measurement method based on the time channel fusion network of the present invention uses a multi-electrode high-frequency capacitance sensor to collect the fluctuation information of the two-phase flow in the pipeline, uses the time channel fusion network to fully explore the time characteristics and spatial characteristics of the two-phase flow signal and predict the parameters, and uses a supervised learning method to use the actual value as a label to predict the gas-solid two-phase flow rate or the water content of the oil-water two-phase flow. Specifically, it includes the following steps:

[0048] 1) Construct a multi-electrode high-frequency capacitance sensor for measuring gas-solid or oil-water two-phase flow fluctuation signals;

[0049] like Figure 1 As shown, the multi-electrode high-frequency capacitance sensor includes four electromagnetic excitation positive electrodes 1, 3, 5, 7 and four electromagnetic excitation negative electrodes 2, 4, 6, 8 installed at equal intervals on the same circumferential surface of the outer wall of the measuring pipe section 9 of the non-conductive pipeline, and one electromagnetic excitation positive electrode is arranged adjacent to one electromagnetic excitation negative electrode, wherein one electromagnetic excitation positive electrode and one electromagnetic excitation negative electrode constitute a group of measuring electrode pairs, and there are four groups of measuring electrode pairs in total, and each group of measuring electrode pairs is located on the same diameter on the radial cross section of the measuring pipe section 9.

[0050] 2) Use a multi-electrode high-frequency capacitance sensor to measure the fluctuation signal of gas-solid or oil-water two-phase flow and convert it into a microwave difference frequency signal; including:

[0051] When the gas-solid or oil-water two-phase flow enters the measuring pipe section 9 from the pipeline, the multi-electrode high-frequency capacitance sensor installed on the measuring pipe section 9 measures the two-phase flow fluctuation signal; the fluctuation information is obtained by measuring the frequency value returned by each pair of electrodes, and the fluctuation of the frequency value reflects the cross-sectional fluctuation information of the two-phase flow in the pipeline; during measurement, the excitation signal is generated by the sinusoidal excitation signal source, which is sent to the detection module through the power divider in one way and to the electromagnetic excitation positive electrode of each group of measuring electrode pairs of the multi-electrode high-frequency capacitance sensor in the other way to excite the gas-solid or oil-water two-phase flow, and the electromagnetic excitation negative electrode of each group of measuring electrode pairs sends the measured fluctuation signal under different flow rate conditions in the pipeline to the detection module, and the excitation signal and the fluctuation signal enter the mixer for signal mixing, and the mixed signal is passed through the adder to obtain a microwave difference frequency signal; each group of measuring electrode pairs respectively uses an independent circuit to measure the fluctuation information of the two-phase flow in the pipeline, and the four groups of measuring electrodes of the multi-electrode high-frequency capacitance sensor obtain a total of 4 microwave difference frequency signals.

[0052] Specifically, four groups of measuring electrode pairs use a cyclic sampling method to measure the gas-solid or oil-water two-phase flow fluctuation signal. The measurement strategy is: within a sampling period, four groups of measuring electrode pairs sample sequentially. After completing one cycle of sampling, the next cycle of sampling is performed. Each group of measuring electrode pairs can collect one flow fluctuation time series signal, and a total of four multi-electrode high-frequency capacitance sensors can be used to measure the water content fluctuation signal. During the measurement process, the sampling period is set to 0.001 seconds.

[0053] The gas-solid or oil-water two-phase flow fluctuation signal is obtained by the frequency value returned by each set of measuring electrode pairs. The fluctuation of the frequency value reflects the fluctuation of the cross-sectional water content of the gas-solid or oil-water two-phase flow fluctuation signal in the pipeline.

[0054] 3) Construct a data set, specifically preprocess the microwave difference frequency signal, then use a sliding window to obtain samples from the data, add corresponding labels to the samples, and then use 80% of the samples as a training set, 10% of the samples as a validation set, and the remaining 10% of the samples as a test set; including:

[0055] (3.1) Preprocess the microwave difference frequency signal, the formula is as follows:

[0056]

[0057] Among them, I i Represents the i-th data, I mean and I std are the mean and standard deviation of the data, is the i-th data after preprocessing;

[0058] The four microwave difference frequency signals are segmented by non-overlapping windowing, and the microwave difference frequency signals with window lengths of H and L are obtained. samples, among which, represents round down;

[0059] For gas-solid two-phase flow, the actual flow value is used as the label, and for oil-water two-phase flow, the actual water content value is used as the label, and a total of N fluctuation samples with label values ​​are obtained;

[0060] (3.2) Randomly divide N fluctuation samples with labeled values ​​into training set, validation set and test set to form a data set. The specific ratio is [training set: validation set: test set] = [8:1:1].

[0061] 4) Construct a time channel fusion convolutional neural network to extract features of the microwave difference frequency signal of the two-phase flow, including time features and spatial features, and fuse them to predict the flow rate of the gas-solid two-phase flow and measure the water content of the oil-water two-phase flow; the constructed time channel fusion convolutional neural network includes 16 network layers for feature extraction and parameter prediction, among which,

[0062] The 1st to 13th layers are the feature extraction and fusion part, which includes four time-channel convolution modules and one attention layer, which are arranged in sequence to extract time-domain and space-domain features. The four time-channel convolution modules have the same structure, and the convolution kernel size decreases in the time domain direction to avoid excessive expansion of the receptive field.

[0063] Each time-channel convolution module has 3 layers, the first of which is the time convolution layer, which uses N T Convolution kernels, the formula of the temporal convolution layer is as follows:

[0064]

[0065] in is the output of the temporal convolutional layer, is the input of the temporal convolutional layer, f T is the activation function of the temporal convolutional layer, B T is the size of the convolution kernel of the temporal convolution layer, Z T is the dimension of the temporal convolution layer input, u is the index of the midpoint of the convolution kernel of the temporal convolution layer, v is the index of the temporal convolution layer input, C T is the index of the convolution kernel, P T and Q T is the position index of the temporal convolutional layer output, SE T is the step size of the temporal convolutional layer, and is the Cth temporal convolutional layer T The weights and biases of the convolution kernels. The temporal convolution layer is used to extract the time domain features of the signal, and the convolution kernel is calculated in the time dimension;

[0066] The second layer is the channel convolution layer, which uses N C Convolution kernel, the formula of channel convolution layer is as follows:

[0067]

[0068] in is the output of the channel convolutional layer, is the input of the channel convolution layer, f C is the activation function of the channel convolution layer, B C is the size of the convolution kernel of the channel convolution layer, Z Cis the dimension of the channel convolution layer input, m is the index of the convolution kernel midpoint of the channel convolution layer, n is the index of the temporal convolution layer input, C C is the index of the convolution kernel, P C and Q C is the position index of the channel convolution layer output, SE C is the stride of the channel convolution layer, and is the Cth channel convolution layer C The weights and biases of the convolution kernels. The channel convolution layer is used to extract the spatial features of the signal, and the convolution kernel is calculated on the channel degree;

[0069] The third layer is the pooling layer, and the formula is described as follows:

[0070]

[0071] in, is the feature output by the pooling layer, is the input of the pooling layer, K P is the size of the pooling kernel of the pooling layer, C P ,P P ,Q P is the index of the pooling layer feature, O P is the index of the midpoint of the pooling kernel; this layer compresses the features extracted by the convolution layer, and uses average pooling to integrate local features to obtain higher-level features;

[0072] The attention layer assigns different attention weights to the output features of the pooling layer according to their different importance. The formula is described as follows:

[0073] y at =V at tanh(w at *X at +b at )

[0074] Where tanh is a nonlinear activation function, V at is the weight vector after activation by the activation function, w at represents the weight parameter matrix, b at is the bias term;

[0075] The 14th to 16th layers are the output part, including a flattening layer, a dense layer and an output layer in sequence. The flattening layer expands the attention-weighted features into one-dimensional features; the dense layer is provided with 50 neurons, and Relu is used as an activation function for fusion analysis of the features expanded into one dimension. The Sigmoid activation function is set in the output layer to realize the numerical measurement of the flow parameters.

[0076] In the time channel fusion convolutional neural network,

[0077] Setting hyperparameters, wherein the hyperparameters are obtained by using the AMSGrad optimization algorithm to update the weights of the time channel fusion convolutional neural network through gradient back-up based on the training set and the validation set;

[0078] The mean square error (MSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) are used as loss functions to measure the gap between the output of the time channel fusion convolutional neural network and the true label. Minimizing the gap is used as the criterion to guide network training and optimization in the right direction.

Claims

1. A multi-electrode two-phase flow parameter measurement method based on a time channel fusion network, characterized in that: The steps include: 1) Construct a multi-electrode high-frequency capacitance sensor for measuring gas-solid or oil-water two-phase flow fluctuation signals; The multi-electrode high-frequency capacitance sensor comprises four electromagnetic excitation positive electrodes (1, 3, 5, 7) and four electromagnetic excitation negative electrodes (2, 4, 6, 8) which are installed at equal intervals on the same circumferential surface of the outer wall of a measuring pipe section (9) of a non-conductive pipeline, and one electromagnetic excitation positive electrode is arranged adjacent to one electromagnetic excitation negative electrode, wherein one electromagnetic excitation positive electrode and one electromagnetic excitation negative electrode form a group of measuring electrode pairs, and there are four groups of measuring electrode pairs in total, and each group of measuring electrode pairs is located on the same diameter on the radial cross section of the measuring pipe section (9); 2) Use a multi-electrode high-frequency capacitance sensor to measure the fluctuation signal of gas-solid or oil-water two-phase flow and convert it into a microwave difference frequency signal; including: When a gas-solid or oil-water two-phase flow enters a measuring pipe section (9) from a pipeline, a multi-electrode high-frequency capacitance sensor installed on the measuring pipe section (9) measures the two-phase flow fluctuation signal; during measurement, an excitation signal is generated by a sinusoidal excitation signal source, one of which is sent to a detection module through a power divider, and the other is sent to the electromagnetic excitation positive electrode of each group of measuring electrode pairs of the multi-electrode high-frequency capacitance sensor to excite the gas-solid or oil-water two-phase flow, and the electromagnetic excitation negative electrode of each group of measuring electrode pairs sends the measured fluctuation signal under different flow rate conditions in the pipeline to the detection module, and the excitation signal and the fluctuation signal enter a mixer for signal mixing, and the mixed signal is passed through an adder to obtain a microwave difference frequency signal; four groups of measuring electrodes of the multi-electrode high-frequency capacitance sensor obtain a total of four microwave difference frequency signals; 3) Construct a data set, specifically preprocess the microwave difference frequency signal, then use a sliding window to obtain samples from the data, add corresponding labels to the samples, and then use 80% of the samples as a training set, 10% of the samples as a validation set, and the remaining 10% of the samples as a test set; 4) Construct a time channel fusion convolutional neural network to extract the features of the microwave difference frequency signal of the two-phase flow, including the time features and spatial features, and fuse them to predict the flow rate of the gas-solid two-phase flow and measure the water content of the oil-water two-phase flow; The constructed time channel fusion convolutional neural network includes 16 network layers for feature extraction and parameter prediction, among which, The 1st to 13th layers are the feature extraction and fusion part, which includes four time-channel convolution modules and one attention layer, which are arranged in sequence to extract time-domain and space-domain features. The four time-channel convolution modules have the same structure, and the convolution kernel size decreases in the time domain direction to avoid excessive expansion of the receptive field. Each time-channel convolution module has 3 layers, the first of which is the time convolution layer. The time convolution layer uses N T Convolution kernels, the formula of the temporal convolution layer is as follows: in is the output of the temporal convolutional layer, is the input of the temporal convolutional layer, f T is the activation function of the temporal convolutional layer, B T is the size of the convolution kernel of the temporal convolution layer, Z T is the dimension of the temporal convolution layer input, u is the index of the midpoint of the convolution kernel of the temporal convolution layer, v is the index of the temporal convolution layer input, C T is the index of the convolution kernel, P T and Q T is the position index of the temporal convolutional layer output, SE T is the step size of the temporal convolutional layer, and is the Cth temporal convolutional layer T The weights and biases of the convolution kernels; the temporal convolution layer is used to extract the time domain features of the signal, and the convolution kernel is calculated in the time dimension; The second layer is the channel convolution layer, which uses N C Convolution kernel, the formula of channel convolution layer is as follows: in is the output of the channel convolutional layer, is the input of the channel convolution layer, f C is the activation function of the channel convolution layer, B C is the size of the convolution kernel of the channel convolution layer, Z C is the dimension of the channel convolution layer input, m is the index of the convolution kernel midpoint of the channel convolution layer, n is the index of the temporal convolution layer input, C C is the index of the convolution kernel, P C and Q C is the position index of the channel convolution layer output, SE C is the stride of the channel convolution layer, and is the Cth channel convolution layer C The weights and biases of the convolution kernels; the channel convolution layer is used to extract the spatial features of the signal, and the convolution kernel is calculated on the channel degree; The third layer is the pooling layer, and the formula is described as follows: in, is the feature output by the pooling layer, is the input of the pooling layer, K P is the size of the pooling kernel of the pooling layer, C P ,P P ,Q P is the index of the pooling layer feature, O P is the index of the midpoint of the pooling kernel; this layer compresses the features extracted by the convolution layer, and uses average pooling to integrate local features to obtain higher-level features; The attention layer assigns different attention weights to the output features of the pooling layer according to their different importance. The formula is described as follows: y at =V at tanh(w at *X at +b at ) Where tanh is a nonlinear activation function, V at is the weight vector after activation by the activation function, w at represents the weight parameter matrix, b at is the bias term; The 14th to 16th layers are the output part, including a flattening layer, a dense layer and an output layer in sequence. The flattening layer expands the attention-weighted features into one-dimensional features; the dense layer is provided with 50 neurons, and Relu is used as an activation function to perform fusion analysis on the features expanded into one dimension. The output layer is provided with a Sigmoid activation function to realize the numerical measurement of flow parameters; In the time channel fusion convolutional neural network, hyperparameters are set. The hyperparameters are obtained by using the AMSGrad optimization algorithm to update the weights of the time channel fusion convolutional neural network through gradient reverse based on the training set and the validation set; mean square error, mean absolute error and mean absolute percentage error are used as loss functions to measure the gap between the output of the time channel fusion convolutional neural network and the true label, and minimizing the gap is used as the criterion to guide network training and optimization in the right direction.

2. The multi-electrode two-phase flow parameter measurement method based on time channel fusion network according to claim 1 is characterized in that: Step 3) includes: (3.1) Preprocess the microwave difference frequency signal, the formula is as follows: Among them, I i Represents the i-th data, I mean and I std are the mean and standard deviation of the data, is the i-th data after preprocessing; The four microwave difference frequency signals are segmented by non-overlapping windowing, and the microwave difference frequency signals with window lengths of H and L are obtained. samples, among which, represents round down; For gas-solid two-phase flow, the actual flow value is used as the label, and for oil-water two-phase flow, the actual water content value is used as the label, and a total of N fluctuation samples with label values ​​are obtained; (3.2) Randomly divide N fluctuation samples with labeled values ​​into training set, validation set and test set to form a data set. The specific ratio is [training set: validation set: test set] = [8:1:1].