A method for measuring oil-water two-phase flow parameters of a dual-flow differential network

Abstract

The application relates to a measurement method for measuring parameters of oil-water two-phase flow, in particular to an oil-water two-phase flow parameter measurement method based on a double-flow differential network, which is composed of a high-precision oil-water two-phase flow sensor and a soft measurement model. The designed oil-water two-phase flow sensor is used to acquire parameter fluctuation signals in the oil-water two-phase flow process, then a structured deep learning network is used to extract and analyze the fluctuation signals collected by the sensor, and finally corrected oil-water two-phase flow parameters can be obtained.

Description

Technical Field

[0001] A method for measuring parameters of oil-water two-phase flow, particularly a method for measuring parameters of oil-water two-phase flow based on a two-flow differential network. Technical Background

[0002] Oil-water two-phase flow refers to the simultaneous flow of oil and water within the same pipeline, and it is widely used in industries such as petroleum, chemical, and power. Accurate measurement of oil-water two-phase flow parameters is crucial for the control and optimization of production processes. Traditional methods for measuring oil-water two-phase flow include Venturi flow meters and turbine flow meters, but these methods suffer from low measurement accuracy and susceptibility to fluid properties. Therefore, developing a high-precision, interference-resistant method for measuring oil-water two-phase flow parameters is of significant practical importance.

[0003] Deep learning-based soft measurement methods have broadened the application scope of soft measurement models for multiphase flows. Deep learning technology, a relatively new theory that has emerged in recent years, extracts features of the measured object layer by layer through unsupervised or supervised methods. These features are highly objective and can accurately and comprehensively reflect the essence of the measured object. Soft measurement techniques can fuse data measured by sensors, and intelligent and deep learning methods can accurately and efficiently extract feature information of multiphase flows. Summary of the Invention

[0004] A method for measuring parameters of oil-water two-phase flow, particularly a method for measuring parameters of oil-water two-phase flow based on a two-flow differential network.

[0005] The water content of the oil-water two-phase flow is measured by using a spiral electrode sensor. Then, the flow information features in the measurement signal are extracted and fused using a dual-flow differential network to achieve accurate measurement of the water content of the oil-water two-phase flow.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0007] 1. A method for measuring parameters of oil-water two-phase flow, particularly a method for measuring parameters of oil-water two-phase flow based on a two-flow differential network. The method is characterized by the following steps:

[0008] (1) The fluctuation signal of oil-water two-phase flow parameters was measured using a spiral electrode sensor;

[0009] (2) Construct a two-flow differential network model to realize the feature fusion of oil-water two-phase flow parameter fluctuation signals and content measurement.

[0010] 2. A method for measuring parameters of an oil-water two-phase flow, particularly a method for measuring parameters of an oil-water two-phase flow based on a two-flow differential network. The method is characterized in that step (1) includes:

[0011] When material enters the measuring pipe section through the pipeline, a spiral electrode sensor installed on the pipeline measures the fluctuation signal of the oil-water two-phase flow parameters. The designed electrostatic sensor consists of a spiral electrode, an insulating layer, and the material pipeline. First, based on the working principle of microwave sensors, we designed a spiral sensing electrode, located in the middle of the pipeline. Then, a high-frequency excitation source excites one side of the spiral sensing electrode, while the other side receives the signal through an AD8302 module. Simultaneously, the entire sensor is covered with a shielding layer made of copper sheets. The spiral electrode sensor can effectively acquire the measurement signal of the oil-water two-phase flow. This sensor utilizes the principle of electrostatic induction to capture the mass flow rate information of the oil-water two-phase flow from a microscopic perspective.

[0012] In actual material flow processes, factors such as gas pressure and material pressure exist. Therefore, we selected PEEK as the material for the material pipeline to meet pressure resistance requirements. Simultaneously, to meet the requirements of high flow rates, we set the pipeline inner diameter to DN50, and chose copper as the electrode material. The model diagram is attached. Figure 1 As shown, the two electrodes are 7 cm long and 15 cm high. Experiments show that this sensor can effectively detect dynamic changes in the flow through the spiral electrode, achieving accurate measurement of dynamic parameters of oil-water two-phase flow.

[0013] 3. A method for measuring parameters of oil-water two-phase flow, particularly a method for measuring parameters of oil-water two-phase flow based on a deep adaptive network.

[0014] The characteristic feature is that step (2) includes:

[0015] ① The constructed two-phase flow differential network model first preprocesses the oil-water two-phase flow signal measured by the spiral electrode sensor. The fluctuating signal is segmented using a non-overlapping windowing method, and sample signals with window length H and length L are obtained. 10 samples, of which This represents rounding down. The fluctuation signal acquired by the sensor is windowed and truncated, with the actual value as the data label, resulting in N parameter fluctuation samples with labeled values. The windowed signal is then subjected to discrete wavelet decomposition to obtain N preprocessed signals.

[0016] ② Randomly divide the original signal and the 2N samples after discrete wavelet decomposition into a dataset with the specific ratio of [training set: validation set: test set] = [8:1:1].

[0017] 4. A method for measuring parameters of oil-water two-phase flow, particularly a method for measuring parameters of oil-water two-phase flow based on a deep adaptive network.

[0018] The characteristic feature is that step (2) includes:

[0019] ① The constructed two-stream differential network model is characterized by its structure, consisting of two parts: a two-stream autoencoder network and a multi-scale differential fusion network. The two-stream autoencoder network comprises the original data stream and the discrete wavelet data stream. Encoded features are extracted using the autoencoder network. The multi-scale differential fusion network decouples feature attributes under different operating conditions by fusing multi-scale differential features from different levels. Simultaneously, the multilayer perceptron model fuses feature differences from intermediate multi-scale layers to obtain the final content prediction result. Its structure is shown in the attached figure. Figure 2 As shown.

[0020] ② The constructed two-stream differential network model is characterized by its network structure, where the autoencoder model employs a deep residual neural network, primarily comprising five convolutional modules that encode the signal into high-dimensional feature representations. These high-dimensional features are then sampled by a decoder composed of deconvolutional modules to the corresponding feature sizes in the encoder model. Each of the five convolutional modules is 3×3 in size, and each of the five deconvolutional modules is 3×3 in size. The autoencoder process is as follows:

[0021]

[0022] Where f(x) is the encoder and g(h) is the decoder. The signal is reconstructed through the decoding function. During the pre-training of the autoencoder model, the signal is first normalized and input into the encoder model f(x) to extract its high-dimensional feature representation. Then, the reconstructed signal is obtained through the decoder model g(h). Model training uses the mean squared error (MSE) loss function to iteratively update the model's parameters. The formula for the loss function is:

[0023]

[0024] Where, x i,j and Let n and m be the values ​​of the original and generated signals, respectively, and n and m be the signal dimensions. After the pre-trained model has finished training, the encoder is used to extract the high-dimensional feature representation of the signal. First, the signal is input into the encoder model to construct the original signal flow network, and the high-dimensional feature distribution is obtained through the encoder. Similarly, the wavelet decomposed signal is input into the encoder model to construct the wavelet decomposition network, and the mixed feature distribution is obtained.

[0025] ③ The constructed two-stream differential network model is characterized by its network structure, where the multi-scale differential fusion network mainly consists of seven local differential networks. The differential features of the input layer of the two-stream encoder network and the five intermediate layers and encoding layer of the convolutional neural network are input into the corresponding local differential network model. Finally, the results of each differential network are concatenated and passed through a multilayer perceptron module to output the prediction result. The method for calculating the differential features at each scale is shown in the following formula.

[0026]

[0027]

[0028] y = f mlp (z d )

[0029] Where y is the prediction result. and For the i-th convolutional layer in a two-stream system, For the differential features of the i-th layer, Indicates the difference features of the original signal, To encode the differences in features, the dimension of each difference feature corresponds to the dimension of the convolutional layer, and the parameter θ ranges from [0,1]. The influence of Euclidean distance and absolute distance on the difference features is determined by adjusting the change of the θ parameter.

[0030] 5. An intelligent method for measuring the flow rate of oil-water two-phase flow, particularly a method for measuring the flow rate of oil-water two-phase flow based on a two-flow differential network model, the advantages of which are:

[0031] (1) The spiral electrode sensor used has a fast response speed, can perform real-time monitoring and dynamic processing, and can quickly and accurately obtain the content sequence fluctuation signal.

[0032] (2) The dual-flow differential network used in this invention is objective in predicting content and has strong predictive performance for oil-water two-phase flow data. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the spiral electrode sensor structure of the present invention;

[0034] Figure 2 This is the two-stream differential network model of the present invention. Detailed Implementation

[0035] The present invention will be further illustrated below with reference to the embodiments. The embodiments described below are illustrative and not limiting, and should not be used to limit the scope of protection of the present invention.

[0036] The oil-water two-phase flow prediction method involved in this invention uses a spiral electrode sensor to collect the fluctuation information of the oil-water two-phase flow content in the pipeline. The collected flow fluctuation signal is used as the input of the two-phase flow differential network model for feature extraction and fusion. A supervised learning method is adopted, with the actual water content as the label, to predict the oil-water two-phase flow content.

[0037] The oil-water two-phase flow measurement method involved in this invention uses a spiral electrode sensor to measure the multi-element flow fluctuation signal in the pipeline. The internal structure of the instrument used for oil-water two-phase flow measurement is shown in the attached figure. Figure 1 As shown, when material enters the measuring pipe section through the pipeline, a spiral electrode sensor installed on the pipeline measures the fluctuation signal of the oil-water two-phase flow parameters. The designed electrostatic sensor consists of a spiral electrode, an insulating layer, and a material pipeline. First, based on the working principle of microwave sensors, we designed a spiral induction electrode, located in the middle of the pipeline. Then, a high-frequency excitation source excites one side of the spiral induction electrode, while the other side receives the signal through an AD8302 module. Simultaneously, the entire sensor is covered with a shielding layer made of copper sheets. The spiral electrode sensor can effectively acquire the measurement signal of the oil-water two-phase flow. This sensor utilizes the principle of electrostatic induction to capture the mass flow rate information of the oil-water two-phase flow from a microscopic perspective.

[0038] In actual material flow processes, factors such as gas pressure and material pressure exist. Therefore, we selected PEEK as the material for the material pipeline to meet pressure resistance requirements. Simultaneously, to meet the requirements of high flow rates, we set the pipeline inner diameter to DN50, and chose copper as the electrode material. The model diagram is attached. Figure 1 As shown, the two electrodes are 7 cm long and 15 cm high. Experiments show that this sensor can effectively detect dynamic changes in the flow through the spiral electrode, achieving accurate measurement of dynamic parameters of oil-water two-phase flow.

[0039] In this invention, the collected flow fluctuation signal is used as input to a deep learning model, and the flow rate of the measured oil-water two-phase flow can be obtained through model calculation. During the measurement process, the sampling period is set to 0.001 seconds.

[0040] The present invention relates to an intelligent flow metering method for oil-water two-phase flow, particularly a flow metering method for oil-water two-phase flow based on a multi-network feature fusion model:

[0041] (1) Collect measurement data, preprocess the data, and construct a dataset.

[0042] ① The constructed two-phase flow differential network model first preprocesses the oil-water two-phase flow signal measured by the spiral electrode sensor. The fluctuating signal is segmented using a non-overlapping windowing method, and sample signals with window length H and length L are obtained. 10 samples, of which This represents rounding down. The fluctuation signal acquired by the sensor is windowed and truncated, with the actual value as the data label, resulting in N parameter fluctuation samples with labeled values. The windowed signal is then subjected to discrete wavelet decomposition to obtain N preprocessed signals.

[0043] ② Randomly divide the original signal and the 2N samples after discrete wavelet decomposition into a dataset with the specific ratio of [training set: validation set: test set] = [8:1:1].

[0044] (2) Constructing a two-stream differential network model

[0045] ① The constructed two-stream differential network model is characterized by its structure, consisting of two parts: a two-stream autoencoder network and a multi-scale differential fusion network. The two-stream autoencoder network comprises the original data stream and the discrete wavelet data stream. Encoded features are extracted using the autoencoder network. The multi-scale differential fusion network decouples feature attributes under different operating conditions by fusing multi-scale differential features from different levels. Simultaneously, the multilayer perceptron model fuses feature differences from intermediate multi-scale layers to obtain the final content prediction result. Its structure is shown in the attached figure. Figure 2 As shown.

[0046] ② The constructed two-stream differential network model is characterized by its network structure, where the autoencoder model employs a deep residual neural network, primarily comprising five convolutional modules that encode the signal into high-dimensional feature representations. These high-dimensional features are then sampled by a decoder composed of deconvolutional modules to the corresponding feature sizes in the encoder model. Each of the five convolutional modules is 3×3 in size, and each of the five deconvolutional modules is 3×3 in size. The autoencoder process is as follows:

[0047]

[0048] Where f(x) is the encoder and g(h) is the decoder. The signal is reconstructed through the decoding function. During the pre-training of the autoencoder model, the signal is first normalized and input into the encoder model f(x) to extract its high-dimensional feature representation. Then, the reconstructed signal is obtained through the decoder model g(h). Model training uses the mean squared error (MSE) loss function to iteratively update the model's parameters. The formula for the loss function is:

[0049]

[0050] Where, x i,j and Let n and m be the values ​​of the original and generated signals, respectively, and n and m be the signal dimensions. After the pre-trained model has finished training, the encoder is used to extract the high-dimensional feature representation of the signal. First, the signal is input into the encoder model to construct the original signal flow network, and the high-dimensional feature distribution is obtained through the encoder. Similarly, the wavelet decomposed signal is input into the encoder model to construct the wavelet decomposition network, and the mixed feature distribution is obtained.

[0051] ③ The constructed two-stream differential network model is characterized by its network structure, where the multi-scale differential fusion network mainly consists of seven local differential networks. The differential features of the input layer of the two-stream encoder network and the five intermediate layers and encoding layer of the convolutional neural network are input into the corresponding local differential network model. Finally, the results of each differential network are concatenated and passed through a multilayer perceptron module to output the prediction result. The method for calculating the differential features at each scale is shown in the following formula.

[0052]

[0053]

[0054] y = f mlp (z d )

[0055] Where y is the prediction result. and For the i-th convolutional layer in a two-stream system, For the differential features of the i-th layer, Indicates the difference features of the original signal, To encode the differences in features, the dimension of each difference feature corresponds to the dimension of the convolutional layer, and the parameter θ ranges from [0,1]. The influence of Euclidean distance and absolute distance on the difference features is determined by adjusting the change of the θ parameter.

[0056] The advantages of the oil-water two-phase flow content measurement method designed in this invention are:

[0057] (1) The spiral electrode sensor used has a fast response speed, can perform real-time monitoring and dynamic processing, and can quickly and accurately obtain the content sequence fluctuation signal.

[0058] (2) The dual-flow differential network used in this invention is objective in predicting content and has strong predictive performance for oil-water two-phase flow data.

Claims

1. A method for measuring parameters of oil-water two-phase flow based on a two-flow differential network, characterized in that... The following steps are required: (1) The fluctuation signal of the oil-water two-phase flow parameters was measured using a spiral electrode sensor; (2) Construct a dual-stream differential network model to realize feature fusion and content measurement of oil-water two-phase flow parameter fluctuation signals. The dual-stream differential network model consists of two parts: a dual-stream autoencoder network and a multi-scale differential fusion network. The dual-stream autoencoder network includes the original data stream and the discrete wavelet data stream. The autoencoder network is used to extract the encoding features. The multi-scale differential fusion network decouples the feature attributes under different working conditions by fusing multi-scale differential features at different levels. At the same time, the multilayer perceptron model fuses the feature differences of the multi-scale intermediate layers to obtain the final content prediction result.

2. The method for measuring oil-water two-phase flow parameters based on a two-flow differential network according to claim 1, characterized in that, Step (1) includes: When the material enters the measuring pipe section through the pipeline, the oil-water two-phase flow parameter fluctuation signal is measured by the spiral electrode sensor installed on the pipeline. The spiral electrode sensor consists of a spiral electrode, an insulating layer, and a material pipeline. First, we designed the spiral sensing electrode based on the working principle of the microwave sensor. The spiral sensing electrode is located in the middle of the pipeline. Then, a high-frequency excitation source excites the spiral sensing electrode on one side, and the other side receives the signal through the AD8302 module. At the same time, the entire sensor is covered with a shielding layer made of copper sheet. The oil-water two-phase flow measurement signal is effectively collected through the spiral electrode sensor.

3. The method for measuring oil-water two-phase flow parameters based on a two-flow differential network according to claim 1, characterized in that, Step (2) includes: ① The constructed two-phase flow differential network model first preprocesses the oil-water two-phase flow signal measured by the spiral electrode sensor. The fluctuating signal is segmented using a non-overlapping windowing method, and sample signals with a window length of H and a length of L can be obtained. 10 samples, of which The function represents rounding down. The fluctuation signal collected by the sensor is windowed and truncated, and the actual value is used as the data label. A total of N parameter fluctuation samples with label values ​​are obtained. The signal after sliding windowing is decomposed into discrete wavelet decomposition to obtain N preprocessed signals. ② Randomly divide the original signal and the 2N samples after discrete wavelet decomposition into a dataset with the specific ratio of [training set: validation set: test set] = [8:1:1].

4. The method for measuring oil-water two-phase flow parameters based on a two-flow differential network according to claim 1, characterized in that, Step (2) includes: ① The two-stream differential network model includes a two-stream autoencoder network and a multi-scale differential fusion network. The two-stream autoencoder network includes the original data stream and the discrete wavelet data stream. The autoencoder network adopts a deep residual neural network. The multi-scale differential fusion network consists of seven local differential networks, which fuse the differential features corresponding to the input layer, five intermediate convolutional layers and the encoding layer, and output the content prediction result through a multilayer perceptron. ② The constructed two-stream differential network model is characterized by the following network structure: the autoencoder model in the two-stream autoencoder network adopts a deep residual neural network, including five convolutional modules and five deconvolutional modules. The five convolutional modules are used to encode the signal into a high-dimensional feature representation, and the decoder composed of the five deconvolutional modules is used to sample the high-dimensional features to the same size as the corresponding features in the encoder model. The size of each of the five convolutional modules is 3×3, and the size of each of the five deconvolutional modules is 3×3. The autoencoding process is as follows: in For encoder, For decoder, The signal is reconstructed through the decoding function. During the pre-training process of the autoencoder model, the signal is first normalized and input into the encoder model. Extract high-dimensional feature representations of the signal, and then pass them through a decoder model. After obtaining the reconstructed signal, the model training iterates through parameter updates using the mean squared error (MSE) loss function. The formula for the loss function is as follows: in, and The values ​​of the original signal and the generated signal are given. and The size of the signal; ③ The constructed two-stream differential network model is characterized by the following network structure: the multi-scale differential fusion network consists of seven local differential networks. The differential features of the input layer of the two-stream encoder network and the five intermediate layers and the encoding layer of the convolutional neural network are input into the corresponding local differential network model. Finally, the results of each differential network are concatenated and passed through a multilayer perceptron module to output the prediction result. The method for calculating the differential features at each scale is shown in the following formula. in, For the predicted results, and For Shuangliu Middle School Features of each convolutional layer For the first Layer differences Indicates the difference features of the original signal, To encode the differences in features, the dimension of each difference feature corresponds to the dimension of the convolutional layer, and the parameters... The range is [0,1], and it can be adjusted by... The effects of parameter variations on the Euclidean distance and absolute distance on the difference features are determined.

Images