CN107392250A - A kind of Method for Discriminating Gas-liquid Two Phase Flow - Google Patents

Abstract

The invention relates to a flow pattern identification method of gas-liquid two-phase flow. It includes calculating the interface wave signal based on the collected ultrasonic echo; decomposing the interface wave signal into several eigenmode functions according to the empirical mode decomposition method; extracting the energy percentage characteristics of the eigenmode function of the interface wave signal to obtain The energy of each eigenmode function component, and the energy of each eigenmode function component constitutes the eigenvector T; based on the normalization of the eigenvector, the total energy of the eigenmode function and the normalized characteristic Vector T'; by sending the feature vector T' as a training sample into the support vector machine for training, the flow classification model is obtained. The principle of the pulsed ultrasonic echo method is simple, and it is easy to measure the interface wave of the two-phase flow; the support vector machine method used is based on the statistical learning theory, which is suitable for the case of fewer samples. get a better recognition rate.

Description

A flow pattern recognition method for gas-liquid two-phase flow

technical field

The invention relates to a flow pattern identification method of gas-liquid two-phase flow.

Background technique

Many industrial production processes (power, petrochemical, metallurgy, etc.) involve gas-liquid two-phase flow phenomenon. The spatial distribution of the phase interface in a two-phase fluid medium is called the flow configuration or flow pattern. The flow pattern of two-phase flow is an important index to analyze the flow characteristics and heat transfer characteristics of two-phase flow, and the accurate identification of flow pattern also plays a decisive role in the measurement of various parameters of two-phase flow. Because the flow pattern of two-phase flow is affected by many factors, the flow pattern is complex and changeable, which not only creates resistance to the measurement of two-phase flow parameters, but also makes the design of field systems and equipment more complicated. The complexity and randomness of the two-phase flow system make it difficult to accurately identify the flow pattern, and the accurate identification of the flow pattern has always been an unsolved problem. As the industrial process has higher and higher requirements for two-phase flow measurement, energy saving and control, the demand for flow pattern identification of two-phase flow has become more and more urgent. The existing technologies are mainly flow diagram method, direct measurement method and indirect measurement method. Flow diagrams are obtained under different experimental conditions, have limitations and applicable conditions, and cannot be widely used in industrial processes. The disadvantage of the high-speed photography method in the direct measurement method is that transparent pipes or windows are required, and the complex reflection and refraction effects reduce the measurement accuracy; the radiation attenuation method uses radioactive sources, which is not conducive to storage and storage, and safety issues need special attention. The indirect measurement method needs to select parameters that can reflect the dynamic characteristics of the fluid. The differential pressure signal is widely used, but the differential pressure signal shows great randomness. Further signal analysis is needed to extract the characteristic parameters that can better identify the flow pattern, and use Signature information identifies flow patterns using intelligent classification algorithms. The selection of features largely determines the success rate of identifying flow patterns.

Contents of the invention

The present invention aims to solve the above problems, and provides a flow pattern identification method of gas-liquid two-phase flow. The principle of the pulse ultrasonic echo method adopted is simple, and it is easy to measure the interface wave of the two-phase flow; the interface wave signal has a strong correlation with the flow pattern feature extraction of the interface wave signal can be better used for flow pattern recognition; the support vector machine method used is based on statistical learning theory and is suitable for the case of fewer samples. Compared with neural network and other algorithms, it can be used in fewer training samples In the case of better recognition rate, the technical scheme adopted is as follows:

A flow pattern identification method of gas-liquid two-phase flow, comprising obtaining interface wave signals according to the collected ultrasonic echo propagation time and the propagation velocity of ultrasonic waves in liquid;

According to the empirical mode decomposition method, the interface wave signal is decomposed into several intrinsic mode functions;

By extracting the energy percentage characteristics of the first 8 intrinsic mode functions of the interface wave signal, the energy of each intrinsic mode function component is obtained, and the energy of each intrinsic mode function component is composed into a feature vector T;

Based on the normalization of the eigenvectors, the total energy of the first 8 eigenmode functions and the normalized eigenvector T′ are obtained;

By sending the feature vector T' as a training sample into the support vector machine for training, the flow classification model is obtained and the flow pattern is identified.

On the basis of the above technical scheme, the collected ultrasonic wave is emitted from the bottom of the pipeline under test, and the echo is received after being reflected by the interface of the two-phase flow fluid; the interface wave signal is calculated by the formula h=ct/2, and recorded as the interface wave signal is x(t).

On the basis of the above technical solutions, the interface wave signal is decomposed into several intrinsic mode functions according to the empirical mode decomposition method, including:

(1) Take all the maximum and minimum points of the interface wave signal, respectively use the cubic spline function to fit the upper envelope and the lower envelope, and the mean value of the upper and lower envelopes is m ₁ (t);

(2) Treat h ₁₁ (t)=x(t)-m ₁ (t) as a new signal x(t), repeat step (1), and pass k times of screening to make h _1k (t) satisfy the The number of extreme points and zero-crossing points is equal or the difference is at most 1, and the mean value of the local maximum envelope and local minimum envelope of the signal is 0. Record c ₁ (t) = h ₁ -k(t), which is the first intrinsic mode function;

(3) Taking r ₁ (t)=x(t)-c ₁ (t) as new data, repeat steps (1) and (2) until all eigenmode functions are decomposed, and finally the interface wave signal is decomposed are n eigenmode function functions and a residual component.

On the basis of the above technical scheme, the energy of each eigenmode function component is Eigenvector T=[E ₁ , E ₂ , E ₃ , E ₄ , E ₅ , E ₆ , E ₇ , E ₈ ].

On the basis of the above technical solutions, the total energy of the first 8 eigenmode functions Normalized feature vector T'=[E ₁ /E, E ₂ /E, E ₃ /E, E ₄ /E, E ₅ /E, E ₆ /E, E ₇ /E, E ₈ /E ].

On the basis of the above technical solution, by sending the feature vector T' as a training sample into the support vector machine for training, the steps to obtain the flow classification model and identify the flow pattern are as follows:

1) Prepare training samples {(X ₁ ,d ₁ ),(X ₂ ,d ₂ ),...,(X _n ,d _n )}, where X ₁ ,X ₂ ,...,X _n are features Vector, d ₁ , d ₂ ,...,d _n are target values;

2) In constraints Solve below so that the objective function

maximized αo _i

Where C is the selected penalty parameter, K(X _i ,X _j ) is the corresponding element of the i-th row and column j of the inner product kernel function K(x, _xi ) matrix, and the inner product kernel function is selected as the radial basis function

3) Calculate the optimal weight Y is the hidden layer output vector;

4) For the pattern X to be classified, calculate the classification discriminant function f(X) is 1 or -1, which determines the category of X;

5) Use the trained support vector machine to process the feature vector of the test sample to complete the recognition of the flow pattern.

The invention has the following advantages: the principle of the pulsed ultrasonic echo method is simple, and it is easy to measure the interface wave of two-phase flow; the interface wave signal has a strong correlation with the flow pattern, and the feature extraction of the interface wave signal can be better used for Flow pattern recognition; the support vector machine method used is based on statistical learning theory and is suitable for situations with fewer samples. Compared with algorithms such as neural networks, it can obtain better recognition rates with fewer training samples.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings that are required in the description of the embodiments or the prior art. Apparently, the drawings in the following description are only one embodiment of the present invention, and those skilled in the art can obtain other implementation attachments according to the provided drawings without creative work. picture.

Fig. 1: method flowchart of the present invention;

Figure 2: a schematic diagram of the ultrasonic measurement interface of the present invention;

Fig. 3: schematic diagram of laminar flow boundary wave signal obtained according to the present invention;

Fig. 4: Schematic diagram of interface wave signal of plug flow obtained according to the present invention;

Fig. 5: Schematic diagram of the interface wave signal of wavy flow obtained according to the present invention.

detailed description

Below in conjunction with accompanying drawing and embodiment the present invention will be further described:

As shown in Figures 1 to 5, a method for identifying a gas-liquid two-phase flow pattern in this embodiment includes obtaining the interface wave signal according to the collected ultrasonic echo propagation time and the propagation speed of the ultrasonic wave in the liquid;

According to the empirical mode decomposition method, the interface wave signal is decomposed into several intrinsic mode functions;

By extracting the energy percentage characteristics of the first 8 intrinsic mode functions of the interface wave signal, the energy of each intrinsic mode function component is obtained, and the energy of each intrinsic mode function component is composed into a feature vector T;

Based on the normalization of the eigenvectors, the total energy of the first 8 eigenmode functions and the normalized eigenvector T′ are obtained;

The flow type classification model is obtained by sending the feature vector T' as a training sample into the support vector machine for training.

Preferably, the ultrasonic wave collected is sent out from the bottom of the pipeline under test, and the echo is received after being reflected by the two-phase flow fluid interface; the interface wave signal is calculated by the formula h=ct/2, and the interface wave signal is recorded as x(t) .

Preferably, the interface wave signal is decomposed into several intrinsic mode functions according to the empirical mode decomposition method, including:

(1) Take all the maximum and minimum points of the interface wave signal, respectively use the cubic spline function to fit the upper envelope and the lower envelope, and the mean value of the upper and lower envelopes is m ₁ (t);

(2) Treat h ₁₁ (t)=x(t)-m ₁ (t) as a new signal x(t), repeat step (1), and pass k times of screening to make h _1k (t) satisfy the The number of extreme points and zero-crossing points is equal or the difference is at most 1, and the mean value of the local maximum envelope and local minimum envelope of the signal is 0. Record c ₁ (t) = h ₁ -k(t), which is the first intrinsic mode function;

(3) Taking r ₁ (t)=x(t)-c ₁ (t) as new data, repeat steps (1) and (2) until all eigenmode functions are decomposed, and finally the interface wave signal is decomposed are n eigenmode function functions and a residual component.

Preferably, the energy of each eigenmode function component is Eigenvector T=[E ₁ , E ₂ , E ₃ , E ₄ , E ₅ , E ₆ , E ₇ , E ₈ ].

Further, the total energy of the first 8 eigenmode functions Normalized eigenvectors

T'=[E ₁ /E, E ₂ /E, E ₃ /E, E ₄ /E, E ₅ /E, E ₆ /E, E ₇ /E, E ₈ /E].

Furthermore, by sending the feature vector T′ as a training sample into the support vector machine for training, the steps to obtain the flow pattern classification model and identify the flow pattern are as follows:

1) Prepare training samples {(X ₁ ,d ₁ ),(X ₂ ,d ₂ ),...,(X _n ,d _n )}, where X ₁ ,X ₂ ,...,X _n are features Vector, d ₁ , d ₂ ,...,d _n are target values;

2) In constraints Solve below so that the objective function

maximized αo _i

Where C is the selected penalty parameter, K(X _i ,X _j ) is the corresponding element of the i-th row and column j of the inner product kernel function K(x, _xi ) matrix, and the inner product kernel function is selected as the radial basis function

3) Calculate the optimal weight Y is the hidden layer output vector;

4) For the pattern X to be classified, calculate the classification discriminant function f(X) is 1 or -1, which determines the category of X;

5) Use the trained support vector machine to process the feature vector of the test sample to complete the recognition of the flow pattern.

6) Flow chart of support vector machine multi-classification algorithm

Table 1 The energy percentage eigenvectors of some samples of different flow patterns

Table 2 Flow pattern recognition results under laminar flow, wavy flow and plug flow conditions

The present invention has been described above by way of examples, but the present invention is not limited to the above specific embodiments, and any changes or modifications made based on the present invention fall within the scope of protection of the present invention.

Claims (6)

1. A gas-liquid two-phase flow pattern recognition method, characterized in that the method comprises: According to the collected ultrasonic echo propagation time and the propagation speed of ultrasonic waves in the liquid, the interface wave signal is obtained; According to the empirical mode decomposition method, the interface wave signal is decomposed into several intrinsic mode functions; By extracting the energy percentage characteristics of the first 8 intrinsic mode functions of the interface wave signal, the energy of each intrinsic mode function component is obtained, and the energy of each intrinsic mode function component is composed into a feature vector T; Based on the normalization of the eigenvectors, the total energy of the first 8 eigenmode functions and the normalized eigenvector T′ are obtained; By sending the feature vector T' as a training sample into the support vector machine for training, the flow classification model is obtained and the flow pattern is identified. 2. A method for identifying a gas-liquid two-phase flow pattern according to claim 1, characterized in that: the collected ultrasonic wave is sent from the bottom of the pipeline under test, and the echo is received after being reflected by the interface of the two-phase flow fluid; The formula h=ct/2 calculates the interface wave signal, and records the interface wave signal as x(t). 3. A method for identifying a gas-liquid two-phase flow pattern according to claim 1, wherein the interface wave signal is decomposed into several intrinsic mode functions according to the empirical mode decomposition method, including: (1) Take all the maximum and minimum points of the interface wave signal, respectively use the cubic spline function to fit the upper envelope and the lower envelope, and the mean value of the upper and lower envelopes is m ₁ (t); (2) Treat h ₁₁ (t)=x(t)-m ₁ (t) as a new signal x(t), repeat step (1), and pass k times of screening to make h _1k (t) satisfy the The number of extreme points and zero-crossing points is equal or the difference is at most 1 and the mean value of the local maximum envelope and local minimum envelope of the signal is 0, write c ₁ (t)=h ₁ -k(t) , is the first eigenmode function; (3) Taking r ₁ (t)=x(t)-c ₁ (t) as new data, repeat steps (1) and (2) until all eigenmode functions are decomposed, and finally the interface wave signal is decomposed are n eigenmode function functions and a residual component. 4. a kind of gas-liquid two-phase flow pattern identification method according to claim 1, is characterized in that, the energy of each intrinsic mode function component is Eigenvector T=[E ₁ , E ₂ , E ₃ , E ₄ , E ₅ , E ₆ , E ₇ , E ₈ ]. 5. A method for identifying gas-liquid two-phase flow pattern according to claim 1, characterized in that: the total energy of the first 8 eigenmode functions Normalized feature vector T'=[E ₁ /E, E ₂ /E, E ₃ /E, E ₄ /E, E ₅ /E, E ₆ /E, E ₇ /E, E ₈ /E ]. 6. A method for flow pattern identification of gas-liquid two-phase flow according to claim 1, characterized in that: by sending the feature vector T' as a training sample into a support vector machine for training, a flow pattern classification model is obtained and the flow pattern is identified The steps are: 1) Prepare training samples {(X ₁ ,d ₁ ),(X ₂ ,d ₂ ),...,(X _n ,d _n )}, where X ₁ ,X ₂ ,...,X _n are features Vector, d ₁ , d ₂ ,...,d _n are target values; 2) In constraints Solve below so that the objective function maximized αo _i Where C is the selected penalty parameter, K(X _i ,X _j ) is the corresponding element of the i-th row and column j of the inner product kernel function K(x, _xi ) matrix, and the inner product kernel function is selected as the radial basis function 3) Calculate the optimal weight Y is the hidden layer output vector; 4) For the pattern X to be classified, calculate the classification discriminant function f(X) is 1 or -1, which determines the category of X; 5) Use the trained support vector machine to process the feature vector of the test sample to complete the recognition of the flow pattern.