CN1904580A - Gas liquid two phase flow type identification method based on information coalescence and flow type signal collection device - Google Patents

Abstract

The invention relates to a gas phase/ liquid phase flow type identification method and flow type signal collection device based on information fusion. The feather is that: taking denoising processing to pressure difference fluctuation signal between the collected different pressure sampling distance, using wavelet theory to take feature distilling to signal and form three wavelet package information entropy eigenvectors that would realize the mapping from feature space to flow type space, using the outputs of the sub-neural network as the independent evidence identities and using D-S evidence theory to take information fusion to gain the identification result. The device includes horizontal tubes, pressure sampling ring, pressure sampling tube, pressure difference transmitter, data collection card and computer. The pressure difference signal would be sent to computer through data collecting card, and the computer would accomplish the data process for the pressure difference signal to realize the collection of the flow type signal of the gas phase/ liquid phase. The advantages of the invention is that it has high recognition rate, strong commonality and is suitable for different tube diameter and gas phase/ liquid phase flow system of different medium.

Description

Method for Discriminating Gas-liquid Two Phase Flow and flow pattern signal pickup assembly based on information fusion

Technical field

The present invention relates to the biphase gas and liquid flow field of measuring technique, relate in particular to a kind of Method for Discriminating Gas-liquid Two Phase Flow and flow pattern signal pickup assembly based on information fusion.

Background technology

The diphasic flow system extensively exists in fields such as oil, chemical industry, metallurgy, power, and the distribution situation of phase medium is called flow pattern in the two-phase flow, and different flow patterns has different flowing and heat transfer characteristic, influences the accurate measurement of other parameters of two-phase flow simultaneously.Therefore the research of flow pattern identification is the important foundation problem of two-phase flow research always, have important industrial application value and learning value, and the safety and automated production, the design of pipe system and the exploitation of operation and two-phase flow meter that can be related industries provide technical support.

Traditional flow type identification method has two big classes: a class is to adopt experimental technique to make flow regime map; Another kind of is the transformation criterion relational expression that the analysis according to the convection transformation mechanism obtains, and utilizes on-the-spot flow parameter to determine flow pattern.Because flow parameter needs the problem that solves often in production reality, therefore traditional method can't be used widely.

The modern surveying method of flow pattern can be divided into direct measurement and indirect measurement method on principle of work.Direct measuring method commonly used has ocular estimate, high-speed photography method etc., and this method has certain subjectivity, and no observer may draw different conclusions.Indirect measurement method is the fluctuation signals such as pressure, pressure reduction and void fraction that utilize the instrument and equipment measurement of measuring, and by signal being analyzed, extracted feature, discerns flow pattern in conjunction with neural network isotype recognition technology.This method has two key issues: the one, and the extraction of flow pattern proper vector; The 2nd, the choosing of network model, wherein feature extraction is particularly crucial.Existing method mainly is to utilize single pressure differential pressure fluctuation signal to extract the single feature of signal, and the feature that causes extracting can't reflect the information of flow pattern comprehensively, causes the flow pattern discrimination not high.As invention " based on the Identification of Gas-Liquid Two-Phase method and system of soft-measuring technique " is to discern at the energy feature of pressure difference signal (CN1246683C).

Summary of the invention

The objective of the invention is to overcome the defective of above-mentioned prior art, propose a kind of flow pattern identification accuracy, reliability height, the Method for Discriminating Gas-liquid Two Phase Flow based on information fusion of highly versatile; Another object of the present invention is to provide a kind of rational in infrastructure, can satisfy flow pattern of gas-liquid two-phase flow signal pickup assembly based on information fusion based on the flow pattern of gas-liquid two-phase flow identification convection signals collecting requirement of information fusion.

The objective of the invention is to be realized by following technical scheme: a kind of Method for Discriminating Gas-liquid Two Phase Flow based on information fusion is characterized in that it may further comprise the steps:

(1) data acquisition and pre-service: one by horizontal pipeline, differential pressure transmitter and the flow pattern of gas-liquid two-phase flow signal pickup assembly formed by computing machine based on information fusion in, the pressure spacing of choosing is respectively 5D, 10D and 15D, D is a caliber, fixing respectively and change liquid phase, gas phase flow, obtain stratified flow, intermittent flow, bubble flow and 4 kinds of flow patterns of annular flow under the different gas-liquid flows, flow parameter to 4 kinds of flow patterns is gathered, and the pressure-difference fluctuation signal of three different pressure spacings is carried out the denoising processing.

(2) 3 pressure-difference fluctuation signals after utilizing wavelet theory to denoising carry out feature extraction, and form 3 wavelet packet information entropy proper vectors;

Pressure-difference fluctuation signal S to 3 different pressure spacings ₁, S ₂And S ₃Carry out 4 layers of WAVELET PACKET DECOMPOSITION according to following recursion (1):

$\left\{\begin{array}{c}{u}_{2n}\left(t\right)=\sqrt{2}\underset{k}{\mathrm{\Σ}}h\left(k\right)u_n(2t-k)\\ u_{2n-1}\left(t\right)=\sqrt{2}\underset{k}{\mathrm{\Σ}}g\left(k\right)u_n(2t-k)\end{array}\right.---\left(1\right)$

H in the formula (k) is the Hi-pass filter group; G (k) is the low-pass filter group; u ₀(t)=and  (t), be scaling function; u ₁(t)=and ψ (t), be wavelet function; The yardstick of k for decomposing.

To being reconstructed, respectively obtain 16 wavelet package reconstruction signal S through the sequence in 16 frequency bands that obtain after decomposing _{1j, k}, S _{2j, k}And S _{3j, k}(k=0～15), then original signal S ₁, S ₂And S ₃Can be expressed as

$S_1=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="14"\; label="S"\; filenames="A20061001709100182.png"\; state="\{\{state\}\}">S</figure-callout>}}_{14,k};S_2=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}{S}_{24,k};S_3=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}{S}_{34,k}---\left(2\right)$

The WAVELET PACKET DECOMPOSITION of signal is regarded as a kind of division to signal, define estimating of this division

${\mathrm{\&epsiv;}}_{(j,k)}\left(i\right)=\frac{S_{F(j,k)}\left(i\right)}{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{S}_{F(j,k)}\left(i\right)}---\left(3\right)$

In the formula: S _{F (j, k)}(i) be S _{J, k}(k=0～2 ^jI value of Fourier transform sequence-1), the yardstick of j for decomposing; N is the length of original signal.

According to the basic theories of information entropy, definition wavelet packet information entropy

$H_{j,k}=-\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&epsiv;}}_{(j,k)}\left(i\right)\log {\mathrm{\&epsiv;}}_{(j,k)}\left(i\right),k=0~2^j-1---\left(4\right)$

The ENERGY E of the 4th layer of k wavelet packet signal of signal _{J, k}For

$E_{4,k}={\&Integral;\left|S_{4,k}\left(t\right)\right|}^2\mathrm{dt}=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left|x_{\mathrm{ki}}\right|}^2---\left(5\right)$

The WAVELET PACKET DECOMPOSITION of signal is regarded as a kind of division to signal, define estimating of this division

${\mathrm{\&epsiv;}}_{(i,k)}\left(i\right)=\frac{S_{F(j,k)}\left(i\right)}{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{S}_{F(j,k)}\left(i\right)}---\left(6\right)$

According to the basic theories of information entropy, definition wavelet packet information entropy

$H_{j,k}=-\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&epsiv;}}_{(j,k)}\left(i\right)\log {\mathrm{\&epsiv;}}_{(j,k)}\left(i\right),k=0~2^j-1---\left(7\right)$

Obtain the wavelet packet information entropy of 16 reconstruction signals by formula (7), and then be that element can the structural attitude vector, be expressed as T with these 16 wavelet packet information entropys ₁, then

T ₁＝[H ₁₄₀，H ₁₄₁，Λ，H ₁₄₁₄，H ₁₄₁₅] (8)

T ₂＝[H ₂₄₀，H ₂₄₁，Λ，H ₂₄₁₄，H ₂₄₁₅] (9)

T ₃＝[H ₃₄₀，H ₁₃₄₁，Λ，H ₃₄₁₄，H ₃₄₁₅] (10)

When wavelet-packet energy and information entropy are big, can give to analyze to use and bring inconvenience, can make normalized to proper vector for this reason.Order

$H_{1\mathrm{total}}={\left(\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{\left|{\mathrm{<figure-callout\; id="14"\; label="H"\; filenames="A20061001709100182.png"\; state="\{\{state\}\}">H</figure-callout>}}_{14,j}\right|}^2\right)}^{1/2};H_{2\mathrm{total}}={\left(\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{\left|H_{24,j}\right|}^2\right)}^{1/2};H_{3\mathrm{total}}={\left(\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{\left|H_{34,j}\right|}^2\right)}^{1/2}---\left(11\right)$

Vector T ₁' be normalization wavelet-packet energy and information entropy proper vector.

$T_1^{\&prime;}=[\frac{H_{140}}{H_{1\mathrm{total}}},\frac{H_{141}}{H_{1\mathrm{total}}},\mathrm{\&Lambda;},\frac{H_{1414}}{H_{1\mathrm{total}}},\frac{H_{1415}}{H_{1\mathrm{total}}}]---\left(12\right)$

$T_2^{\&prime;}=[\frac{H_{240}}{H_{2\mathrm{total}}},\frac{H_{241}}{H_{2\mathrm{total}}},\mathrm{\&Lambda;},\frac{H_{2414}}{H_{2\mathrm{total}}},\frac{H_{2415}}{H_{2\mathrm{total}}}]---\left(13\right)$

$T_3^{\&prime;}=[\frac{H_{340}}{H_{3\mathrm{total}}},\frac{H_{341}}{H_{3\mathrm{total}}},\mathrm{\&Lambda;},\frac{H_{3414}}{H_{3\mathrm{total}}},\frac{H_{3415}}{H_{3\mathrm{total}}}]---\left(14\right)$

(3) neural network of flow pattern identification: by above-mentioned 3 wavelet packet information entropy proper vectors that obtain, as the input sample of 3 RBF neural networks, it is (1,0 that the output of 4 kinds of flow patterns is decided to be stratified flow respectively respectively, 0,0), intermittent flow is (0,1,0,0), bubble flow is (0,0,1,0), annular flow is (0,0,0,1), set up 3 neural networks and it is trained, the network that trains can be finished the mapping from feature space to the flow pattern space.

(4) with the output result of each neural network as evidence independent of each other, utilize the D-S evidence theory to carry out information fusion, obtain final recognition result.

1. determine the flow pattern identification framework, according to the stratified flow u in the biphase gas and liquid flow in the horizontal tube ₁, intermittent flow u ₂, bubble flow u ₃With annular flow u ₄Four kinds of flow patterns, the identification framework Θ in the conclusion evidence theory is designated as Θ={ u ₁, u ₂, u ₃, u ₄.Because can only occur a kind of flow pattern simultaneously in flow process, therefore the incident in the flow pattern identification framework is mutual exclusion.

Proposition u is certain flow pattern, and is the subclass under the identification framework.If set function m:2 ^Θ→ [0,1] (2 ^ΘPower set for Θ) satisfies

$m\left(\mathrm{\&phi;}\right)=0;\underset{u}{\mathrm{\&Sigma;}}m\left(u\right)=1(\&ForAll;u\&SubsetEqual;\mathrm{\&Theta;})---\left(15\right)$

Claim that then m is the basic reliability distribution function of framework Θ, φ is an empty set herein.To  u  Θ, m (u) is called the basic confidence level score value of u, when m (u)＞0, claims that u is the burnt unit on the Θ, and m (u) has reflected the affirmation degree to certain flow pattern.Belief function corresponding to the mass function m is defined as

$\left(u\right)=\underset{A\&SubsetEqual;u}{\mathrm{\&Sigma;}}m\left(A\right)(\&ForAll;u\&SubsetEqual;\mathrm{\&Theta;})$ $\left(16\right)$

The certainty of Bel (u) reflection incident has been described the total degree of belief to u, and it comprises the trusting degree of all subclass of u.

2. evidence fusion rule

If Bel ₁And Bel ₂Be two belief functions of same identification framework Θ, m ₁And m ₂Be respectively its corresponding basic reliability distribution, burnt unit is respectively A ₁, A ₂, Λ, A _kAnd B ₁, B ₂, Λ, B _n

If $\underset{A_i{\&cap;B}_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)<1$ So, two belief function Bel ₁And Bel ₂Synthetic basic reliability distribution can be by the function m of following formula definition: 2 ^ΘCalculate → [0,1]:

$m\left(A\right)=\left\{\begin{array}{cc}0& A=\mathrm{\&phi;}\\ \frac{\underset{A_i\&cap;B_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)}{1-\underset{A_i\&cap;B_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)}& A\&NotEqual;\mathrm{\&phi;}\end{array}\right.---\left(17\right)$

In the formula, φ is an empty set, m (φ)=0.If $N=\underset{A_i\&cap;B_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)$ Be to comprise conflict hypothesis A _iAnd B _jAll belief function sum of products, so-called conflict A _iAnd B _j, be meant the target pattern A of hypothesis _iAnd B _jIn Θ, can not exist simultaneously, promptly repel mutually.In the formula (17), A refers to the target pattern A that supposes _iAnd B _jA synthetic proposition of boolean combination, the belief function value m (A) of A comprises the hypothesis A that do not conflict _iAnd B _jAll belief function sum of products.Formula (17) reflection be that two evidences merge, can obtain by formula (17) recursion for the fusion of a plurality of evidences.

3. fusion recognition.The flow pattern of recognition system is 4 kinds of flow patterns, and the identification framework Θ of corresponding evidence theory just comprises 4 flow patterns.Simultaneously, system has 3 recognition networks, and each network correspondence is output as 4, respectively corresponding 4 kinds of flow patterns.Each neural network is as a corroboration of evidence theory, through conversion, becomes the confidence level of various flow patterns under the evidence for this reason to distribute the output valve of neural network, lays the foundation for evidence theory is synthetic.The ability of each recognition network is different, so discount that reliability coefficient is an evidence of each network existence, has represented the trusting degree to recognition result.

If j output valve of i network is O _i(j), the reliability to state j of its correspondence on the basis of this evidence is assigned as so

$m_i\left(j\right)=\frac{O_i\left(j\right){\mathrm{\&alpha;}}_i}{\underset{j=1}{\overset{q}{\mathrm{\&Sigma;}}}{O}_i\left(j\right)}---\left(18\right)$

m(Θ)＝1-α _i，i＝1，2，Λ，p (19)

In the formula: α _iBe the discrimination of each network, m (Θ) represents the uncertain degree of each Network Recognition.

Each flow pattern basic reliability distribution after can obtaining merging according to evidence fusion formula (17) can realize the fusion recognition of convection thus, thereby obtain final recognition result.

A kind of flow pattern of gas-liquid two-phase flow signal pickup assembly based on information fusion, it is characterized in that: it is included in pressure ring 1a, 2b, 3c, 4d, 5e, the 6f that is nested with on the horizontal pipeline 14 of two-phase flow, pressure ring 1a, 2b are connected with pressure pipe 7a, pressure ring 3c, 4d are connected with pressure pipe 7b, pressure ring 5e, 6f are connected with pressure pipe 7c, pressure pipe 7a, 7b, 7c are connected with differential pressure transmitter 8a, 9b, 10c respectively, differential pressure transmitter 8a, 9b, 10c are electrically connected with data collecting card 11, and data collecting card 11 is electrically connected with computing machine 12.

The spacing of described pressure ring 1a, 2b is 5 times of caliber D, and the spacing of pressure ring 3c, 4d is 10 times of caliber D, and the spacing of pressure ring 5e, 6f is 15 times of caliber D.

The beneficial effect that the present invention is based on the Method for Discriminating Gas-liquid Two Phase Flow of information fusion is:

1) the neural network recognition technology is combined with wavelet analysis technology, realized the Intelligent Recognition of flow pattern, overcome the strong shortcoming of classic method subjectivity;

2) this method is carried out information fusion to the output of 3 neural networks, obtains final recognition result, discrimination height, and good reliability.

The present invention is based on flow pattern of gas-liquid two-phase flow signal pickup assembly rational in infrastructure of information fusion, gather the flow pattern signal accurately, true, can satisfy requirement based on the flow pattern of gas-liquid two-phase flow identification convection signals collecting of information fusion.

Description of drawings

Fig. 1 is based on the Method for Discriminating Gas-liquid Two Phase Flow block diagram of information fusion.

Fig. 2 is based on the structural representation of the flow pattern of gas-liquid two-phase flow signal pickup assembly of information fusion.

Fig. 3 is based on the flow chart of the flow pattern of gas-liquid two-phase flow signal pickup assembly of information fusion.

Embodiment

Utilize drawings and Examples that the flow pattern of gas-liquid two-phase flow signal pickup assembly that the Method for Discriminating Gas-liquid Two Phase Flow that the present invention is based on information fusion reaches based on information fusion is described further below.

As shown in Figure 1, the Method for Discriminating Gas-liquid Two Phase Flow that the present invention is based on information fusion comprises the steps:

(1) data acquisition and pre-service: one by horizontal pipeline, differential pressure transmitter and the flow pattern of gas-liquid two-phase flow signal pickup assembly formed by computing machine based on information fusion in, the pressure spacing of choosing is respectively 5D, 10D and 15D, D is a caliber, differential pressure signal to three kinds of different pressure spacings is gathered, and the setting sample frequency is 256Hz; Sampling number is 1024 points; Sampling time is 16s.And the differential pressure fluctuation signal application wavelet theory of gathering is carried out the small echo denoising handle, for feature extraction lays the foundation.

(2) 3 pressure-difference fluctuation signals after utilizing wavelet theory to denoising carry out feature extraction, and form 3 wavelet packet information entropy proper vectors; Pressure-difference fluctuation signal S to 3 different pressure spacings ₁, S ₂And S ₃Carry out 4 layers of WAVELET PACKET DECOMPOSITION according to following recursion (1):

$\left\{\begin{array}{c}{u}_{2n}\left(t\right)=\sqrt{2}\underset{k}{\mathrm{\&Sigma;}}h\left(k\right)u_n(2t-k)\\ u_{2n-1}\left(t\right)=\sqrt{2}\underset{k}{\mathrm{\&Sigma;}}g\left(k\right)u_n(2t-k)\end{array}\right.---\left(1\right)$

H in the formula (k) is the Hi-pass filter group; G (k) is the low-pass filter group; u ₀(t)=and  (t), be scaling function; u ₁(t)=and ψ (t), be wavelet function; The yardstick of k for decomposing.

To being reconstructed, respectively obtain 16 wavelet package reconstruction signal S through the sequence in 16 frequency bands that obtain after decomposing _{1j, k}, S _{2j, k}And S _{3j, k}(k=0～15), then original signal S ₁, S ₂And S ₃Can be expressed as

$S_1=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="14"\; label="S"\; filenames="A20061001709100182.png"\; state="\{\{state\}\}">S</figure-callout>}}_{14,k};S_2=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}{S}_{24,k};S_3=\underset{k=0}{\overset{15}{\mathrm{\&Sigma;}}}{S}_{34,k}---\left(2\right)$

The WAVELET PACKET DECOMPOSITION of signal is regarded as a kind of division to signal, define estimating of this division

${\mathrm{\&epsiv;}}_{(j,k)}\left(i\right)=\frac{S_{F(j,k)}\left(i\right)}{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{S}_{F(j,k)}\left(i\right)}---\left(3\right)$

In the formula: S _{F (j, k)}(i) be S _{J, k}(k=0～2 ^jI value of Fourier transform sequence-1), the yardstick of j for decomposing; N is the length of original signal.

According to the basic theories of information entropy, definition wavelet packet information entropy

$H_{j,k}=-\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&epsiv;}}_{(j,k)}\left(i\right)\log {\mathrm{\&epsiv;}}_{(j,k)}\left(i\right),k=0~2^j-1---\left(4\right)$

The ENERGY E of the 4th layer of k wavelet packet signal of signal _{J, k}For

$E_{4,k}={\&Integral;\left|S_{4,k}\left(t\right)\right|}^2\mathrm{dt}=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left|x_{\mathrm{ki}}\right|}^2---\left(5\right)$

The WAVELET PACKET DECOMPOSITION of signal is regarded as a kind of division to signal, define estimating of this division

${\mathrm{\&epsiv;}}_{(i,k)}\left(i\right)=\frac{S_{F(j,k)}\left(i\right)}{\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{S}_{F(j,k)}\left(i\right)}---\left(6\right)$

According to the basic theories of information entropy, definition wavelet packet information entropy

$H_{j,k}=-\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&epsiv;}}_{(j,k)}\left(i\right)\log {\mathrm{\&epsiv;}}_{(j,k)}\left(i\right),k=0~2^j-1---\left(7\right)$

Obtain the wavelet packet information entropy of 16 reconstruction signals by formula (7), and then be that element can the structural attitude vector, be expressed as T with these 16 wavelet packet information entropys ₁, then

T ₁＝[H ₁₄₀，H ₁₄₁，Λ，H ₁₄₁₄，H ₁₄₁₅] (8)

T ₂＝H ₂₄₀，H ₂₄₁Λ，H ₂₄₁₄，H ₂₄₁₅] (9)

T ₃＝[H ₃₄₀，H ₁₃₄₁，Λ，H ₃₄₁₄，H ₃₄₁₅] (10)

When wavelet-packet energy and information entropy are big, can give to analyze to use and bring inconvenience, can make normalized to proper vector for this reason.Order

$H_{1\mathrm{total}}={\left(\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{\left|{\mathrm{<figure-callout\; id="14"\; label="H"\; filenames="A20061001709100182.png"\; state="\{\{state\}\}">H</figure-callout>}}_{14,j}\right|}^2\right)}^{1/2};H_{2\mathrm{total}}={\left(\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{\left|H_{24,j}\right|}^2\right)}^{1/2};H_{3\mathrm{total}}={\left(\underset{j=0}{\overset{15}{\mathrm{\&Sigma;}}}{\left|H_{34,j}\right|}^2\right)}^{1/2}---\left(11\right)$

Vector T ₁' be normalization wavelet-packet energy and information entropy proper vector.

$T_1^{\&prime;}=[\frac{H_{140}}{H_{1\mathrm{total}}},\frac{H_{141}}{H_{1\mathrm{total}}},\mathrm{\&Lambda;},\frac{H_{1414}}{H_{1\mathrm{total}}},\frac{H_{1415}}{H_{1\mathrm{total}}}]---\left(12\right)$

$T_2^{\&prime;}=[\frac{H_{240}}{H_{2\mathrm{total}}},\frac{H_{241}}{H_{2\mathrm{total}}},\mathrm{\&Lambda;},\frac{H_{2414}}{H_{2\mathrm{total}}},\frac{H_{2415}}{H_{2\mathrm{total}}}]---\left(13\right)$

$T_3^{\&prime;}=[\frac{H_{340}}{H_{3\mathrm{total}}},\frac{H_{341}}{H_{3\mathrm{total}}},\mathrm{\&Lambda;},\frac{H_{3414}}{H_{3\mathrm{total}}},\frac{H_{3415}}{H_{3\mathrm{total}}}]---\left(14\right)$

(3) neural network of flow pattern identification: respectively as the input vector of 3 RBF neural networks, the output of different flow patterns is defined as stratified flow (1,0,0,0), intermittent flow (0 respectively with above-mentioned 3 proper vectors, 1,0,0), bubble flow (0,0,1,0), annular flow (0,0,0,1).With the 200 stack features samples that collect (4 kinds flow pattern each 50 groups), the total error e=0.0001 of system.Choose 160 groups (every kinds flow pattern each 40 groups) more in addition and carry out emulation, check the recognition effect of each network as test sample book.

(4) on the local base of recognition of RBF neural network, add the D-S evidence theory and carry out information fusion identification, obtain final recognition result.

1. determine the flow pattern identification framework, according to the stratified flow u in the biphase gas and liquid flow in the horizontal tube ₁, intermittent flow u ₂, bubble flow u ₃With annular flow u ₄Four kinds of flow patterns, the identification framework Θ in the conclusion evidence theory is designated as Θ={ u ₁, u ₂, u ₃, u ₄.Because can only occur a kind of flow pattern simultaneously in flow process, therefore the incident in the flow pattern identification framework is mutual exclusion.

Proposition u is certain flow pattern, and is the subclass under the identification framework.If set function m:2 ^Θ→ [0,1] (2 ^ΘPower set for Θ) satisfies

$m\left(\mathrm{\&phi;}\right)=0;\underset{u}{\mathrm{\&Sigma;}}m\left(u\right)=1(\&ForAll;u\&SubsetEqual;\mathrm{\&Theta;})---\left(15\right)$

Claim that then m is the basic reliability distribution function of framework Θ, φ is an empty set herein.To  u  Θ, m (u) is called the basic confidence level score value of u, when m (u)＞0, claims that u is the burnt unit on the Θ, and m (u) has reflected the affirmation degree to certain flow pattern.Belief function corresponding to the mass function m is defined as

$\left(u\right)=\underset{A\&SubsetEqual;u}{\mathrm{\&Sigma;}}m\left(A\right)(\&ForAll;u\&SubsetEqual;\mathrm{\&Theta;})$ $\left(16\right)$

The certainty of Bel (u) reflection incident has been described the total degree of belief to u, and it comprises the trusting degree of all subclass of u.

2. evidence fusion rule

If Bel ₁And Bel ₂Be two belief functions of same identification framework Θ, m ₁And m ₂Be respectively its corresponding basic reliability distribution, burnt unit is respectively A ₁, A ₂, Λ, A _kAnd B ₁, B ₂, Λ, B _n

If $\underset{A_i{\&cap;B}_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)<1$ So, two belief function Bel ₁And Bel ₂Synthetic basic reliability distribution can be by the function m of following formula definition: 2 ^ΘCalculate → [0,1]:

$m\left(A\right)=\left\{\begin{array}{cc}0& A=\mathrm{\&phi;}\\ \frac{\underset{A_i\&cap;B_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)}{1-\underset{A_i\&cap;B_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)}& A\&NotEqual;\mathrm{\&phi;}\end{array}\right.---\left(17\right)$

In the formula, φ is an empty set, m (φ)=0.If $N=\underset{A_i\&cap;B_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)$ Be to comprise conflict hypothesis A _iAnd B _jAll belief function sum of products, so-called conflict A _iAnd B _j, be meant the target pattern A of hypothesis _iAnd B _jIn Θ, can not exist simultaneously, promptly repel mutually.In the formula (17), A refers to the target pattern A that supposes _iAnd B _jA synthetic proposition of boolean combination, the belief function value m (A) of A comprises the hypothesis A that do not conflict _iAnd B _jAll belief function sum of products.Formula (17) reflection be that two evidences merge, can obtain by formula (17) recursion for the fusion of a plurality of evidences.

3. fusion recognition.The flow pattern of recognition system is 4 kinds of flow patterns, and the identification framework Θ of corresponding evidence theory just comprises 4 flow patterns.Simultaneously, system has 3 recognition networks, and each network correspondence is output as 4, respectively corresponding 4 kinds of flow patterns.Each neural network is as a corroboration of evidence theory, through conversion, becomes the confidence level of various flow patterns under the evidence for this reason to distribute the output valve of neural network, lays the foundation for evidence theory is synthetic.The ability of each recognition network is different, so discount that reliability coefficient is an evidence of each network existence, has represented the trusting degree to recognition result.

If j output valve of i network is O _i(j), the reliability to state j of its correspondence on the basis of this evidence is assigned as so

$m_i\left(j\right)=\frac{O_i\left(j\right){\mathrm{\&alpha;}}_i}{\underset{j=1}{\overset{q}{\mathrm{\&Sigma;}}}{O}_i\left(j\right)}---\left(18\right)$

m(Θ)＝1-α _i，i＝1，2，Λ，p (19)

In the formula: α _iBe the discrimination of each network, m (Θ) represents the uncertain degree of each Network Recognition.

Each flow pattern basic reliability distribution after can obtaining merging according to evidence fusion formula (17) can realize the fusion recognition of convection thus, thereby obtain final recognition result.

The output result of table 1 localized network

Network	α i	Neural network output result
						O 1	O 2	O 3	O 4
		RBF network 1 RBF network 2 RBF networks 3	0.925 0.938 0.931	0.4277 0.3874 0.5287	0.5854 0.7534 0.6527					0.0337 0.0464 0.0374	0.0207 0.0454 0.0572

Determine the trusting degree that each localized network recognition result is correct according to formula (18), (19), table 1 provides 3 localized network real output value of a sample to be identified of intermittent flow, and the trusting degree that 3 different pressure spacing pressure-difference fluctuation signal characteristics are extracted.

By the identification framework Θ of aforementioned definitions, proposition u is an intermittent flow, and is the subclass under its identification framework.M (u) is the basic reliability distribution of proposition u.The real output value of each local RBF network is carried out normalized, multiply by the total discrimination of neural network again, it is as shown in table 2 to obtain basic reliability distribution.The recognition result combination of 3 RBF networks is comprehensively discerned, and it is as shown in table 3 to obtain final recognition result.

The basic reliability distribution of table 2 localized network

Network	m(u 1)	m(u 2)	m(u 2)	m(u 4)	m(Θ)
						RBF network 1 RBF network 2 RBF networks 3	0.3115 0.2456 0.3954	0.5644 0.6778 0.4885	0.0304 0.0344 0.0376	0.0193 0.0367 0.0465	0.075 0.065 0.070

The recognition result that table 3 evidence merges

Combination of network	m(u 1)	m(u 2)	m(u 2)	m(u 4)	m(Θ)
						Network 1+ network 2+ network 3	0.2234	0.7637	0.0026	0.0024	0.0012

Each Network Recognition of 160 groups of test sample books the results are shown in Table 4.Recognition result from table, the recognition methods that utilization RBF neural network and D-S evidence theory merge owing to made full use of the flow pattern information of 3 different pressure-difference fluctuation signals, has realized the accurate identification of flow pattern, recognition capability obviously strengthens.

Each Network Recognition rate of table 4 test sample book

Combination of network	Stratified flow	Intermittent flow	Bubble flow	Annular flow	Total discrimination
						Sample number	40	40	40	40	160
Discrimination	39	38	38	39	154
						Network 1+ network 2+ network 3	97.5％	92.5％	92.5％	97.5％	96.25％

As shown in Figure 2, a kind of flow pattern of gas-liquid two-phase flow signal pickup assembly based on information fusion, has the horizontal pipeline 14 of two-phase flow, the pressure ring 1a, 2b, 3c, 4d, 5e, the 6f that on the horizontal pipeline 14 of two-phase flow, are nested with, pressure ring 1a, 2b are connected with pressure pipe 7a, pressure ring 3c, 4d are connected with pressure pipe 7b, pressure ring 5e, 6f are connected with pressure pipe 7c, pressure pipe 7a, 7b, 7c are connected with differential pressure transmitter 8a, 9b, 10c respectively, differential pressure transmitter 8a, 9b, 10c are electrically connected with data collecting card 11, and data collecting card 11 is electrically connected with computing machine 12.The spacing of described pressure ring 1a, 2b is 5 times of caliber D, and the spacing of pressure ring 3c, 4d is 10 times of caliber D, and the spacing of pressure ring 5e, 6f is 15 times of caliber D.Wherein pressure ring 1a is apart from inlet 500mm, pressure ring 2b and the pressure ring 3c 300mm of being separated by, pressure ring 4d and the pressure ring 5e 300mm of being separated by.Differential pressure transmitter 8a, 9b, 10c model are that outputting standard signal and the model of PD-23 is that the data collecting card 11 of IMP3595 links to each other, data collecting card 11 links to each other with computing machine 12, finish the data processing of pressure difference signal by computing machine, and then realize collection the flow pattern of gas-liquid two-phase flow signal.Computing machine 12 also can connect printer 13.

As shown in Figure 3, software block diagram based on the flow pattern of gas-liquid two-phase flow signal pickup assembly of information fusion has been described.Software program is the technology that those skilled in the art were familiar with according to Automatic Measurement Technique and microcomputer data processing establishment.Program is at first carried out the initialization of hardware, checks whether the driving of hardware is normal, and whether hardware all connects, if success then carry out following operation, if would get nowhere then need check.If the parameter setting of software is then carried out in hardware device initialization success, the channel number of comprise sample frequency, sampling number (or sampling time), selecting for use etc.After setting these parameters on request, just can carry out data acquisition, the signal after the collection takes out contained noise in the signal earlier through pre-service, and then realizes the collection to the flow pattern of gas-liquid two-phase flow signal.

Claims (7)

1. the Method for Discriminating Gas-liquid Two Phase Flow based on information fusion is characterized in that it comprises the steps:

(1) one by horizontal pipeline, differential pressure transmitter and the flow pattern of gas-liquid two-phase flow signal pickup assembly formed by computing machine based on information fusion in, by gathering the pressure-difference fluctuation signal of 3 different pressure spacings on the horizontal pipeline, fix respectively and change liquid and gas flow, obtain stratified flow, intermittent flow, bubble flow and 4 kinds of flow patterns of annular flow under the different gas-liquid flows, flow parameter to 4 kinds of flow patterns is gathered, and the pressure-difference fluctuation signal of three different pressure spacings is carried out the denoising processing;

(2) 3 pressure-difference fluctuation signals after utilizing wavelet theory to denoising carry out feature extraction, and form 3 wavelet packet information entropy proper vectors;

(3) proper vector of extracting is sent into neural network respectively and train, the neural network that trains can be discerned unknown flow pattern;

(4) with the output result of each neural network as evidence independent of each other, utilize the D-S evidence theory to carry out information fusion, obtain final recognition result.

2. the Method for Discriminating Gas-liquid Two Phase Flow based on information fusion according to claim 1 is characterized in that: described step (1) is to utilize the differential pressure transmitter of 3 different pressure spacings that flow parameter is gathered in real time; The pressure spacing is respectively 5 times, 10 times and 15 times of calibers; Sample frequency is 256Hz; Sampling number is 1024 points; Utilizing wavelet transformation that signal is carried out denoising handles.

3. the Method for Discriminating Gas-liquid Two Phase Flow based on information fusion according to claim 1 is characterized in that: described step (2) is that pressure-difference fluctuation is carried out 4 layers of WAVELET PACKET DECOMPOSITION, extracts the wavelet packet information entropy H of 16 reconstruction signals _4i(i=0,1, Λ, 15) as the proper vector of flow pattern, and carries out normalized.

4. the Method for Discriminating Gas-liquid Two Phase Flow based on information fusion according to claim 1 is characterized in that: the neural network that adopts in the described step (3) is radial basis function (RBF) neural network, and the wavelet packet information entropy of 3 pressure-difference fluctuation signals that step (2) is obtained is as input, export correspondingly with flow pattern, i.e. stratified flow is (1,0,0,0), intermittent flow is (0,1,0,0), bubble flow is (0,0,1,0), annular flow is (0,0,0,1), set up 3 neural networks and it is trained, the network that trains can be finished the mapping from feature space to the flow pattern space.

5. the Method for Discriminating Gas-liquid Two Phase Flow based on information fusion according to claim 1 is characterized in that: the information fusion that adopts in the described step (4) is finished by the D-S evidence theory, according to the stratified flow u in the biphase gas and liquid flow in the horizontal tube ₁, intermittent flow u ₂, bubble flow u ₃With annular flow u ₄Four kinds of flow patterns, the identification framework Θ in the conclusion evidence theory is designated as Θ={ u ₁, u ₂, u ₃, u ₄.

If Bel ₁And Bel ₂Be two belief functions of same identification framework Θ, m ₁And m ₂Be respectively its corresponding basic reliability distribution, burnt unit is respectively A ₁, A ₂, Λ, A _kAnd B ₁, B ₂, Λ, B _n

If $\underset{A_i\&cap;B_i=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)<1$ So, two belief function Bel ₁And Bel ₂Synthetic basic reliability distribution can be by the function m of following formula definition: 2 ^ΘCalculate → [0,1]:

$m\left(A\right)=\left\{\begin{array}{cc}0& A=\mathrm{\&phi;}\\ \frac{\underset{A_i\&cap;B_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)}{1-\underset{A_i\&cap;B_j=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}{m}_1\left(A_i\right)m_2\left(B_j\right)}& \&NotEqual;\mathrm{\&phi;}\end{array}\right.---\left(10\right)$

In the formula, φ is an empty set, m (φ)=0, following formula reflection be that two evidences merge, can obtain by this formula recursion for the fusion of a plurality of evidences:

If the individual output valve of j (j=1,2,3,4) of the individual network of i (i=1,2,3) is O _i(j), the reliability to state j of its correspondence on the basis of this evidence is assigned as so

$m_i\left(j\right)=\frac{O_i\left(j\right){\mathrm{\&alpha;}}_i}{\underset{j=1}{\overset{4}{\mathrm{\&Sigma;}}}{O}_i\left(j\right)}---\left(11\right)$

m(Θ)＝1-α _i，i＝1，2，3 (12)

In the formula: α _iBe the discrimination of each network, m (Θ) represents the uncertain degree of each Network Recognition;

Each flow pattern basic reliability distribution after can obtaining merging according to evidence fusion formula (10) can realize the accurate identification of convection thus.

6. flow pattern of gas-liquid two-phase flow signal pickup assembly based on information fusion, it is characterized in that: it is included in the pressure ring (1a that is nested with on the horizontal pipeline of two-phase flow (14), 2b, 3c, 4d, 5e, 6f), pressure ring (1a, 2b) be connected with pressure pipe (7a), pressure ring (3c, 4d) be connected with pressure pipe (7b), pressure ring (5e, 6f) be connected with pressure pipe (7c), pressure pipe (7a, 7b, 7c) respectively with differential pressure transmitter (8a, 9b, 10c) connect, differential pressure transmitter (8a, 9b, 10c) be electrically connected with data collecting card (11), data collecting card (11) is electrically connected with computing machine (12).

7. the flow pattern of gas-liquid two-phase flow signal pickup assembly based on information fusion according to claim 6, it is characterized in that: the spacing of described pressure ring (1a, 2b) is 5 times of calibers (D), the spacing of pressure ring (3c, 4d) is 10 times of calibers (D), and the spacing of pressure ring (5e, 6f) is 15 times of calibers (D).

Images