Non-straight pipeline leakage positioning detection technology based on instantaneous linear model blind separation

Abstract

The invention relates to the technical field of water supply pipeline leakage positioning, and discloses a non-straight pipeline leakage positioning detection technology based on instantaneous linear model blind separation, and the technology comprises the following steps: S1, a three-way branch pipe (a three-way elbow) and a non-straight leakage pipeline are connected, and the two ends of the non-straight leakage pipeline are respectively provided with an acoustic vibration sensor with a magnetic pedestal; compared with a convolution hybrid time domain algorithm, a convolution hybrid separation frequency domain algorithm has great superiority, especially when the order of a hybrid filter is relatively high, the solution of each order of a separation filter depends on the solution of other orders in the filter, so that the calculation amount is greatly changed, and in the frequency domain algorithm, the calculation amount is greatly reduced. Each fixed frequency is decomposed by using an instantaneous blind separation algorithm, the calculation of each frequency band is mutually independent, the algorithm is slightly influenced by the order of the hybrid filter, and under the same condition, the frequency domain algorithm is much smaller than the time domain algorithm in calculation amount and is easier to converge.

Description

Technical Field

[0001] The invention relates to the technical field of water supply pipeline leakage positioning, and in particular to a non-straight pipeline leakage positioning detection technology of instantaneous linear model blind separation. Background Art

[0002] Water supply pipeline leakage is a waste of water resources, increases water supply costs, and reduces the quality of urban life. Therefore, it is particularly important to develop pipeline leakage detection technology.

[0003] In the process of leak location detection, the non-straight working condition of the pipeline is an important reason for the generation of internal noise. When the fluid flows through the non-straight pipeline, the fluid will be affected by diversion, pressure division, and speed change. The fluid will form multiple vortices in the non-straight pipeline. These vortices influence each other, and they stimulate the pipeline vibration to generate noise. When the pipeline leaks, the leakage signal and these internal noise sources will adjust to each other to a certain extent. Therefore, the leakage signal and the internal noise source have a certain correlation, which causes great difficulties for the detection and positioning of the pipeline.

[0004] The existence of non-straight pipelines increases the difficulty of leak detection and positioning. Traditional detection methods have limited leak detection capabilities for non-straight pipelines. It is necessary to propose a method to solve the problem of leak positioning in non-straight pipelines.

[0005] The current mainstream detection methods mainly include flow detection, pressure detection, light detection and sound detection.

[0006] The flow detection method refers to the use of a flow sensor to obtain flow change data in the pipe, and the leak location information is obtained by analyzing the upstream and downstream flow data. The Chinese patent (CN106352246A) installs an electromagnetic flow meter upstream and downstream of the pipeline, and the pipeline contains a U-shaped bend pipe condition. The leak location is obtained by performing a correlation analysis on the upstream and downstream flow data. This method does not process the noise generated by the U-shaped elbow, and the leak location error is large; the pressure detection method uses a pressure sensor to obtain the negative pressure wave data in the leaking pipeline. Since this negative pressure wave is caused by leakage, the leak location information can be obtained by analyzing the negative pressure wave data.

[0007] The Chinese patent (CN105844051A) collects negative pressure wave data inside the pipeline, uses the inversion method to learn and train the spatiotemporal variation operator fitting mathematical model, compares the parameters of the mathematical model with the negative pressure wave signal sample data, uses the parameterized collaborative method to perform sparse coding, realizes the training and learning of the over-complete dictionary, judges the bit error of the sparse coding, establishes a sparse representation method for the negative pressure wave signal sample data, and identifies the operating state of the pipeline to be tested based on the sparse representation feature extraction method, and locates the leakage point of the pipeline to be tested based on the sparse representation waveform decomposition method. This method is limited to leakage detection of straight pipelines, and does not process the internal noise source signal in non-straight pipelines, so the positioning effect is not ideal; the optical detection method refers to laying optical sensors such as optical fibers during pipeline construction, and locating the leakage point of the pipeline based on the reflected optical fiber signal.

[0008] The Chinese patent (CN106641739A) installs a distributed optical fiber sensor next to the water supply pipeline, and uses a Raman scattering optical fiber sensor to collect distributed temperature data of the water supply pipeline in real time. The location of the pipeline leakage point can be determined based on the temperature distribution diagram of the pipeline. Although this method can perform distributed temperature measurement on the pipeline, the temperature difference of the fluid in the pipe is small and almost constant under actual working conditions, so it cannot be used for leakage detection of non-straight pipes; the acoustic detection method refers to the use of an acoustic vibration sensor to pick up the acoustic signal generated by the pipeline leakage, and the acoustic signal is analyzed to obtain the leakage point location information.

[0009] The Chinese patent (CN105675216A) uses a vibration sensor to collect noisy leakage sound signals near the leakage point, then extracts features from the noisy leakage sound signals, performs approximate entropy calculations on the selected feature data, obtains a measure of signal complexity, and uses the obtained entropy value as the input of the neural network algorithm to identify the leakage event, and then uses a blind system to locate it. Although this method also uses a blind separation algorithm, it fails to effectively process the internal noise generated by non-straight pipe factors, and the positioning error is large.

[0010] The above methods have certain theoretical and technical innovations for straight water supply pipes, but none of them effectively deals with the internal noise sources of non-straight water supply pipes, resulting in large positioning errors. Summary of the invention

[0011] The present invention provides a non-straight pipeline leakage location detection technology for instantaneous linear model blind separation, which has the advantages of good pipeline leakage location detection effect and solves the problems raised by the above background technology.

[0012] The present invention provides the following technical solution: a non-straight pipeline leakage location detection technology based on instantaneous linear model blind separation, comprising the following steps:

[0013] Step S1: a three-way branch pipe (three-way elbow) and a non-straight leakage pipeline, wherein both ends of the non-straight leakage pipeline are equipped with an acoustic vibration sensor with a magnetic base;

[0014] Step S2:

[0015] The analog signal output by the sensor is amplified and filtered by the data collector. The time domain propagation model of the noisy leakage signal collected by the sensor conforms to the convolution model, and the time domain convolution model is transformed into the frequency domain instantaneous product model.

[0016] Step S3:

[0017] Let the separation matrix of X(z) be W(z), omitting z from W and writing it as W;

[0018] Initialize W with random values ​​and set the iteration variable k;

[0019] Start iteration:

[0020]

[0021] Then calculate:

[0022]

[0023] Update iteratively:

[0024] W k =W1-W2(12)

[0025] For the W obtained by single-step iteration k Do normalization:

[0026]

[0027] Until the convergence condition is met, stop;

[0028] After the iteration is completed, the frequency domain separation matrix is ​​transformed back to the time domain: W t ←IFFT(W N );

[0029] Separate the time domain solution of this section of the signal: s t =W t X;

[0030] Step S4:

[0031] Separate the m segments of X observation signal, and the separation result of each segment is based on the formula:

[0032]

[0033] Step S5:

[0034] The signal complexity value of the source signal obtained from the two observation signals is calculated. For the sample entropy calculation, the spectrum occupied by the leakage signal is wider than the internal noise and has greater randomness, so its sample entropy value is larger. This feature can be used to distinguish the leakage source signal;

[0035] The cross-correlation estimation of the leakage source signal is performed to obtain the time delay value τ, and then combined with the empirical sound velocity v of the pipeline material, the leakage point positioning formula is: Positioning can be completed, L is the length of the propagation path from the leak point to a certain sensor, and L1 is a known path.

[0036] Preferably, step S2 specifically includes:

[0037] Establish the observation signal matrix model, s=[s1,s2…s n ], and establish the following convolution model:

[0038]

[0039] Where x=[x1,x2…x n ],y=[y1,y2…y n ], the convolution mixed signal is windowed and divided into m windows, and then the time domain signal of each window is fast Fourier transformed, that is, the leakage convolution model is transformed into the frequency domain product model, and the model can be obtained:

[0040]

[0041] Preferably, step S3 specifically includes:

[0042] For the above iterative formula, the nonlinear function of the cost function is taken as G(y)=lg(a+y), where a is 0.1, we can get: g(y)=G'(y)=1 / (a+y) and g'=-1 / (a+y) 2 .

[0043] Preferably, step S4 specifically includes:

[0044] Calculate the correlation coefficient c between the output of the previous segment and the output of this segment ij , adjust the order of the output vectors to keep the signals of each output channel consistent, connect the obtained time domain output signals in sequence, and then separate the observed signal Y in the same way.

[0045] Preferably, step S3 specifically includes:

[0046] The complex FastICA separation algorithm is used as the core algorithm in this detection method:

[0047] Let source signal S = S r +jS iis the whitened signal, and the noisy mixed signal X = X r +jX i , separation matrix W = W r +jW i are all complex signals. Since both the source signal and the noisy mixed signal are whitened data, we have E(xx H )=I, I is the unit matrix. According to the Kuhn-Tucker condition, under the constraint condition E((W H X) 2 )=1, the cost function of the separation method can be derived as follows:

[0048]

[0049] The Jacobian iteration matrix of the first term in the above formula can be approximately expressed as:

[0050]

[0051] where g(·) = G'(·), and the matrix The Jacobian iteration matrix can be expressed as:

[0052]

[0053] Combining the above equations (2) and (3), we can get the Jacobian iteration matrix of the cost function (1):

[0054] J=2(E(g(|W H X| 2 )+|W H | 2 g'(||W H X| 2 ))-β)I (4)

[0055] The iteration rule of the separation matrix W is derived from the Newton iteration method:

[0056] W + =E(X(W H X)*g(|W H X| 2 ))-E(g(|W H X| 2 )+|W H X| 2 g'(||W H X| 2 ))W (5)

[0058] In order to correct the output signal amplitude distortion problem, the W obtained in each iteration is normalized:

[0059]

[0060] The above is the separation and iteration process of the core algorithm of this method, FastICA.

[0061] The signal complexity value is calculated for the source signal obtained from the two observation signals.

[0062] The cross-correlation estimation of the leakage source signal is performed to obtain the time delay value τ, and then combined with the empirical sound velocity v of the pipeline material, the leakage point positioning formula is: Positioning can be completed, L is the length of the propagation path from the leak point to a certain sensor, and L1 is a known path.

[0063] The present invention has the following beneficial effects:

[0064] 1. The non-straight pipe leakage location detection technology based on instantaneous linear model blind separation has great advantages over the convolution mixed time domain algorithm in the frequency domain algorithm of convolution mixed separation, especially when the order of the mixed filter is high. The solution of each order of the separation filter depends on the solution of the remaining orders in the filter, which makes the amount of calculation very large. In the frequency domain algorithm, each fixed frequency is decomposed by the instantaneous blind separation algorithm, and the calculation of each frequency band is independent of each other. The algorithm is less affected by the order of the mixed filter. Under the same conditions, the frequency domain algorithm is much smaller than the time domain algorithm in terms of calculation amount and is easier to converge.

[0065] 2. The non-straight pipeline leakage location detection technology based on instantaneous linear model blind separation normalizes the separation matrix obtained in each iteration through this method, avoiding the amplitude uncertainty of the output vector, and then uses the optimization algorithm to solve the problem of output vector order uncertainty. Since the traditional blind separation algorithm is sensitive to noise, especially to white noise and other wide spectrum noise, the separation effect is not ideal. The blind separation algorithm used in this method shows strong anti-interference ability to noise and can achieve excellent separation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a schematic diagram of the arrangement structure of the perception sensor 1 of the present invention;

[0067] Figure 2 This is a schematic diagram of the arrangement structure of the perception sensor 2 of the present invention;

[0068] Figure 3 This is a schematic diagram of the power spectrum structure of a noisy leakage signal according to the present invention;

[0069] Figure 4 It is a schematic diagram of the structure of the leakage detection process of the present invention;

[0070] Figure 5It is a schematic diagram of the structure of the calculation results of the randomness complexity of the leakage signal and the noise source signal of the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] See also Figure 1 - Figure 5 , a non-straight pipeline leakage location detection technology based on instantaneous linear model blind separation, comprising the following steps:

[0073] Step S1: a three-way branch pipe (three-way elbow) and a non-straight leakage pipeline, wherein both ends of the non-straight leakage pipeline are equipped with an acoustic vibration sensor with a magnetic base;

[0074] Step S2:

[0075] The analog signal output by the sensor is amplified and filtered by the data collector. The time domain propagation model of the noisy leakage signal collected by the sensor conforms to the convolution model, and the time domain convolution model is transformed into the frequency domain instantaneous product model.

[0076] Step S3:

[0077] Let the separation matrix of X(z) be W(z), omitting z from W and writing it as W;

[0078] Initialize W with random values ​​and set the iteration variable k;

[0079] Start iteration:

[0080]

[0081] Then calculate:

[0082]

[0083] Update iteratively:

[0084] W k =W1-W2(12)

[0085] For the W obtained by single-step iteration k Do normalization:

[0086]

[0087] Until the convergence condition is met, stop;

[0088] After the iteration is completed, the frequency domain separation matrix is ​​transformed back to the time domain: W t ←IFFT(W N );

[0089] Separate the time domain solution of this section of the signal: s t =W t X;

[0090] Step S4:

[0091] Separate the m segments of X observation signal, and the separation result of each segment is based on the formula:

[0092]

[0093] Step S5:

[0094] The signal complexity value of the source signal obtained from the two observation signals is calculated. For the sample entropy calculation, the spectrum occupied by the leakage signal is wider than the internal noise and has greater randomness, so its sample entropy value is larger. This feature can be used to distinguish the leakage source signal;

[0095] The cross-correlation estimation of the leakage source signal is performed to obtain the time delay value τ, and then combined with the empirical sound velocity v of the pipeline material, the leakage point positioning formula is: Positioning can be completed, L is the length of the propagation path from the leak point to a certain sensor, and L1 is a known path.

[0096] In a preferred embodiment, step S2 specifically includes:

[0097] Establish the observation signal matrix model, s=[s1,s2…s n ], and establish the following convolution model:

[0098]

[0099] Where x=[x1,x2…x n ],y=[y1,y2…y n ], the convolution mixed signal is windowed and divided into m windows, and then the time domain signal of each window is fast Fourier transformed, that is, the leakage convolution model is transformed into the frequency domain product model, and the model can be obtained:

[0100]

[0101] In a preferred embodiment, step S3 specifically includes:

[0102] For the above iterative formula, the nonlinear function of the cost function is taken as G(y)=lg(a+y), where a is 0.1, we can get: g(y)=G'(y)=1(a+y) and g'=-1 / (a+y) 2.

[0103] In a preferred embodiment, step S4 specifically includes:

[0104] Calculate the correlation coefficient c between the output of the previous segment and the output of this segment ij , adjust the order of the output vectors to keep the signals of each output channel consistent, connect the obtained time domain output signals in sequence, and then separate the observed signal Y in the same way.

[0105] In a preferred embodiment, step S3 specifically includes:

[0106] The complex FastICA separation algorithm is used as the core algorithm in this detection method:

[0107] Let source signal S = S r +jS i is the whitened signal, and the noisy mixed signal X = X r +jX i , separation matrix W = W r +jW i are all complex signals. Since both the source signal and the noisy mixed signal are whitened data, we have E(xx H )=I, I is the unit matrix. According to the Kuhn-Tucker condition, under the constraint condition E((W H X) 2 )=1, the cost function of the separation method can be derived as follows:

[0108]

[0109] The Jacobian iteration matrix of the first term in the above formula can be approximately expressed as:

[0110]

[0111] where g(·) = G'(·), and the matrix The Jacobian iteration matrix can be expressed as:

[0112]

[0113] Combining the above equations (2) and (3), we can get the Jacobian iteration matrix of the cost function (1):

[0114] J=2(E(g(|W H X| 2 )+|W H | 2 g'(||W H X| 2 ))-β)I (4)

[0115] The iteration rule of the separation matrix W is derived from the Newton iteration method:

[0116] W + =E(X(W H X)*g(|W H X| 2 ))-E(g(|W H X| 2 )+|W H X| 2 g'(||W H X| 2 ))W (5)

[0118] In order to correct the output signal amplitude distortion problem, the W obtained in each iteration is normalized:

[0119]

[0120] The above is the separation and iteration process of the core algorithm of this method, FastICA.

[0121] The signal complexity value is calculated for the source signal obtained from the two observation signals.

[0122] The cross-correlation estimation of the leakage source signal is performed to obtain the time delay value τ, and then combined with the empirical sound velocity v of the pipeline material, the leakage point positioning formula is: Positioning can be completed, L is the length of the propagation path from the leak point to a certain sensor, and L1 is a known path.

[0123] Working principle: Under field conditions, a non-straight leaking pipeline with a three-way branch or elbow is used as the research object. Acoustic vibration sensors with magnetic bases are installed at both ends of the pipeline. The analog signal output by the sensor is amplified and filtered by the data collector to establish the observation signal matrix model, s = [s1, s2…s n ], and establish the following convolution model:

[0124]

[0125] Where x=[x1,x2…x n ],y=[y1,y2…y n ], the convolution mixed signal is windowed and divided into m windows, and then the time domain signal of each window is fast Fourier transformed, that is, the leakage convolution model is transformed into the frequency domain product model, and the model can be obtained:

[0126]

[0127] Let the separation matrix of X(z) be W(z), omitting z from W and writing it as W;

[0128] Initialize W with random values ​​and set the iteration variable k;

[0129] Start iteration:

[0130]

[0131] Then calculate:

[0132]

[0133] For the above iterative formula, the nonlinear function of the cost function is taken as G(y)=lg(a+y), where a is 0.1, we can get: g(y)=G'(y)=1 / (a+y) and g'=-1 / (a+y) 2

[0134] Update iteratively:

[0135] W k =W1-W2 (12)

[0136] For the W obtained by single-step iteration k Do normalization:

[0137]

[0138] Until the convergence conditions are met, stop!

[0139] After the iteration is completed, the frequency domain separation matrix is ​​transformed back to the time domain: W t ←IFFT(W N );

[0140] Separate the time domain solution of this section of the signal: s t =W t X;

[0141] The above steps are used to separate the m segments of X observation signals, and the separation results of each segment are based on the formula:

[0142]

[0143] Calculate the correlation coefficient c between the output of the previous segment and the output of this segment ij , adjust the order of the output vectors to keep the signals of each output channel consistent, connect the obtained time domain output signals in sequence, and then separate the observed signal Y in the same way.

[0144] The signal complexity value is calculated for the source signal obtained by the two observation signals. Here, the calculation of sample entropy is taken as an example. The calculation result is as follows: Figure 5 The spectrum occupied by the leakage signal is wider than the internal noise and has greater randomness, so its sample entropy value is larger, and this feature is used to distinguish the leakage source signal.

[0145] The cross-correlation estimation of the leakage source signal is performed to obtain the time delay value τ, and then combined with the empirical sound velocity v of the pipeline material, the leakage point positioning formula is: Positioning can be completed, L is the length of the propagation path from the leak point to a certain sensor, and L1 is a known path.

[0146] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0147] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A non-straight pipeline leakage location detection technology based on instantaneous linear model blind separation, characterized in that: The following steps are involved: Step S1: a three-way branch pipe (three-way elbow) and a non-straight leakage pipeline, wherein both ends of the non-straight leakage pipeline are equipped with an acoustic vibration sensor with a magnetic base; Step S2: The analog signal output by the sensor is amplified and filtered by the data collector. The time domain propagation model of the noisy leakage signal collected by the sensor conforms to the convolution model, and the time domain convolution model is transformed into the frequency domain instantaneous product model. Step S3: Let the separation matrix of X(z) be W(z), omitting z from W and writing it as W; Initialize W with random values ​​and set the iteration variable k; Start iteration: Then calculate: Update iteratively: IN k =W1-W2(12) For the W obtained by single-step iteration k Do normalization: Until the convergence condition is met, stop; After the iteration is completed, the frequency domain separation matrix is ​​transformed back to the time domain: W t ←IFFT(W N ); Separate the time domain solution of this section of the signal: s t =W t X; Step S4: Separate the m segments of X observation signal, and the separation result of each segment is based on the formula: Step S5: The signal complexity value of the source signal obtained from the two observation signals is calculated. For the sample entropy calculation, the spectrum occupied by the leakage signal is wider than the internal noise and has greater randomness, so its sample entropy value is larger. This feature can be used to distinguish the leakage source signal; The cross-correlation estimation of the leakage source signal is performed to obtain the time delay value τ, and then combined with the empirical sound velocity v of the pipeline material, the leakage point positioning formula is: Positioning can be completed, L is the length of the propagation path from the leak point to a certain sensor, and L1 is a known path.

2. According to claim 1, a non-straight pipeline leakage location detection technology based on instantaneous linear model blind separation is characterized by: Step S2 specifically includes: Establish the observation signal matrix model, s=[s1,s2…s n ], and establish the following convolution model: Where x=[x1,x2…x n ],y=[y1,y2…y n ], the convolution mixed signal is windowed and divided into m windows, and then the time domain signal of each window is fast Fourier transformed, that is, the leakage convolution model is transformed into the frequency domain product model, and the model can be obtained:

3. According to claim 1, the non-straight pipeline leakage location detection technology based on instantaneous linear model blind separation is characterized by: Step S3 specifically includes: For the above iterative formula, the nonlinear function of the cost function is taken as G(y)=lg(a+y), where a is 0.1, we can get: g(y)=G'(y)=1 / (a+y) and g'=-1 / (a+y) 2 .

4. The non-straight pipeline leakage location detection technology based on instantaneous linear model blind separation according to claim 1 is characterized by: Step S4 specifically includes: Calculate the correlation coefficient c between the output of the previous segment and the output of this segment ij , adjust the order of the output vectors to keep the signals of each output channel consistent, connect the obtained time domain output signals in sequence, and then separate the observed signal Y in the same way.

5. The non-straight pipeline leakage location detection technology based on instantaneous linear model blind separation according to claim 1 is characterized by: Step S3 specifically includes: The complex FastICA separation algorithm is used as the core algorithm in this detection method: Let source signal S = S r +jS i is the whitened signal, and the noisy mixed signal X = X r +jX i , separation matrix W = W r +jW i are all complex signals. Since both the source signal and the noisy mixed signal are whitened data, we have E(xx H )=I, I is the unit matrix. According to the Kuhn-Tucker condition, under the constraint condition E((W H X) 2 )=1, the cost function of the separation method can be derived as follows: The Jacobian iteration matrix of the first term in the above formula can be approximately expressed as: where g(·) = G'(·), and the matrix The Jacobian iteration matrix can be expressed as: Combining the above equations (2) and (3), we can get the Jacobian iteration matrix of the cost function (1): J=2(E(g(|W H X| 2 )+|W H | 2 g'(||W H X| 2 ))-β)I (4) The iteration rule of the separation matrix W is derived from the Newton iteration method: W + =E(X(W H X)*g(||W H X| 2 ))-E(g(||W H X| 2 )+|W H X| 2 g'(||W H X| 2 ))W (5) In order to correct the output signal amplitude distortion problem, the W obtained in each iteration is normalized: The above is the separation and iteration process of the core algorithm of this method, FastICA. The signal complexity value is calculated for the source signal obtained from the two observation signals. The cross-correlation estimation of the leakage source signal is performed to obtain the time delay value τ, and then combined with the empirical sound velocity v of the pipeline material, the leakage point positioning formula is: Positioning can be completed, L is the length of the propagation path from the leak point to a certain sensor, and L1 is a known path.

Images