CN105864643A - Gas pipeline leakage positioning experimental device and method based on RBF neural network - Google Patents

Abstract

The invention belongs to the technical field of gas pipeline leakage positioning and particularly relates to a gas pipeline leakage positioning experimental device and method based on an RBF neural network. The gas pipeline leakage positioning experimental method comprises the steps that a pressure sensor is installed at the tail end of a gas transmission pipeline, and five equal-precision acoustic emission sensors are installed on one side of a preset leakage point at equal intervals; leakage signals collected by five channels under different leakage pressure are subjected to feature extraction and processing, and the maximum peak value, the (3, 0) node wavelet packet energy and the power spectrum peak value corresponding to the leakage feature frequency of the leakage signal of each channel are obtained to construct feature vectors under the different leakage pressure; and after being subjected to normalization processing, the feature vectors serve as the input of the RBF combined neural network, the leakage positioning value L serves as the output of the RBF combined neural network, and an RBF combined neural network pipeline leakage positioning model is built under the different pressure, so that gas pipeline leakage point positioning and error analysis are achieved.

Description

Gas pipe leakage positioning experiment device based on RBF neural and method

Technical field

The invention belongs to gas pipe leakage field of locating technology, particularly relate to a kind of based on RBF neural Gas pipe leakage positioning experiment device and method.

Background technology

Along with energy crisis and environmental problem are day by day serious, carbon captures and seals (Carbon Capture and up for safekeeping Storage, CCS) technology be considered as process global warming, reduce the maximally effective one of greenhouse gases Method.CCS technology is the carbon dioxide separation that industry, energy industry are produced to be gone out, then passes through Carbon storage means, are transported to the inferior place with atmospheric isolation in seabed or ground.Use pipeline transportation is caught The CO2 obtained is the mode most economical, most reliable generally acknowledged at present, but often occurs in course of conveying letting out Leakage problem.Therefore, a kind of suitably leakage locating method is found, it is possible in time leakage point is positioned, Improve the safety of carbon dioxide conveyance conduit.

Acoustic emission is to solve carbon dioxide conveying pipe with the combination of RBF (RBF) neutral net The important research direction of road leakage location difficulty.Rely on acoustic emission, by the feature extraction of leakage signal, The leakage location of carbon dioxide conveyance conduit is carried out, it is possible to solve carbon dioxide and let out in conjunction with RBF neural The problem that leakage location is difficult, method based on RBF neural has higher accuracy.

Summary of the invention

The problem difficult in order to solve gas transmission pipeline leakage location, the present invention proposes a kind of based on RBF god Gas pipe leakage positioning experiment device and method through network.

Experimental provision includes: the compressor that is sequentially connected, filter, the first ball valve, high-pressure buffering pot, gas Liquid/gas separator, the second ball valve, the first matrix pipeline, simulated leakage unit, the second matrix pipeline, the 3rd ball Valve, recycling can；By Flange joint between described matrix pipeline and simulated leakage unit, simulated leakage unit Being sealed in protective cover, simulated leakage unit is a straight pipeline, and its one end is provided with leakage point, leakage point Opposite side is equidistantly provided with five equally accurate acoustic emission sensors, and simulated leakage cell end is provided with temperature and passes Sensor, pressure transducer, simulated leakage unit includes zone of heating and heat-insulation layer；Various sensings in protective cover Device is connected with analyzer by the capture card outside protective cover；Protective cover top connects mozzle.

Localization method includes:

Step 1: measure setup is at the gas temperature of the simulated leakage cell end of experimental provision and gas pressure；

Step 2: gather five equally accurate acoustic emission sensor signals, and be transferred to adopt by five signalling channels Truck；

Gas pressure in step 3, regulation simulated leakage unit, is extracted in different lower five signals of leak pressure The feature of the gas leakage signal that passage is collected；

Step 4, structural feature vector, and it is normalized；

Step 5, using the characteristic vector under leak pressures different after normalization as RBF combination neural net Input, to leak the locator value L output variable as RBF combination neural net, sets up RBF combination nerve Network model, chooses training sample and carries out network training；

Step 6: realize letting out by the RBF neural that the test sample input under different leak pressures trains Leak source positions, and calculation of position errors.

Described step 3 includes:

The leakage signal that five signalling channels collect is divided into multiple sections, finds the peak-peak of every section, The peak-peak of each section is averaging, is used as the leakage that under different leak pressure, i-th signalling channel collects The peak-peak of signalAnd as first input element of RBF combination neural net；

Primary signal carries out Fourier transform obtain the i-th signalling channel power spectrum under different leak pressure and letting out Spectrum peak at leakage frequencyAnd as second input element of RBF combination neural net；

By primary signal through using db1 wavelet basis to carry out three layers of WAVELET PACKET DECOMPOSITION, obtain different leak pressure The wavelet-packet energy of lower i-th signalling channel (3,0) nodeAnd as the 3rd of RBF combination neural net Individual input element；Wherein, k is the numbering of corresponding different leak pressure.

Described step 3 obtains the gas under different condition except the gas pressure in regulation simulated leakage unit Outside the feature of body leakage signal, the mode also by change temperature, leakage aperture obtains different condition Under the feature of leakage signal.

The invention has the beneficial effects as follows:

(1) according to the feature of carbon dioxide conveyance conduit, based on RBF combination neural net, in conjunction with acoustic emission Technology, on the basis of acoustic emission, it is proposed that a kind of method that novel feature vector extracts, for dioxy Change the development of carbon conveyance conduit Leak Locating Technology and provide new direction.

(2) wavelet packet is utilized to carry out characteristic vector pickup, using wavelet-packet energy as characteristic vector unit Element, carries out leakage location by RBF combination neural net, has possessed and has realized pipe leakage standard under different pressures Determine the ability of position.

Accompanying drawing explanation

Fig. 1 is experimental provision schematic diagram；

Fig. 2 is the method flow diagram of the present invention；

Fig. 3 is leakage signal Characteristic Extraction flow chart；

Fig. 4 is RBF combination neural net structure chart；

Detailed description of the invention

Below in conjunction with the accompanying drawings, embodiment is elaborated.

As it is shown in figure 1, the gas pipe leakage positioning experiment device bag based on RBF neural of the present invention Include: the compressor that is sequentially connected, filter, the first ball valve, high-pressure buffering pot, gas-liquid separator, second Ball valve, the first matrix pipeline, simulated leakage unit, the second matrix pipeline, the 3rd ball valve, recycling can；Institute Stating by Flange joint between matrix pipeline and simulated leakage unit, simulated leakage unit is sealed in protective cover, Simulated leakage unit is a straight pipeline, and its one end is provided with leakage point, and the opposite side of leakage point is equidistantly installed Having five equally accurate acoustic emission sensors, simulated leakage cell end is provided with temperature sensor, pressure transducer, Simulated leakage unit includes zone of heating and heat-insulation layer；Various sensors in protective cover are by adopting outside protective cover Truck is connected with analyzer；Protective cover top connects mozzle.

Change temperature, pressure, the situation in leakage aperture respectively, can obtain multiple in the case of carbon dioxide Leakage situation, is gathered by acoustic emission signal and pretreatment unit accesses analyzer, and according to RBF combination god Source leak point positioning is carried out through network.

As in figure 2 it is shown, the gas pipe leakage localization method based on RBF neural of the present invention includes: Feature is leaked according to carbon dioxide conveyance conduit, in conjunction with acoustic emission and leakage signal Feature Extraction Technology, On the basis of laboratory simulation carbon dioxide conveyance conduit leak test, carry out RBF combination neural net Modeling, carry out carbon dioxide conveyance conduit leakage positioning analysis.

Specifically include following steps:

Step 1: measure setup is at the gas temperature of the simulated leakage cell end of experimental provision and gas pressure；

Step 2: gather five equally accurate acoustic emission sensor signals, and be transferred to adopt by five signalling channels Truck；

Gas pressure in step 3, regulation simulated leakage unit, is extracted in different lower five signals of leak pressure The feature of the gas leakage signal that passage is collected；

Step 4, structural feature vector, and it is normalized；

Step 5, using the characteristic vector under leak pressures different after normalization as RBF combination neural net Input, to leak the locator value L output variable as RBF combination neural net, sets up RBF combination nerve Network model, chooses training sample and carries out network training；

Step 6: realize letting out by the RBF neural that the test sample input under different leak pressures trains Leak source positions, and calculation of position errors.

Described step 3 includes:

The leakage signal that five signalling channels collect is divided into multiple sections, finds the peak-peak of every section, The peak-peak of each section is averaging, is used as the leakage that under different leak pressure, i-th signalling channel collects The peak-peak of signalAnd as first input element of RBF combination neural net；

Primary signal carries out Fourier transform obtain the i-th signalling channel power spectrum under different leak pressure and letting out Spectrum peak at leakage frequencyAnd as second input element of RBF combination neural net；

By primary signal through using db1 wavelet basis to carry out three layers of WAVELET PACKET DECOMPOSITION, obtain different leak pressure The wavelet-packet energy of lower i-th signalling channel (3,0) nodeAnd as the 3rd of RBF combination neural net Individual input element；Wherein, k is the numbering of corresponding different leak pressure.

Described step 3 obtains the gas under different condition except the gas pressure in regulation simulated leakage unit Outside the feature of body leakage signal, the mode also by change temperature, leakage aperture obtains different condition Under the feature of leakage signal.

In described step 3, eigenvector algorithm formula is as follows:

(1) average maxima peak algorithmic formula:

$P_i^k\[m\]=\max a\[j\]\left\{\begin{array}{c}m=1,j=1,2...\frac{N}{3}\\ m=2,j=\frac{N}{3}+1,\frac{N}{3}+2...\frac{2N}{3}\\ m=3,j=\frac{2N}{3}+1,\frac{2N}{3}+2...N\end{array}\right.$

X1^k _i=mean P_i ^k[m]

Wherein, m is the sequence number in one section of certain interval of leakage signal；A [j] is the signal amplitude of jth sampled point； N is total data acquisition sampling point number；P_i ^k[m] is the i-th passage m interval leakage signal maximum peak under different pressures Value；For the i-th channel signal peak value under different leak pressures.

(2) wavelet packet decomposition algorithm formula:

$\left\{\begin{array}{c}{d}_{i,j,2n}=\frac{1}{\sqrt{2}}\underset{l}{\Σ}{h}_{0(l-2j)}{d}_{i+1,l,n}\\ d_{i,j,2n+1}=\frac{1}{\sqrt{2}}\underset{l}{\Σ}{h}_{1(l-2j)}{d}_{i+1,l,n}\end{array}\right.$

Wherein, d_i+1,l,nFor upper strata WAVELET PACKET DECOMPOSITION result；d_i,j,2nWith d_i,j,2n+1For next stage decomposition result；I is Scale index；J is positioning index；N is Frequency Index；L is variable；h₀And h₁For decomposing the many resolutions used Rate filter coefficient.(3,0) node wavelet-packet energy under different leak pressures:

$X{2^k}_i=\left|\right|d_{i,j}|{|}^2=\underset{-\∝}{\overset{+\∝}{\∫}}{\left|d_{i,j}\right|}^{2}dt$

Wherein d_i,jIt it is the signal amplitude of each sampled point in (3,0) node after three layers of wavelet packet joint.

(3) spectrum peak algorithmic formula:

Sampled data is x (n), and its Fourier conversion and power spectral density are estimatedThere is following relation:

$\left\{\begin{array}{c}{\hat{s}}_x\left(f\right)=\frac{1}{N}{\left|X{\left(k\right)}^{\′}\right|}^2\\ {\hat{S}}_x\left(k^{\′}\right)=\frac{1}{N}{\left|X{\left(k\right)}^{\′}\right|}^2=\frac{1}{N}{|FFT\[x\left(n\right)\]|}^2,k^{\′}=0,1...N-1\\ X{3^k}_i={\hat{S}}_x\left(\frac{f^{\′}}{\mathrm{\Δ}f}\right)\end{array}\right.$

Wherein N is the length of x (n), f=k' Δ f, and FFT [x (n)] is the Fourier conversion of sampled data x (n), f' Characteristic frequency for leakage signal.

As it is shown on figure 3, the embodiment of the present invention based on leakage signal Characteristic Extraction flow chart, this flow chart Mainly include herein below:

Characteristic Extraction is divided into three parts, Part I, will leak out signal and is divided into three sections, asks for each district Between peak-peak, afterwards the peak-peak that three are interval is averaging；Part II, with FFT to leakage Signal carries out power spectrumanalysis, asks for the spectrum peak that leakage signal characteristic frequency is corresponding；Part III, Use db1 wavelet basis that leakage signal carries out 3 layers of WAVELET PACKET DECOMPOSITION, obtain the decomposition result of (3,0) node, Ask for the energy of (3,0) node, the i.e. wavelet-packet energy of (3,0) node.

As shown in Figure 4, the embodiment of the present invention based on RBF combination neural net structure chart, this structure chart master Herein below to be included:

This RBF combination neural net is by three sub-networks, and each sub-network has three layers, respectively input layer, Obscuring layer, output layer；Wherein input layer is three inputs, and output layer is an output.

Corresponding input isIt is output as the position L of source leakage point.

In described step 5, RBF neural structure is described as follows:

$f=\underset{j=1}{\overset{n}{\Σ}}{c}_j{R}_j\left(X\right)$

$R_j\left(X\right)=\exp (-\frac{\left|\right|X-m_j|{|}^2}{{\mathrm{\δ}}_j^2})$

Wherein, c_jFor the output weight of jth hidden node, R_j(X) it is the Gaussian function of jth hidden node, m_j∈R^NFor jth center, δ_jFor jth standard deviation.

In described step 6, the relative average error formula of calculating leak position L:

$RMS=(\underset{i}{\overset{N}{\Σ}}\frac{|L\left(i\right)-L_0\left(i\right)|}{L_0\left(i\right)}\×100\%)/N$

Wherein, L (i) is predictive value, L₀I () is measured value, N is the total number of samples of test.

In described step 4, structural feature vectorCharacteristic vector is normalized place Reason,Wherein, ENERGY E^kFor

This embodiment is only the present invention preferably detailed description of the invention, but protection scope of the present invention is not limited to In this, any those familiar with the art, in the technical scope that the invention discloses, can think easily The change arrived or replacement, all should contain within protection scope of the present invention.Therefore, the protection model of the present invention Enclose and be as the criterion with scope of the claims.

Claims (4)

1. a gas pipe leakage positioning experiment device based on RBF neural, it is characterised in that bag Include: the compressor that is sequentially connected, filter, the first ball valve, high-pressure buffering pot, gas-liquid separator, second Ball valve, the first matrix pipeline, simulated leakage unit, the second matrix pipeline, the 3rd ball valve, recycling can；Institute Stating by Flange joint between matrix pipeline and simulated leakage unit, simulated leakage unit is sealed in protective cover, Simulated leakage unit is a straight pipeline, and its one end is provided with leakage point, and the opposite side of leakage point is equidistantly installed Having five equally accurate acoustic emission sensors, simulated leakage cell end is provided with temperature sensor, pressure transducer, Simulated leakage unit includes zone of heating and heat-insulation layer；Various sensors in protective cover are by adopting outside protective cover Truck is connected with analyzer；Protective cover top connects mozzle.

2. a gas pipe leakage localization method based on experimental provision described in claim 1, its feature exists In, including:

Step 1: measure setup is at the gas temperature of the simulated leakage cell end of experimental provision and gas pressure；

Step 2: gather five equally accurate acoustic emission sensor signals, and be transferred to adopt by five signalling channels Truck；

Gas pressure in step 3, regulation simulated leakage unit, is extracted in different lower five signals of leak pressure The feature of the gas leakage signal that passage is collected；

Step 4, structural feature vector, and it is normalized；

Step 5, using the characteristic vector under leak pressures different after normalization as RBF combination neural net Input, to leak the locator value L output variable as RBF combination neural net, sets up RBF combination nerve Network model, chooses training sample and carries out network training；

Step 6: realize letting out by the RBF neural that the test sample input under different leak pressures trains Leak source positions, and calculation of position errors.

Method the most according to claim 2, it is characterised in that described step 3 includes:

The leakage signal that five signalling channels collect is divided into multiple sections, finds the peak-peak of every section, The peak-peak of each section is averaging, is used as the leakage that under different leak pressure, i-th signalling channel collects The peak-peak of signalAnd as first input element of RBF combination neural net；

Primary signal carries out Fourier transform obtain the i-th signalling channel power spectrum under different leak pressure and letting out Spectrum peak at leakage frequencyAnd as second input element of RBF combination neural net；

By primary signal through using db1 wavelet basis to carry out three layers of WAVELET PACKET DECOMPOSITION, obtain different leak pressure The wavelet-packet energy of lower i-th signalling channel (3,0) nodeAnd as RBF combination neural net 3rd input element；Wherein, k is the numbering of corresponding different leak pressure.

Method the most according to claim 2, it is characterised in that let out except regulation simulation in described step 3 Gas pressure in leakage unit obtains outside the feature of the gas leakage signal under different condition, also by Change temperature, the mode in leakage aperture obtain the feature of the leakage signal under different condition.