CN109084186A - Pipeline leakage signal recognition methods based on improved ELMD multi-scale entropy - Google Patents
Pipeline leakage signal recognition methods based on improved ELMD multi-scale entropy Download PDFInfo
- Publication number
- CN109084186A CN109084186A CN201810937852.1A CN201810937852A CN109084186A CN 109084186 A CN109084186 A CN 109084186A CN 201810937852 A CN201810937852 A CN 201810937852A CN 109084186 A CN109084186 A CN 109084186A
- Authority
- CN
- China
- Prior art keywords
- signal
- component
- entropy
- elmd
- waveform
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F17—STORING OR DISTRIBUTING GASES OR LIQUIDS
- F17D—PIPE-LINE SYSTEMS; PIPE-LINES
- F17D5/00—Protection or supervision of installations
- F17D5/02—Preventing, monitoring, or locating loss
- F17D5/06—Preventing, monitoring, or locating loss using electric or acoustic means
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M3/00—Investigating fluid-tightness of structures
- G01M3/02—Investigating fluid-tightness of structures by using fluid or vacuum
- G01M3/04—Investigating fluid-tightness of structures by using fluid or vacuum by detecting the presence of fluid at the leakage point
- G01M3/24—Investigating fluid-tightness of structures by using fluid or vacuum by detecting the presence of fluid at the leakage point using infrasonic, sonic, or ultrasonic vibrations
- G01M3/243—Investigating fluid-tightness of structures by using fluid or vacuum by detecting the presence of fluid at the leakage point using infrasonic, sonic, or ultrasonic vibrations for pipes
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Examining Or Testing Airtightness (AREA)
- Complex Calculations (AREA)
Abstract
The present invention provides a kind of pipeline leakage signal recognition methods based on improved ELMD multi-scale entropy, first to experimental data progress noise pretreatment is obtained, eliminates low Related Component in signal；Then ELMD processing is carried out to preprocessed signal, obtains each PF component；Weaken ELMD by peak value waveform matching method and decomposes the end effect problem retained；The multi-scale entropy of PF component is calculated separately, arrangement compares the multiple dimensioned entropy of leakage signal to eliminate ambient noise；Main PF component construction feature vector is selected according to multiple dimensioned entropy；Using feature vector as the input vector of BP neural network, network is trained；Sample to be tested is inputted into acquisition pipe leakage recognition result in trained BP neural network.Method provided by the invention can adapt to the various situations of pipeline, have preferable detection accuracy.
Description
Technical field
The present invention relates to Discussion on Pipe Leakage Detection Technology field, more particularly to a kind of based on improved ELMD multi-scale entropy
Pipeline leakage signal recognition methods.
Background technique
Urban duct has become the indispensable tool of Modern Urban Development, with the continuous expansion of its scale, due to setting
Standby natural aging, climatic environment and artificial destruction etc. influence, and pipeline fault event is in rising trend, especially gas pipeline one
Denier leakage, it is easy to cause the serious accidents such as fire, explosion, poisoning, environmental pollution.Therefore effective pipeline leakage testing is found
Leakage hidden danger is recognized accurately in method, has good economic value and social effect.
In recent years, with the development of computer technology, the direction of Discussion on Pipe Leakage Detection Technology forward direction software and hardware combining is developed,
Currently, various new pipeline leakage detection methods are still the forward position research direction of various countries.Local mean value decomposes (local mean
Decomposition, LMD) method is a kind of new adaptive Time Frequency Analysis method, it can be adaptive by a multicomponent data processing
Ground resolves into multiple amplitude modulationfrequency modulation component signals, has preferable time-frequency characteristics extractability, but LMD decomposition result equally exists
Modal overlap phenomenon.Its innovatory algorithm: overall local mean value decompose (ensenmble local mean decomposition,
ELMD modal overlap phenomenon) is then overcome, is more suitable for the complicated leakage signal of analysis, and survey accuracy rate and improve a lot.
It is relatively difficult due to obtaining entire pipeline whole pipeline section pressure data, especially pipeline can not be close pipeline section it is several
Pressure data can not be got, it is existing that pipeline is carried out according to pressure data in the case where lacking whole tube section pressure data
The method of leak detection is all to be calculated that it is very ideal that this, which results in detection effect not, according to few force samples point.
In recent years, with the development of detection technique, the method for occurring getting entire pipeline pressure data a bit, but it is based on whole tube section pressure
The method that force data carries out leak detection but has no proposition, and BP neural network be then can according to the pressure data of whole tube section into
The method of row leak detection, BP neural network have Serial Distribution Processing, self-organizing, adaptive, self study and zmodem etc.
Unique excellent performance.
Summary of the invention
The technical problems to be solved by the present invention are: for the problems in existing Discussion on Pipe Leakage Detection Technology, the present invention
A kind of pipeline leakage signal recognition methods based on improved ELMD multi-scale entropy is provided.This method is with whole tube section in specific pipeline
Pressure data based on, first according to auto-correlation function eliminate signal in low Related Component；Then using ELMD method to letter
It number is decomposed, can effectively overcome modal overlap phenomenon, obtain accurate PF component；It is solved using extreme point continuation method
Decompose the end effect problem generated；Ambient noise is handled by calculating the multiple dimensioned entropy of each PF component；Finally by BP
Neural network can quickly and accurately judge whether pipeline leaks, and ensure the safe operation of urban duct, so that it is guaranteed that people
The property and life security of the people masses.
The present invention solves its technical problem technical solution to be taken: since the signal to acquisition is decomposed using LMD
After can generate modal overlap phenomenon, be described using Conventional mathematical model there are extremely difficult, the present invention from artificial intelligence and
The angle of signal processing is set out, and a kind of pipeline leakage signal recognition methods based on improved ELMD multi-scale entropy is provided, to whole
The pressure data of a pipeline is analyzed, and has preferable superiority in terms of signal processing, feature extraction, accident analysis, is wrapped
Include following steps:
S1: obtaining experimental data, in material due to crack extension, plastic deformation or phase transformation etc. cause strain energy quick release
The referred to as sound emission of the phenomenon that generation.Material property or structural integrity are evaluated by the acoustic emission signal of reception and analysis material
The lossless detection method of property.Signal data is acquired respectively using sound emission leakage system, obtains original leakage signal x (b)；
S2: auto-correlation pretreatment is carried out to original leakage signal x (b), calculates auto-correlation function rx, and carry out time domain point
Analysis eliminates low Related Component in signal, obtains pretreated leakage signal x (t)；It must not in the signal acquired by sound emission
Ambient noise and some low relevant ingredients can be contained less, and although ratio that the low Related Component in signal accounts for is little,
But signal processing and reconstruct can be had an important influence on.The auto-correlation function of random signal reflect signal and its own not
With the degree of correlation at time point.Its method particularly includes:
In formula, x (b) is original leakage signal, and x (b+a) is the leakage signal of delay time a, and a, b are positive integer；
Auto-correlation function reflects signal x (b) and its own has made the similarity degree of the x (b+a) after one section of delay, by signal
Auto-correlation is sought to carry out Signal Pretreatment, eliminates some low Related Components of signal itself.
S3:ELMD method has the property for equably " polluting " the entire time frequency space of echo signal using white noise signal,
Limited different white noise signal is adulterated in echo signal every time, then mixed signal is decomposed using LMD method,
All scales in echo signal will automatically be decomposed associated in filter group determined by white noise signal
In passband, the signal-to-noise ratio of decomposition result is reduced, also therefore alleviates modal aliasing phenomenon.But due to the difference adulterated every time
It is uncorrelated between white noise, and only its mean value of the enough white noises of quantity can just go to zero, therefore more using LMD method
It is secondary that the mixed signal for adulterating different white noises is decomposed, population mean then is made to the PF component obtained after multiple decompose, and
As final decomposition result, such white noise is cancelled during population mean, therefore the signal-to-noise ratio of signal also obtains
To raising.
ELMD (overall local mean value is decomposed) processing is carried out to pretreated leakage signal x (t), obtains each PF component yn
With residual components εr；Wherein ELMD decomposition step includes:
S3.1: determining the white noise grade of population mean number M and addition, if initial decomposition number m=1.M value and white noise
Sound grade is selected according to the actual situation, preferably M=100, and the white noise grade of addition is 0.1.
S3.2: the white noise n for determining grade is added in preprocessed signal x (t)m(t), then leakage signal can indicate
Are as follows:
xm(t)=x (t)+nm(t) (2)
Wherein, wherein m is to decompose number, and t is time, nmIt (t) is white noise signal, xm(t) mixed after white noise to be added
Close leakage signal.
S3.3: to mixing leakage signal xm(t) it carries out m LMD to decompose, obtains multiple PF component εn,m(n=1,2 ...,
.N), N is positive integer.εn,mN-th obtained of PF component is decomposed for the m times.
S3.4: if m < M, repeatedly step S3.2, S3.3, decomposes number and adds 1.
The PF component population mean of S3.5:M decomposition are as follows:
Wherein, n=1,2 ..., N, m=1,2 ..., M, N, M are positive integer；
The mean value y that N number of PF component M times is decomposedn(n=1,2 ..., N) as final PF component.As M=100, i.e.,
The mean value y that N number of PF component 100 times are decomposedn(n=1,2 ..., N) as final PF component.
S4: continuation method is matched by peak value waveform to pretreated signal x (t) and weakens the endpoint effect that ELMD decomposition retains
It answers, the signal after obtaining continuation is denoted as x ' (t), carries out ELMD decomposition, obtained PF component y ' againnWith residual components ε 'r。
ELMD is decomposed overcomes modal overlap problem compared to LMD decomposition to a certain extent, but there are still reconstructed errors
Greatly, the noise information adulterated in component is more, this is because in LMD decomposing program, since local mean value envelope function is at endpoint
There is one section of unknown spurious signal, if ignoring this or processing method is improper, can finally be generated when program is run
Deceptive information influences decomposition result.In response to this problem, continuation method is matched using peak value waveform to weaken end effect problem:
Continuation is carried out to the data at endpoint, key seeks to determine variation tendency of the initial data at endpoint.Peak value
Waveform Matching continuation method is based on adaptive Waveform Matching method, i.e., the waveform pair for being best suitable for signal trend is found out inside original signal
Signal carries out continuation, the inherent trend of maximum maintenance signal, and according to the peak factor of waveform, by given threshold come
It realizes.It is illustrated below with the continuation of left end point.
Left end point continuation is first carried out, if pretreated signal is x (t), p1、q1Respectively the waveform of signal x (t) when
Between be tp、tqThe left end point value of the maximum and minimum at place, x (t) is x (1), with x (1)-p1—q13 points constitute one three
Angle waveform, and referred to as signature waveform, then along signal x (t) search and signature waveform triangular waveform the most matched；General
Continuation waveform with the data before waveform as x (t), will meet the natural tendency of signal；Specific step is as follows:
S4.1: finding the starting point x (i) of the triangular waveform in addition to signature waveform, and corresponding time point is tx(i), then
S4.2: the matching error e (i) of all triangular waveforms and signature waveform, error formula are calculated are as follows:
E (i)=| pi-p1|+|qi-q1|+|xi-x(1)| (5)
S4.3: the smallest error amount e (i) is found out, e is denoted asmin(i), and given threshold α, if emin(i) < α, then by emin
(i) corresponding triangular waveform is as matching waveform, before the Data extension to original signal before matching waveform, if emin(i)≥
α is carried out in next step.
S4.4: if emin(i) >=α calculates the peak factor F in each triangular waveform,
And given threshold β, if Fmin< β, then repeatedly S4.3；If Fmin>=β, the extreme value at direct setting signal endpoint, into
Row is in next step.
S4.5: if Fmin>=β, the extreme value at direct setting signal endpoint, i.e., find out adjacent near signal left end respectively
The average value of two maximum value minimums, respectively as the maximum of signal x (t) and minimum.
S4.6: carrying out right endpoint continuation in the same way, if the signal completely after continuation is x 't, ELMD decomposition is carried out,
Carry out ELMD decomposition, obtained PF component y 'nWith residual components ε 'r.Wherein, threshold alpha is derived from waveform from error amount e (i), β
Peak factor F, the two can be adjusted according to practical programs operating condition, and α is smaller to illustrate that original signal has stronger regularity, β
It is bigger to illustrate that the data boundary between matched waveform and original signal is more identical.
S5: each PF component y ' after calculating continuationnMultiple dimensioned entropy MSEn, the proposition of multi-scale entropy (MSE) is in sample
On the basis of entropy, initial data is subjected to coarse and the sample entropy on each scale is formed into one group of ordered series of numbers, i.e. time sequence
The Sample Entropy being listed under different scale.If a sequence and another sequence are under scale of the same race, the former entropy compares the latter
Height, this illustrates that the complexity of the former time series is higher than the latter.There is scholar that multi-scale entropy is applied to rolling bearing at present
And Fault Diagnosis for Rotor System, the results showed that multi-scale entropy can effectively distinguish various failures.With traditional based on single ruler
The Sample Entropy of degree is compared, and MSE can be preferably from the complexity features of multiple scale reflecting time sequences, and have and calculate institute
The advantages that data are short, stability is good, anti-noise ability is strong is needed, biology, myoelectricity, brain electrically and mechanically failure are widely used in
In the analysis of equal signals.
The following steps are included:
S5.1: a discrete original time series are set as { x1, x2 ..., xn }, coarse change is carried out to original time series
It changes, obtains new time series:
Wherein,μ,For positive integer andS is discrete-time series length, and τ is scale
The factor, it is the coarse grain sequence of s/ τ that original series, which are divided into τ sections and every segment length, and as τ=1, new time series is exactly original
Sequence；It is preferred that τ=3.
S5.2: mould-fixed dimension k and similar tolerance r (r > 0) are given, the k dimensional vector of build time sequence:
xk(μ)={ xμ,xμ+1,…,xμ+k-1I=1,2 ..., S-k (8)
S5.3: vector x is calculatedk(μ) withThe distance between:
Wherein, L=0,1 ..., k-1；μ,And
S5.4: to each μ, x is calculatedk(μ) withDistance, number of the statistical distance less than r and by this number
The ratio of distance sum S-k+1 is denoted as Ckμ(r), i.e.,
S5.5: by Ckμ(r) average value is denoted as
S5.6: dimension plus 1, become k+1, repeat step S5.1 and S5.5, calculate
When S is finite value, sample Entropy estimate when the sequence length obtained by above-mentioned steps is S:
S5.7: repeating the above process, and obtains the sample entropy SampEn under different scale, is chosen at optimal under each scale
As multi-scale entropy.Multi-scale entropy is derived from Sample Entropy, can be used to indicate the complexity of signal, i.e. entropy is bigger, signal
It is more complicated.The multiple dimensioned entropy that each PF component after continuation is finally calculated is denoted as: MSE1, MSE2, MSE3…MSEn。
S6: main PF component construction feature vector E is selected according to multiple dimensioned entropy, using feature vector E as BP neural network
Input vector, network is trained.Artificial neural network (artificial neural networks, ANNs) is simulation
A kind of mathematical model of biological neural network progress information processing.And BP (back propagation) neural network is a kind of base
In the neural network of error back propagation training algorithm, it is advantageous that, in the case where hidden layer and enough number of nodes, it
Arbitrary Nonlinear Mapping relationship can be approached, and there is preferable generalization ability.Sample to be tested is inputted into trained BP mind
Through obtaining pipe leakage result in network.Leakage signal classifying identification method based on BP neural network mainly includes following
Step:
S6.1: signature analysis is carried out to each sound leakage signal to be analyzed, by calculating each PF component y 'nIt is multiple dimensioned
Entropy MSEn, the size for arranging more multiple dimensioned entropy carrys out construction feature vector E=(MSE1, MSE2…MSEn)；
S6.2: by different feature vector E1、E2、EnIt constructs trained input matrix A, tests output matrix B, wherein n is
Positive integer, according to training sample in BP neural network and test sample ratio (such as: ratio 9:1) determine matrix A, B
Sample number.And then according to input matrix A and desired output matrix B, the system parameter of BP neural network is determined；Data are by experiment
The multiple dimensioned entropy composition of the PF component of pipe leakage under different pressures.
S6.3: the input characteristic parameter vector matrix of training sample is input in BP neural network classifier and is learned
It practises, obtains neural network classification model；
S6.4: the input characteristic parameter vector matrix of sample to be identified is input to trained neural network classifier
In, the Classification and Identification of signal is carried out, and export and finally identify true leakage signal.After BP neural network training, obtain pre-
Error between measured value and true value, error are indicated by root-mean-square value.The identification of pipe leakage is finally obtained according to error
Accuracy rate.
The beneficial effects of the present invention are: a kind of pipe leakage letter based on improved ELMD multi-scale entropy provided by the invention
Number recognition methods, this method acquire pipeline leakage signal using acoustic emission system, have carried out Signal Pretreatment first, eliminate letter
Low Related Component in number.Signal is decomposed by the decomposition of overall local mean value, has obtained several PF components.Pass through peak value wave
Shape matching continuation method weakens ELMD and decomposes the end effect problem retained.Then the multiple dimensioned entropy of PF component is calculated, and is arranged
The size of the multi-scale entropy of comparison signal judges its complexity, eliminates existing residual components and background after ELMD is decomposed
Noise.Finally selection carrys out construction feature vector comprising the PF component of main leakage feature, using this feature vector as BP nerve
The input vector of network, is trained network, finally inputs in trained BP neural network sample to be tested and obtains pipeline
Leak result.The present invention can accurately identify pipe leakage, and essentially eliminate the ambient noise in signal, low correlation
Ingredient and residual components, so that leakage recognition result is more accurate.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples.
Fig. 1 is the flow diagram of preferred embodiment；
Fig. 2 is typical BP neural network topology diagram；
Fig. 3 is the time-domain and frequency-domain waveform diagram of original signal；
Fig. 4 is leakage signal autocorrelation analysis figure；
Fig. 5 is leakage signal ELMD decomposition result；
Fig. 6 is PF component variation figure before peak value waveform matching continuation；
Fig. 7 is PF component variation figure after peak value waveform matching continuation；
Fig. 8 is BP neural network algorithm flow chart；
Fig. 9 typical case's BP neural network topological structure；
Figure 10 is BP neuron models figure；
Figure 11 is BP neural network training error change curve；
Figure 12 is predicted value and true value comparison diagram；
Figure 13 is error change figure；
Figure 14 is neural network target to output matched curve.
Specific embodiment
Presently in connection with attached drawing, the present invention is described in detail.This figure is simplified schematic diagram, is only illustrated in a schematic way
Basic structure of the invention, therefore it only shows the composition relevant to the invention.
A kind of pipeline leakage signal recognition methods based on improved ELMD multi-scale entropy of the invention, if Fig. 1 is this hair
Bright specific flow chart.Integrated application overall local mean value decomposition algorithm, multi-scale entropy, BP neural network, this method are specific
Steps are as follows:
As shown in Fig. 2, this experiment simulation duct length is 42m, pipe material is steel, pipeline specifications DN90, medium
For compressed air, pipeline internal medium is in flow regime.No. 1 upstream sensor set-down location is 0m, in No. 1 sensor 18m of distance
Place is to leak the leak that aperture is 1mm.No. 2 downstream sensors are fixed in the place installation of No. 1 upstream sensor 42m of distance, i.e.,
Two sensors spacing 42m.
The original signal waveform of sound emission leak detection system acquisition pipe leakage, time domain, frequency domain are utilized in step S1
Distribution is as shown in Figure 3.
Signal Pretreatment is carried out in step S2, by sound emission acquire signal in it is essential containing ambient noise with
And some low relevant ingredients, the auto-correlation function of random signal reflect signal and its own related journey in different time points
Degree.And although ratio that the low Related Component in signal accounts for is little, but can have an important influence on to signal processing and reconstruct.Benefit
Autocorrelation analysis is carried out to original signal with matlab, analysis result is as shown in Figure 4.
ELMD decomposition is carried out to pretreated signal in step S3.ELMD method refers to before LMD decomposition, and multiple groups are limited
A different white noise signal is added in signal to be decomposed, is zero using white noise mean value, the characteristic that spectrum energy is evenly distributed,
White noise is evenly distributed in entire time frequency space, and the signal of different time scales can be distributed to automatically with
The relevant appropriate scale of ambient noise gets on.The fixed white noise grade being added is 0.1, and population mean number is 100, ELMD points
It is as shown in Figure 5 to solve result.From fig. 5, it can be seen that pretreated signal has obtained a series of PF components after ELMD is decomposed
With a residual components, preceding 4 PF components containing main information are only shown here.
In step S4, ELMD is decomposed overcomes modal overlap problem compared to LMD decomposition to a certain extent, but there are still
Reconstructed error is big, the problem more than noise information adulterated in component.This is because in ELMD decomposing program, due to local mean value packet
There is one section of unknown spurious signal in network function, finally can be if ignoring this or processing method is improper at endpoint
Program generates deceptive information when running, and influences decomposition result.In response to this problem, continuation method is matched using peak value waveform to weaken end
Point effect problem.Peak value waveform matching continuation method mainly calculates two parameter waveform matching error e(i), peak factor F, continuation effect
Fruit sees Fig. 6 and Fig. 7.Fig. 6 and Fig. 7 is that PF component map before and after end effect is solved with peak value waveform matching continuation respectively, from figure
This it appears that waveform of the PF component at endpoint is more preferable in Fig. 7, and end effect is obvious after peak value waveform matches continuation
Weaken.Therefore peak value waveform matching continuation method can weaken ELMD decomposition bring end effect to a certain extent.
In step S5, in order to further eliminate existing ambient noise and residual components after ELMD is decomposed, improves pipeline and let out
Leak the accuracy rate of identification.The multi-scale entropy of PF component is calculated, and arranges the multiple dimensioned entropy of comparison signal, multiple dimensioned entropy reflection
The complexity of leakage signal.
When pipeline leaks, different pore size, different pressures, different carriers frequency distribution variation can cause vibrational energy
Variation, vibration signal virtual value can also change.The amplitude instantaneous value that the virtual value of vibration signal describes signal is being adopted
The size of sample time internal vibration, it is the average value to time series, is adapted to analysis continuous shaking signal.Pipeline leakage signal
Multi-scale entropy is selected with uncertain, complexity and non-linear behavior, the present invention to judge the complexity of pipeline leakage signal
Degree, while calculating sample entropy and comparing.Preprocessed signal 30 are taken at random, calculates separately its multiple dimensioned entropy and sample
Entropy is shown in Table 1, table 2.
Table 1 is the multiple dimensioned entropy of PF component；
Table 2 is the sample entropy of PF component；
The leakage information that PF component is included as can be seen from Table 1 is from PF1To PF4Gradually successively decrease.Compare the more of PF component
Scale entropy, it can be seen that PF1—PF3The multiple dimensioned entropy of component is relative to PF4It is bigger, illustrate the leakage information that the former includes
More.Therefore preceding 3 PF component construction feature vector is selected, ambient noise, residual components pair can be greatly reduced in this way
The influence of leakage signal.Feature vector is input to BP neural network to be trained, identify.
Multiple dimensioned entropy and sample entropy in 1 table 2 of contrast table, it is apparent that, under same PF component, multiple dimensioned entropy
To be slightly larger than sample entropy, this is related with the principle of multi-scale entropy and Sample Entropy, and entropy is bigger, and the effective information for including is more.
So the present invention selects the multiple dimensioned entropy of PF component to construct the input feature value of BP neural network.
BP neural network algorithm flow is shown in Fig. 8, including initiation parameter, input value and desired value, normalization data and tune
Whole weight etc..Fig. 9, Figure 10 are shown in the setting of BP neural network specific structure.The initiation parameter of BP neural network is set as 3 etale topologies
Structure, hidden layer and output layer transmission function select logarithm S-shaped, and training algorithm selects the decline of autoadapted learning rate momentum gradient
Method.The initiation parameter of network is arranged are as follows: frequency of training k=2000, target error coefficient e=0.005, learning rate η=0.01,
Factor of momentum 0.9, node in hidden layer 20.
The calculating of multi-scale entropy is carried out according to the PF component after decomposing in step S5 to ELMD, and selects preceding 3 PF components
Multiple dimensioned entropy construction feature vector.
According to progress BP neural network identification in step S6.30 groups of pretreated samples are taken, respectively from different pressures
Under (0.1mpa, 0.2mpa, 0.3mpa) pipeline leakage signal, every group of sample includes the multiple dimensioned entropy of 3 PF components.Wherein
27 groups are selected as training sample, 3 groups are test sample, and the mean square error performance change situation of training process is as shown in figure 11.
It can be seen from fig. 11 that with the increase of iterative steps, mean square error between network output valve and target value by
Gradual change is small.In the 505th iteration, the mean square error between output valve and target value is 0.0049726, is less than setting value 0.005,
The training effect of setting is reached.
The classification results of test sample are as shown in figure 12.Compared with true value, in Figure 13 accidentally by predicted value in Figure 12
Poor variation diagram can be seen that the multiple dimensioned entropy of the pipeline leakage signal after BP neural network training and actual very close, accidentally
Difference is smaller, very high to the recognition accuracy of pipe leakage, has reached the requirement to leakage signal identification.While BP neural network
Training time is 0.2s, and training process is very rapid.
Figure 14 is shown in BP neural network training matched curve.Abscissa in figure is target value, and ordinate is the output of network
Value.To prevent over-fitting, the method for use is data to be divided into three points, training (training), validation (verifying),
Test (test).As can be seen from Figure 14, training fitting result distribution uniform, is distributed in diagonal line, and fitting effect is good.
Finally, the improved ELMD method of pipeline leakage signal is decomposed, end effect is reduced, it is multiple dimensioned by calculating
Entropy construction feature vector is input to BP neural network training identification, eliminates the low Related Component in test signal, remnants divide
Amount and ambient noise.BP neural network recognition result is shown in Table 3:
Table 3 is BP neural network training recognition result
From table 3 it is observed that constructing BP neural network by the multiple dimensioned entropy of PF component after calculating ELMD decomposition
Input vector, final recognition result be higher than calculate sample entropy.Therefore, multi-scale entropy and BP neural network based on ELMD
Pipe leakage recognition methods first to pipeline leakage signal carry out autocorrelation analysis remove low Related Component, then improve
ELMD decompose, weaken end effect, and calculate multiple dimensioned entropy, choose the PF component comprising main leakage information constitute it is special
Vector is levied, identification is finally trained by BP neural network.By Figure 12-Figure 14 it is found that the method that is mentioned of the present invention is feasible and know
Other effect is good.
Taking the above-mentioned ideal embodiment according to the present invention as inspiration, through the above description, relevant staff
Various changes and amendments can be carried out without departing from the scope of the present invention completely.The technical scope of this invention is not
The content being confined on specification, it is necessary to which the technical scope thereof is determined according to the scope of the claim.
Claims (6)
1. a kind of pipeline leakage signal recognition methods based on improved ELMD multi-scale entropy, it is characterised in that: including following step
Suddenly,
S1: obtaining experimental data, utilizes the leakage signal x (b) of sound emission leak detection system acquisition pipeline；
S2: auto-correlation function r is calculated to leakage signal x (b)x, and time-domain analysis is carried out, low Related Component in signal is eliminated, is obtained
Pretreated leakage signal x (t)；
S3: determining the white noise grade of population mean number M and addition, to the pretreated leakage signal containing ambient noise
X (t) carries out ELMD decomposition, obtains a series of PF component ynWith residual components εr, wherein yn=εn,m(n=1,2 ... .N)；
S4: continuation method is matched by peak value waveform to pretreated leakage signal x (t) and weakens the endpoint effect that ELMD decomposition retains
It answers, the signal after obtaining continuation is x ' (t), carries out ELMD decomposition, obtained PF component y 'nWith residual components ε 'r；
S5: each PF component y ' after calculating continuationnMultiple dimensioned entropy MSEn, n is positive integer, and according to the big of multiple dimensioned entropy
Small selection carrys out construction feature vector E=(MSE comprising the PF component of a large amount of leakage informations1, MSE2…MSEn), it further obviates
Ambient noise and residual components in original leakage signal；
S6: using feature vector E as the input vector of BP neural network, neural network is set as 3 etale topology structures, wherein implying
Layer selects logarithm S-shaped with output layer transmission function, and training algorithm selects autoadapted learning rate momentum gradient descent method, to network
It is trained, tests, it is final to carry out pipeline leakage signal identification.
2. the pipeline leakage signal recognition methods as described in claim 1 based on improved ELMD multi-scale entropy, feature exist
In: the auto-correlation function of leakage signal x (b) indicates in step S2 are as follows:
In formula, x (b) is the pressure signal that the original leakage signal of pipeline is converted by Acoustic radiating instrument, when x (b+a) is delay
Between a leakage signal, and a, b are positive integer；Auto-correlation function reflects signal x (b) and its own has been made one section and has postponed it
The similarity degree of x (b+a) afterwards.
3. the pipeline leakage signal recognition methods as claimed in claim 2 based on improved ELMD multi-scale entropy, feature exist
In: in step S3 ELMD decompose specifically includes the following steps:
S3.1: determining the white noise grade of population mean number M and addition, if initial decomposition number m=1；
S3.2: the white noise n for determining grade is added in preprocessed signal x (t)m(t), then leakage signal indicates are as follows:
xm(t)=x (t)+nm(t) (2)
Wherein, m is to decompose number, and t is time, nmIt (t) is white noise signal, xm(t) letter is leaked for the mixing after white noise is added
Number；
S3.3: to mixing leakage signal xm(t) it carries out m LMD to decompose, obtains multiple PF component εn,m(n=1,2 ... .N), N
For positive integer；εn,mN-th obtained of PF component is decomposed for the m times；
S3.4: if m < M, repeatedly step S3.2, S3.3, decomposes number m and adds 1；
The PF component population mean of S3.5:M decomposition are as follows:
Wherein, n=1,2 ..., N, m=1,2 ..., M, N, M are positive integer；
The mean value y that N number of PF component M times is decomposedn(n=1,2 ..., N) as final PF component.
4. the pipeline leakage signal recognition methods as claimed in claim 3 based on improved ELMD multi-scale entropy, feature exist
In step S4 specifically includes the following steps:
Left end point continuation is first carried out, if pretreated signal is x (t), p1、q1Respectively the waveform of signal x (t) is in time tp、
tqThe maximum and minimum at place, if the left end point value of signal x (t) waveform is x (1), with x (1)-p1—q13 points constitute one
Triangular waveform, and referred to as signature waveform, then along the waveform searching and signature waveform triangle the most matched of signal x (t)
Waveform；Using the data before matching waveform as the continuation waveform of x (t), the natural tendency of signal will be met；Specific steps are such as
Under:
S4.1: finding the starting point x (i) of the triangular waveform in addition to signature waveform, and corresponding time point is tx(i), then
S4.2: the matching error e (i) of all triangular waveforms and signature waveform, error formula are calculated are as follows:
E (i)=| pi-p1|+|qi-q1|+|xi-x(1)| (5)
S4.3: the smallest error amount e (i) is found out, e is denoted asmin(i), and given threshold α, if emin(i) < α, then by emin(i) right
The triangular waveform answered is as matching waveform, before the Data extension to original signal before matching waveform, if emin(i) >=α, into
Row is in next step；
S4.4: if emin(i) >=α calculates the peak factor F in each triangular waveform,
And given threshold β, if Fmin< β, then repeatedly S4.3；If Fmin>=β, the extreme value at direct setting signal endpoint, carries out down
One step；
S4.5: if Fmin>=β, the extreme value at direct setting signal endpoint are found out respectively near adjacent two in signal left end
The average value of maximum value minimum, respectively as the maximum of signal x (t) and minimum；
S4.6: carrying out right endpoint continuation in the same way, if the signal completely after continuation is x 't, ELMD decomposition is carried out, is carried out
ELMD is decomposed, obtained PF component y 'nWith residual components ε 'r。
5. the pipeline leakage signal recognition methods as claimed in claim 4 based on improved ELMD multi-scale entropy, feature exist
In: the multiple dimensioned entropy specific steps that step S5 calculates PF component include:
S5.1: PF component y ' is setnOriginal time series be { x1,x2,…,xn, data are believed by the pressure that Acoustic radiating instrument obtains
It number converts, coarse transformation is carried out to original time series, obtains new time series:
Wherein,For positive integer andS be discrete-time series length, τ be scale because
Son, it is the coarse grain sequence of s/ τ that original series, which are divided into τ sections and every segment length, and as τ=1, new time series is exactly original sequence
Column；
S5.2: mould-fixed dimension k and similar tolerance r (r > 0) are given, the k dimensional vector of build time sequence:
xk(μ)={ xμ,xμ+1,…,xμ+k-1} (8)
Wherein, i=1,2 ..., s-k；
S5.3: vector x is calculatedk(μ) withThe distance between:
Wherein, L=0,1 ..., k-1；And
S5.4: to each μ, x is calculatedk(μ) withDistance, statistical distance is less than the number of the r and distance of this number is total
The ratio of number s-k+1 is denoted as Ckμ(r), then
S5.5: by Ckμ(r) average value is denoted asThen
S5.6: dimension plus 1, becoming k+1, repeats step S5.1 to S5.5, calculates
Sample Entropy estimate when S is finite value, when the sequence length obtained by above-mentioned steps is s are as follows:
S5.7: repeating the above process, and obtains the sample entropy SampEn under different scale, is chosen at and optimal under each scale is
Multi-scale entropy MSEn, the multiple dimensioned entropy that each PF component after continuation is finally calculated is denoted as: MSE1, MSE2, MSE3…
MSEn。
6. the pipeline leakage signal recognition methods as claimed in claim 5 based on improved ELMD multi-scale entropy, feature exist
It is specifically included in the step of: step S6 construction feature vector:
S6.1: each PF component y ' is calculatednMultiple dimensioned entropy MSEn, choose multiple dimensioned entropy it is biggish come construction feature vector E=
(MSE1, MSE2…MSEn；
S6.2: by different feature vector E1、E2…EnCome construct trained input matrix A, test output matrix B, wherein n is positive
Integer determines the sample number of matrix A, B according to training sample in BP neural network and test sample ratio, according to input matrix
A and desired output matrix B, determine the system parameter of BP neural network；
S6.3: the input feature value Input matrix of training sample is learnt into BP neural network classifier, obtains mind
Through network class model；
S6.4: the input characteristic parameter vector matrix of sample to be identified is input in trained neural network classifier,
The Classification and Identification of signal is carried out, and exports and finally identifies true leakage signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810937852.1A CN109084186B (en) | 2018-08-17 | 2018-08-17 | Pipeline leakage signal identification method based on improved ELMD (ensemble empirical mode decomposition) multi-scale entropy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810937852.1A CN109084186B (en) | 2018-08-17 | 2018-08-17 | Pipeline leakage signal identification method based on improved ELMD (ensemble empirical mode decomposition) multi-scale entropy |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109084186A true CN109084186A (en) | 2018-12-25 |
| CN109084186B CN109084186B (en) | 2020-05-26 |
Family
ID=64793731
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810937852.1A Active CN109084186B (en) | 2018-08-17 | 2018-08-17 | Pipeline leakage signal identification method based on improved ELMD (ensemble empirical mode decomposition) multi-scale entropy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109084186B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109630908A (en) * | 2019-01-23 | 2019-04-16 | 常州大学 | A kind of pipeline leakage positioning method of multiple noise reduction |
| CN110146292A (en) * | 2019-06-04 | 2019-08-20 | 昆明理工大学 | A kind of rolling bearing fault testing method that the overall local mean value based on the reconstruct of integrated noise is decomposed |
| CN110263832A (en) * | 2019-06-11 | 2019-09-20 | 哈尔滨工程大学 | A kind of AUV navigation system method for diagnosing faults based on multiscale analysis |
| CN110730037A (en) * | 2019-10-21 | 2020-01-24 | 苏州大学 | Optical signal-to-noise ratio monitoring method of coherent optical communication system based on momentum gradient descent method |
| CN111397700A (en) * | 2020-03-02 | 2020-07-10 | 西北工业大学 | Wall-mounted fault detection method of Coriolis mass flow meter |
| CN111444805A (en) * | 2020-03-19 | 2020-07-24 | 哈尔滨工程大学 | Improved multi-scale wavelet entropy digital signal modulation identification method |
| CN111735583A (en) * | 2020-06-24 | 2020-10-02 | 东北石油大学 | Pipeline working condition identification method based on LCD-EE pipeline sound wave signal characteristic extraction |
| WO2021012986A1 (en) * | 2019-07-22 | 2021-01-28 | 常州大学 | Pipeline multi-point leakage positioning method based on improved vmd |
| CN113722977A (en) * | 2021-05-26 | 2021-11-30 | 上海华电奉贤热电有限公司 | Gas turbine rotor fault early warning method based on hybrid prediction |
| CN114234061A (en) * | 2021-12-20 | 2022-03-25 | 北京工业大学 | Neural network-based intelligent judgment method for water leakage sound of pressurized operation water supply pipeline |
| CN114576568A (en) * | 2022-02-25 | 2022-06-03 | 辽宁石油化工大学 | Pipeline leakage detection method and device based on infrasonic waves |
| CN115111537A (en) * | 2022-08-24 | 2022-09-27 | 北京云庐科技有限公司 | Method, device and medium for determining the position of a leak in a gas pipeline network |
| CN115204243A (en) * | 2022-09-15 | 2022-10-18 | 西南交通大学 | LMD endpoint effect improvement method based on similar triangular waveform matching continuation |
| CN116415119A (en) * | 2023-04-26 | 2023-07-11 | 山东大学 | Entropy aliasing and feature enhancement-based gas abnormal signal detection method and system |
| CN116415119B (en) * | 2023-04-26 | 2024-06-28 | 山东大学 | Entropy aliasing and feature enhancement-based gas abnormal signal detection method and system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20150041564A (en) * | 2013-10-08 | 2015-04-16 | (주)유민에쓰티 | Alkali solution leak detection apparatus |
| US20160023876A1 (en) * | 2013-03-14 | 2016-01-28 | Khs Gmbh | Method and filling system for filling containers |
| CN106598910A (en) * | 2016-12-28 | 2017-04-26 | 四川中烟工业有限责任公司 | EMD (Empirical Mode Decomposition) end effect inhibiting method and system |
| CN106778594A (en) * | 2016-12-12 | 2017-05-31 | 燕山大学 | Mental imagery EEG signal identification method based on LMD entropys feature and LVQ neutral nets |
| CN107228282A (en) * | 2017-07-06 | 2017-10-03 | 东北石油大学 | A kind of gas pipeline leakage localization method and device |
| CN108152363A (en) * | 2017-12-21 | 2018-06-12 | 北京工业大学 | A kind of defect of pipeline recognition methods for pressing down end intrinsic time Scale Decomposition |
-
2018
- 2018-08-17 CN CN201810937852.1A patent/CN109084186B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160023876A1 (en) * | 2013-03-14 | 2016-01-28 | Khs Gmbh | Method and filling system for filling containers |
| KR20150041564A (en) * | 2013-10-08 | 2015-04-16 | (주)유민에쓰티 | Alkali solution leak detection apparatus |
| CN106778594A (en) * | 2016-12-12 | 2017-05-31 | 燕山大学 | Mental imagery EEG signal identification method based on LMD entropys feature and LVQ neutral nets |
| CN106598910A (en) * | 2016-12-28 | 2017-04-26 | 四川中烟工业有限责任公司 | EMD (Empirical Mode Decomposition) end effect inhibiting method and system |
| CN107228282A (en) * | 2017-07-06 | 2017-10-03 | 东北石油大学 | A kind of gas pipeline leakage localization method and device |
| CN108152363A (en) * | 2017-12-21 | 2018-06-12 | 北京工业大学 | A kind of defect of pipeline recognition methods for pressing down end intrinsic time Scale Decomposition |
Non-Patent Citations (2)
| Title |
|---|
| 孙洁娣等: "改进LMD及高阶模糊度函数的管道泄漏定位", 《仪器仪表学报》 * |
| 董国新: "基于ELMD多尺度模糊熵和概率神经网络的暂态电能质量识别", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109630908A (en) * | 2019-01-23 | 2019-04-16 | 常州大学 | A kind of pipeline leakage positioning method of multiple noise reduction |
| CN110146292A (en) * | 2019-06-04 | 2019-08-20 | 昆明理工大学 | A kind of rolling bearing fault testing method that the overall local mean value based on the reconstruct of integrated noise is decomposed |
| CN110146292B (en) * | 2019-06-04 | 2021-08-31 | 昆明理工大学 | Rolling bearing fault detection method based on total local mean decomposition of integrated noise reconstruction |
| CN110263832A (en) * | 2019-06-11 | 2019-09-20 | 哈尔滨工程大学 | A kind of AUV navigation system method for diagnosing faults based on multiscale analysis |
| CN110263832B (en) * | 2019-06-11 | 2023-01-06 | 哈尔滨工程大学 | AUV navigation system fault diagnosis method based on multi-scale analysis |
| WO2021012986A1 (en) * | 2019-07-22 | 2021-01-28 | 常州大学 | Pipeline multi-point leakage positioning method based on improved vmd |
| CN110730037A (en) * | 2019-10-21 | 2020-01-24 | 苏州大学 | Optical signal-to-noise ratio monitoring method of coherent optical communication system based on momentum gradient descent method |
| CN110730037B (en) * | 2019-10-21 | 2021-02-26 | 苏州大学 | Optical signal-to-noise ratio monitoring method of coherent optical communication system based on momentum gradient descent method |
| CN111397700B (en) * | 2020-03-02 | 2021-12-10 | 西北工业大学 | Wall-mounted fault detection method of Coriolis mass flow meter |
| CN111397700A (en) * | 2020-03-02 | 2020-07-10 | 西北工业大学 | Wall-mounted fault detection method of Coriolis mass flow meter |
| CN111444805A (en) * | 2020-03-19 | 2020-07-24 | 哈尔滨工程大学 | Improved multi-scale wavelet entropy digital signal modulation identification method |
| CN111444805B (en) * | 2020-03-19 | 2023-03-17 | 哈尔滨工程大学 | Improved multi-scale wavelet entropy digital signal modulation identification method |
| CN111735583A (en) * | 2020-06-24 | 2020-10-02 | 东北石油大学 | Pipeline working condition identification method based on LCD-EE pipeline sound wave signal characteristic extraction |
| CN113722977A (en) * | 2021-05-26 | 2021-11-30 | 上海华电奉贤热电有限公司 | Gas turbine rotor fault early warning method based on hybrid prediction |
| CN113722977B (en) * | 2021-05-26 | 2023-08-18 | 上海华电奉贤热电有限公司 | Gas turbine rotor fault early warning method based on hybrid prediction |
| CN114234061A (en) * | 2021-12-20 | 2022-03-25 | 北京工业大学 | Neural network-based intelligent judgment method for water leakage sound of pressurized operation water supply pipeline |
| CN114234061B (en) * | 2021-12-20 | 2024-06-21 | 北京工业大学 | Intelligent discrimination method for water leakage sound of pressurized operation water supply pipeline based on neural network |
| CN114576568A (en) * | 2022-02-25 | 2022-06-03 | 辽宁石油化工大学 | Pipeline leakage detection method and device based on infrasonic waves |
| CN114576568B (en) * | 2022-02-25 | 2023-08-29 | 辽宁石油化工大学 | Pipeline leakage detection method and device based on infrasonic wave |
| CN115111537B (en) * | 2022-08-24 | 2022-11-18 | 北京云庐科技有限公司 | Method, device and medium for determining the position of a leak in a gas pipeline network |
| CN115111537A (en) * | 2022-08-24 | 2022-09-27 | 北京云庐科技有限公司 | Method, device and medium for determining the position of a leak in a gas pipeline network |
| CN115204243A (en) * | 2022-09-15 | 2022-10-18 | 西南交通大学 | LMD endpoint effect improvement method based on similar triangular waveform matching continuation |
| CN115204243B (en) * | 2022-09-15 | 2023-02-07 | 西南交通大学 | LMD endpoint effect improvement method based on similar triangular waveform matching continuation |
| CN116415119A (en) * | 2023-04-26 | 2023-07-11 | 山东大学 | Entropy aliasing and feature enhancement-based gas abnormal signal detection method and system |
| CN116415119B (en) * | 2023-04-26 | 2024-06-28 | 山东大学 | Entropy aliasing and feature enhancement-based gas abnormal signal detection method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109084186B (en) | 2020-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109084186A (en) | Pipeline leakage signal recognition methods based on improved ELMD multi-scale entropy | |
| Tsou et al. | Structural damage detection and identification using neural networks | |
| CN111664365B (en) | Oil and gas pipeline leakage detection method based on improved VMD and 1DCNN | |
| CN108664690A (en) | Long-life electron device reliability lifetime estimation method under more stress based on depth belief network | |
| Zhang et al. | Fault identification based on PD ultrasonic signal using RNN, DNN and CNN | |
| CN101226743A (en) | Method for recognizing speaker based on conversion of neutral and affection sound-groove model | |
| CN103245907A (en) | Artificial circuit fault diagnosis pattern sorting algorithm | |
| Chin et al. | Audio event detection based on layered symbolic sequence representations | |
| CN103761965B (en) | A kind of sorting technique of instrument signal | |
| CN115758212A (en) | Mechanical equipment fault diagnosis method based on parallel network and transfer learning | |
| CN110472689B (en) | Sucker-rod pump pumping well moving liquid level soft measurement method based on integrated Gaussian process regression | |
| Kessler et al. | Application of a rectified linear unit (ReLU) based artificial neural network to cetane number predictions | |
| CN103955714A (en) | Navy detection model construction method and system and navy detection method | |
| CN113759323B (en) | Signal sorting method and device based on improved K-Means joint convolution self-encoder | |
| CN109886433A (en) | The method of intelligent recognition city gas pipeline defect | |
| CN103456302A (en) | Emotion speaker recognition method based on emotion GMM model weight synthesis | |
| CN103776901B (en) | Based on the sticky cartridge clip Rotating fields ageing state recognition methods of vibratory response information | |
| CN105094118A (en) | Airplane engine air compressor stall detection method | |
| CN116629431A (en) | Photovoltaic power generation amount prediction method and device based on variation modal decomposition and ensemble learning | |
| CN113268833A (en) | Migration fault diagnosis method based on deep joint distribution alignment | |
| CN115424620A (en) | Voiceprint recognition backdoor sample generation method based on self-adaptive trigger | |
| CN103440275A (en) | Prim-based K-means clustering method | |
| CN102496366B (en) | Speaker identification method irrelevant with text | |
| CN110796047B (en) | Self-adaptive sparse time-frequency analysis method based on machine learning | |
| Cai et al. | EMD and GNN-AdaBoost fault diagnosis for urban rail train rolling bearings |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |