CN109284777A - Recognition methods is leaked based on signal time-frequency characteristics and the water supply line of support vector machines - Google Patents
Recognition methods is leaked based on signal time-frequency characteristics and the water supply line of support vector machines Download PDFInfo
- Publication number
- CN109284777A CN109284777A CN201810988501.3A CN201810988501A CN109284777A CN 109284777 A CN109284777 A CN 109284777A CN 201810988501 A CN201810988501 A CN 201810988501A CN 109284777 A CN109284777 A CN 109284777A
- Authority
- CN
- China
- Prior art keywords
- signal
- support vector
- vector machines
- leakage
- principal component
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Examining Or Testing Airtightness (AREA)
Abstract
Recognition methods is leaked based on signal time-frequency characteristics and the water supply line of support vector machines the invention discloses a kind of, belongs to leak localization technical field.The described method includes: the signal that input is detected;Feature extraction is carried out to the signal of input;Extracted feature set is inputted to the support vector machines optimized, feature is identified using support vector machines;Support vector machines according to the signal characteristic of input export recognition result, determine signal be leakage signal be also non-leakage signal.The features such as this method is relatively concentrated using water leakage randomness, frequency spectrum, proposes three time-frequency characteristics based on signal intrinsic mode function, approximate entropy, principal component.It is inputted using these latent structure eigenmatrixes as support vector machines, support vector machines is identified to signal as classifier and is exported recognition result, to solve the problems such as existing modeling degree-of-difficulty factor of existing Discussion on Pipe Leakage Detection Technology is big, False Rate is high.
Description
Technical field
The present invention relates to leak localization technical field, more particularly to one kind based on signal time-frequency characteristics and support to
The water supply line of amount machine leaks recognition methods.
Background technique
Water is the mankind and the material base that all living things are depended on for existence, and be that human social development is indispensable provides naturally
Source." world water resources development Report in 2018 " display of the United Nations's publication, due to population growth, economic development and consumption pattern
Factors, the demands of global water resources such as transformation are increased with annual 1% speed, and this speed 20 years will also be big in future
Width is accelerated.With population it is continuous increase and pollution it is continuous worsening, the shortage of water resource is increasingly sharpened.It is tight in global water resources
In the case where weight shortage, the waste problem of water resource is extremely serious.Wherein, water resource waste ten caused by water supply line leakage
Divide serious.One of the World Bank is studies have shown that whole world water loss amount as caused by the leakage of water supply line reaches annual
486 billion cubic meters, corresponding economic loss are about annual 14600000000 dollars.Therefore, it is fixed to study efficient water supply line leak detection
Position technology is of great significance for water conservation and promoting economic development.
In order to detect that the leakage situation of groundwater supply pipeline, academia and industry have carried out numerous studies work
Make, develops many effective detection methods.The detection method occurred earliest is that sound listens method.This method is testing staff by listening
Sound detection device judges leak region according to the size of leakage sound and sound quality feature.This method is although easy to operate, but according to
Rely the experience of testing staff, and since water supply network is widely distributed, heavy workload, reliability are low in this way for institute.It wears on ground
Saturating radar can determine the leak position in pipeline by detection interstices of soil caused by leak, however, due to differently
Area's geologic structure is relative complex, and practicability is poor and expensive in this way for institute.Changed according to water supply line internal pressure
Scholars propose pressure gradient method, negative pressure wave method, flow equilibrium method in succession.These methods are more quick to in-pipe flow pressure value
Sense, still, due to water supply network water flow continued jitters, wrong report is easy to produce when fluctuating range is larger.Research finds leakage letter
Number frequency spectrum is more concentrated, and the vibration frequency of pipeline is only related with leakage situation.Passed through using this characteristic to the pressure on pipeline
The signal that electric acceleration transducer obtains carries out spectrum analysis, to carry out leak detection.But when presence and leakage signal frequency spectrum
When similar ambient noise, this method is easy to appear erroneous judgement.Researcher combines leakage acoustical signal linear predictive coding bispectrum system
Number (LPCC) and hidden Markov model (Hidden Markov Models, HMM) are further improved to leakage signal and ring
The identification capability of border interference signal.However limited by algorithm itself, with the increase of system operation time, the mistake of this method
Probability can rise.For large-scale water supply pipe net system, scholars attempt to model on pipe network using Realtime Streaming Transport, will manage
Online measurement data is compared with the predicted value of discharge model, but this method models degree-of-difficulty factor in practical applications
Height, data operation quantity are big.
Summary of the invention
In order to solve problem above, the invention proposes a kind of water supplying pipe based on signal time-frequency characteristics and support vector machines
The features such as road leaks recognition methods, and this method is relatively concentrated using water leakage randomness, frequency spectrum, according to leakage signal and non-leakage
The otherness of signal time-frequency characteristic is based on signal intrinsic mode function, approximate entropy, principal component and proposes three time-frequency characteristics.It utilizes
These latent structure eigenmatrixes are inputted as support vector machines, are identified simultaneously using support vector machines as classifier to signal
Recognition result is exported, is asked to solve that the existing modeling degree-of-difficulty factor of existing Discussion on Pipe Leakage Detection Technology is big, False Rate is high etc.
Topic.
It is according to the present invention to leak recognition methods, the side based on signal time-frequency characteristics and the water supply line of support vector machines
Method is based on signal intrinsic mode function, approximate entropy, principal component according to the otherness of leakage signal and non-leakage signal time-frequency characteristic
Three time-frequency characteristics are proposed, and are inputted using these latent structure eigenmatrixes as support vector machines, by support vector machines
Recognition result is identified and exported to signal as classifier, determine signal be leakage signal be also non-leakage signal.
Further, which comprises
S1: detected signal is inputted;
S2: feature extraction is carried out to the signal of input;
S3: inputting the support vector machines that has optimized for extracted feature set, using support vector machines to feature into
Row identification;
S4: support vector machines exports recognition result according to the signal characteristic of input, determines that signal is leakage signal also right and wrong
Leakage signal.
Further, extracted feature includes following three time-frequency characteristics: the frequency domain based on intrinsic mode function in S2
Feature, the feature based on approximate entropy and the feature based on principal component analysis.
Further, extracting the frequency domain character step based on intrinsic mode function includes:
It is handled using signal of the Empirical mode decomposition to input, decomposes the multiple solid of the signal for obtaining the input
There is mode function;
Obtained intrinsic mode function is handled, the intrinsic mode function power spectrum of signal is obtained;
Calculate the mean value of intrinsic mode function power spectrum, the time-frequency characteristics as leakage signal.
Further, extracting the characterization step based on approximate entropy includes:
According to N number of sample u (1) of acquisition pipe signal, u (2) ..., u (N) construct the sequence that two length are m, x
(i)=[u (i), u (i+1) ..., u (i+m-1)], x (j)=[u (j), u (j+1) ..., u (j+m-1)], wherein i, j≤N-m+
1, the distance between sequence of calculation x (i) and x (j),
D [x (i), x (j)]=maxK=1,2 ..., m[|u(i+m-1)-u(j+k-1)|]
A threshold value r is given, the number of d [x (i), x (j)]≤r is counted for each i < N-m+1, and calculate
Ratio between this number and vector number:
For all i values, askAverage value be φm(r),
M is increased by 1, above step is repeated and obtains φm+1(r), according to φm+1(r) and φm(r) value of approximate entropy is obtained are as follows:
ApEn (m, r)=φm(r)-φm+1(r), as the time-frequency characteristics of leakage signal.
Further, the value of r is 0.1 to 0.2 times of detected signal standard deviation.
Further, extracting the characterization step based on principal component analysis includes:
Acquire m group pipe signal x1, x2..., xm, every group of signal contain n sample, be expressed as xi=(x1i,x2i,…,
xni)T, then n × m rank matrix X=[x for thus constituting1x2…xm] be
Utilize l characteristic value before the covariance matrix of X, wherein 0 < l≤m, the characteristic value arrange from big to small, corresponding
Feature vector αi=(α1i,α2i,…,αmi)T(i=1,2, l) can be new in the hope of l vectors,
yi=X αi, (i=1,2 ..., l)
yiFor the principal component of X;
Utilize the principal component y of signaliConstruct the principal component signal matrix Y=[y of n × l rank1y2…yl], believed according to principal component
Number matrix and original signal matrix inner products gji=[yj,xi] structural matrix G,
G=YT(X-E[X])
Choose gj=[gj1gj2···gjm], time-frequency characteristics of the 0 < j≤l as leakage signal.
Further, the value of l is determined by contribution rate and contribution rate of accumulative total.
Further, the support vector machines optimizes in the following ways:
Signal acquisition construction signal is carried out when to different zones pipeline leakage occurs for different time sections and does not leak
Sample database randomly selects signal as training sample signal and test sample signal from sample of signal library;
Support vector machines is trained using the feature set of training sample signal, support vector machines is formed and recognizes model;
The support vector machines identification model that training is completed is tested using the feature set of test sample signal;
Support vector machines is further optimized according to test result, until the accuracy rate of test output is met the requirements;
Form the supporting vector machine model for pipe leakage identification.
Beneficial effects of the present invention:
(1) the present invention is based on signal intrinsic mode function, approximate entropy, principal components to propose three time-frequency characteristics.Utilize this
A little latent structure eigenmatrixes are inputted as support vector machines, so that detection effect of the invention is more comprehensively, avoid single consideration
It is larger that probability of miscarriage of justice is led to the problem of when a certain feature, effectively improves leak detection accuracy;
(2) present invention is using empirical mode decomposition (empirical mode decomposition, EMD) to pipe signal
Do time frequency analysis, sophisticated signal be decomposed into limited intrinsic mode function (intrinsic mode function, IMF) it
The form of sum carries out multiscale analysis using power spectral density of multiple IMF to leakage signal to realize, realizes high-precision
Leak positioning;
(3) principal component.
The present invention utilizes signal principal component component construction principal component matrix, then utilizes principal component matrix and original signal square
The inner product g of battle arrayji=[yj,xi] eigenmatrix of the construction based on principal component, to realize multiple leak detections that can be used for
Temporal signatures based on principal component.
Detailed description of the invention
Fig. 1 shows the water supply line leak detection side according to the present invention based on signal time-frequency characteristics and support vector machines
Method flow chart;
Fig. 2 a shows power spectrum signal when pipeline operates normally;
Fig. 2 b shows power spectrum signal when pipeline leaks;
Fig. 3 shows the intrinsic mode function of pipe leakage signal;
Fig. 4 shows the approximate entropy of pipeline leakage signal Yu non-leakage signal;
Fig. 5 shows support vector machines identification model training and principle of optimality figure;
Fig. 6 shows the recognition accuracy under support vector machines parameter combination (C, γ);
Fig. 7 shows pipe leakage and non-leak detection result.
Specific embodiment
Below in conjunction with specific attached drawing the present invention is described in detail specific embodiment.It should be noted that in following embodiments
The combination of the technical characteristic or technical characteristic of description is not construed as isolated, they can be combined with each other to reaching
To superior technique effect.In the drawings of the following embodiments, the identical label that each attached drawing occurs represent identical feature or
Person's component, can be apply to different embodiments.
As shown in Figure 1, the water supply line leak detection according to the present invention based on signal time-frequency characteristics and support vector machines
Method the following steps are included:
Step 101: inputting detected signal;
Step 102: feature extraction is carried out to the signal of input;
Step 103: extracted feature set being inputted to the support vector machines optimized, using support vector machines to spy
Sign is identified;
Step 104: support vector machines exports recognition result according to the signal characteristic of input, determines signal for leakage signal also
It is non-leakage signal.
Frequency domain character based on intrinsic mode function
A large number of studies show that signal when not leaking on pipeline has significantly with power spectrum signal when leaking
The power spectrum ingredient of difference, leakage signal is concentrated mainly in special frequency band.It therefore, can be using the power spectrum of signal as pipeline
Another characteristic is known in leakage.In order to extract the otherness of power spectrum density, we use empirical mode decomposition (empirical
Mode decomposition, EMD) time frequency analysis is done to pipe signal.Empirical mode decomposition can select the ruler of decomposed signal
Degree, is decomposed into sophisticated signal the form of the sum of limited intrinsic mode function (intrinsic mode function, IMF),
Multiscale analysis is carried out using power spectral density of multiple IMF to leakage signal to realize.
Hilbert transformation is done to a function s (t), is obtained
Then, it is based on x (t) and y (t) tectonic knot function,
The phase function of analytic signal is
By phase function to time derivation, the instantaneous frequency function of available analytic signal is
From the definition of instantaneous frequency function it is found that instantaneous frequency is the function of time, only has one in a certain particular moment
Frequency values are corresponding thereto.Instantaneous frequency will appear nonsensical negative frequency in some cases, if whole instantaneous frequencys
All it is positive frequency, then this function s (t) is known as intrinsic mode function (IMF).Therefore, intrinsic mode function must satisfy following
Two conditions:
(1) the quantity N of the extreme value of a function pointe(including minimum point and maximum point) it is equal with the quantity of zero crossing or
Difference up to 1, i.e.,
(Ns-1)≤Ne≤(Ns+1) (5)
(2) t at any timeiOn, the determining lower packet of the coenvelope line and local minimum that function local maximum determines
The mean value of winding thread is zero, i.e.,
[smax(t)+smin(t)]/2=0, ti∈[ta,tb] (6)
[t in formulaa,tb] it is a period of time length.
Qualifications (1) show that can neither occur minus maximum in s (t) does not also occur being greater than zero minimum.
Condition (2) is the waveform asymmetry for removing localised waving and generating.In general, a signal may include multiple natural mode of vibration letters
Number, the intrinsic mode function of a sophisticated signal can be extracted using Empirical mode decomposition.In order to acquire the function of leakage signal
Rate spectrum density feature, we are handled signal first with Empirical mode decomposition, obtain the natural mode of vibration letter of the signal
Number.Then, the power spectrum of different modalities function can be found out according to intrinsic mode function, and then extracts frequency domain character.
Firstly, all extreme points of original signal x (t) are connected with cubic spline curve respectively, obtain x's (t)
Upper and lower envelope is in the signal between two envelopes.Meanwhile the function for enabling two envelope mean values form is m (t).
Subtracted from original signal x (t) thereon, lower envelope line mean value m (t), obtain:
h1(t)=x (t) (7)-m (t)
It is then detected that h1(t) two conditions for whether meeting IMF, to h if being unsatisfactory for1(t) aforesaid operations are repeated, directly
To the condition for meeting IMF.By h at this time1(t) it is expressed as c1(t), then c1(t) first IMF for being signal x (t),
c1(t)=h1(t) (8)
Further, c is subtracted from original signal x (t)1(t),
r1(t)=x (t)-c1(t) (9)
By r1(t) signal new as one, acquires r using the above method1(t) first IMF, which is exactly x
(t) second IMF, is denoted as c2(t).And so on, it can gradually acquire n-th of IMFc of signal x (t)n(t) and remainder rn
(t)。
Through the above steps, original signal x (t) is decomposed into the sum of n IMF and remainder,
Under normal conditions, judge that the condition (2) of IMF is unable to satisfy, a stopping criterion is generally set, when satisfaction is stopped
Only when criterion, that is, think that condition (2) are set up.For this purpose, setting the standard deviation S between two continuous processing resultsdFor in this method
As standard deviation SdMeet
Think that condition (2) meet.Wherein, T is the observed length of signal, hk-1(t) and hkIt (t) is in IMF solution procedure
Two continuous processing results.Studies have shown that standard deviation SdThreshold value can usually take 0.2-0.3.
Fig. 2 a shows power spectrum signal when pipeline operates normally, and Fig. 2 b shows signal function when pipeline leaks
Rate spectrum.Analysis the result shows that, the frequency of pipe leakage acoustical signal is concentrated mainly near 1.6kHz.Fig. 3 is pipeline leakage signal
Empirical mode decomposition as a result, experiment Plays difference SdThreshold value be set as 0.3.The experimental results showed that the signal passes through Empirical Mode
State decomposes to have obtained 5 intrinsic mode functions.This 5 intrinsic mode functions be can use from the multiple dimensioned frequency to leakage signal
Characteristic of field is analyzed.
IMF component c after being decomposed to EMDi(n) its Discrete Fourier Transform is asked to obtain Ci(k),
To Ci(k) its mould square is sought, the power spectrum of signal is obtained
Then, the mean value of formula (13) is sought
The present invention is using the mean value of signal intrinsic mode function power spectrum as the frequency domain character of leakage signal.
Feature based on approximate entropy
Pipe leakage is a kind of local small probability event, so leakage signal is with non-leakage signal, there are one in randomness
Determine difference, it can be from the feature of the angle extraction leakage signal of analysis signal randomness.Approximate entropy (Approximate
Entropy, ApEn) it is to continue to keep the conditional probability of its similitude when similarity vector increases to m+1 dimension by m dimension, it is to work as dimension
The size of new model probability is generated when variation.The probability for generating new model is bigger, and signal is more complicated, and corresponding approximate entropy is bigger.
Therefore, the present invention chooses approximate entropy as one of leakage signal knowledge another characteristic.
Firstly, u (2) ..., u (N) construct the sequence that two length are m according to N number of sample u (1) of acquisition pipe signal
It arranges, x (i)=[u (i), u (i+1) ..., u (i+m-1)], x (j)=[u (j), u (j+1) ..., u (j+m-1)], wherein i, j≤
N-m+1.Then, the distance between sequence of calculation x (i) and x (j),
D [x (i), x (j)]=maxK=1,2 ..., m[|u(i+m-1)-u(j+k-1)|] (15)
A threshold value r is given, the number of d [X (i), X (j)]≤r is counted for each i < N-m+1, and calculate
Ratio between this number and vector number:
For all i values, askAverage value be φm(r),
M is increased by 1, (15)-(17) is repeated and obtains φm+1(r), according to φm+1(r) and φm(r) available approximate entropy
Value be,
ApEn (m, r)=φm(r)-φm+1(r) (18)
Above analysis shows approximate entropy is a nondimensional scalar, the size of value is related with m and r.In order to make approximation
The value of entropy has relatively reasonable statistical property, and rule of thumb m=2, the value of r are generally sequence criteria poor (SD, stand
Deviation) 0.1-0.2 times.Fig. 4 is the approximate entropy of signal when cast-iron pipe occurs leakage and do not leak.In experiment,
Two kinds of situations respectively extract 50 groups of data, and every group of data length is 5000, m=2, r=0.2SD.It can be seen that and let out from Fig. 4 result
The ApEn mean value of signal is apparently higher than mean value when not leaking when leakage, this shows that the random nature of leakage signal is higher than nothing and lets out
Leakage signal can be used as leakage and know another characteristic.
Feature based on signal principal component
Principal component analysis (principal components analysis, PCA) method is classical feature extracting method.
This method is the thought using dimensionality reduction, is a few generalized variable (i.e. principal component) by multiple variables transformations.Wherein, Mei Gezhu
Ingredient is all the linear combination of original variable, irrelevant between each principal component.Principal component is able to reflect the exhausted big portion of original variable
Divide information, and all information do not overlap.The present invention is using Principal Component Analysis come analysis conduit leakage and non-leakage signal
Otherness.
Acquire m group pipe signal x1, x2..., xm, every group of signal contain n sample, can be expressed as xi=(x1i,
x2i,…,xni)T, then n × m rank matrix X=[x for thus constituting1x2…xm] be
It is corresponding using l before the covariance matrix of X (0 < l≤m) a characteristic value (arranging from big to small) according to the principle of PCA
Feature vector αi=(α1i,α2i,…,αmi)T(i=1,2, l) can be new in the hope of l vectors,
yi=X αi, (i=1,2 ..., l) (20)
Claim yiFor the principal component of X, formula meets y in (20)iWith yj(i≠j;I, j=1,2 ..., l) it is uncorrelated.y1For x1,
x2..., xmAll linear combinations in variance the maximum, y2For with y1Incoherent x1, x2..., xmSide in all linear combinations
Poor the maximum.And so on, ylFor with y1, y2..., yl-1Incoherent x1, x2..., xmVariance is most in all linear combinations
Big person.In practical applications, the value of l can be determined by contribution rate and contribution rate of accumulative total.The definition of contribution rate is,
Wherein, λ=[λ1λ2 ··· λm] it is the descending arrangement of X covariance matrix characteristic value.Then, pass through setting
One threshold value makes to accumulate contribution rateReach the threshold value of setting to determine the value of l.
We construct the principal component signal matrix Y=[y of n × l rank using the principal component of signal in this method1 y2 …
yl], then, according to principal component signal matrix and original signal matrix inner products gji=[yj,xi] structural matrix G,
G=YT(X-E[X]) (22)
Further choose gj=[gj1 gj2 ··· gjm], 0 < j≤l knows another characteristic as leak.
It is detected based on the leakage signal of time-frequency characteristics and support vector machines
Signal time-frequency characteristics proposed above have the characteristics that when recognizing pipe leakage it is different, still, it is single consider certain
The probability that erroneous judgement is generated when one feature is larger.For example, if the Power Spectrum Distribution of leakage signal and non-leakage signal has obviously
When difference, the power spectrum mean value of intrinsic mode function has preferable recognition effect, but if then should when in the presence of same band interference
Method is easy to appear erroneous judgement.When the leakage rate of pipeline leakage point is smaller, approximate entropy mean value and the non-leakage signal of leakage signal
Difference can be unobvious, this is easy to cause this method to judge by accident.
In order to improve the accuracy rate of leak detection, it is special as identification that the present invention comprehensively utilizes feature combination set forth above
Sign, classifies to signal characteristic using support vector machines (SVM), judges whether pipeline leaks.SVM is one kind to unite
The data digging method based on the theories of learning is counted, the problems such as solving small sample problem, nonlinear problem and high dimensional data
In have advantage, be widely used in the fields such as data prediction, data fitting, pattern-recognition.Assuming that training set data sample is
(xi,yi), wherein 1≤i≤N, each sample xi∈Rd, d is the dimension of the input space, yi∈ { -1,1 } is class label.If
The training set can be by a hyperplane linear partition, then the hyperplane can be expressed as wx+b=0, and wherein w and b is to determine
The position of hyperplane.The sample for meeting following condition is known as supporting vector,
yi(w·xi+ b)=1 (23)
The optimization of sample is actually the Solve problems of optimal separating hyper plane,
In formula, w is the coefficient vector of Optimal Separating Hyperplane in feature space;B is the threshold value of classifying face;ξiIt is to consider that classification misses
Poor and introducing relaxation factor and ξi≥0;C is the penalty factor for error sample.Thus obtained optimal separating hyper plane
It can be expressed as,
w0·x+b0=0 (25)
Using method of Lagrange multipliers, the antithesis that the optimization hyperplane Solve problems of formula (24) are converted into it can be asked
Topic,
α=(α in formula1,…,αN) it is Lagrange multiplier, meet αiThe sample of > 0 is supporting vector.It utilizes formula (26)
Its available corresponding decision function of the optimal hyperlane acquired,
Wherein,α when for formula (33) optimal solutioniValue.
It the case where for Nonlinear separability, can be by a mapping function (being known as kernel function in SVM) by the defeated of low-dimensional
Enter space RdIt is mapped to the feature space H of higher-dimension, training sample is made to be changed into higher dimensional space by the linearly inseparable problem of low-dimensional
Linear separability problem.The dual problem of optimization problem at this time are as follows:
Wherein, K (xi,xj)=Φ (xi)·Φ(xj) it is kernel function.From formula (28) as can be seen that for Nonlinear separability
Problem needs to select suitable kernel function K () construction SVM model.It is using the corresponding decision function of formula (28),
In order to improve the accuracy rate of leak water detdction, it would be desirable to support vector machines is optimized using known signal, it is excellent
Change process is as shown in Figure 5.In view of the influence of groundwater supply pipeline local environment factor, need in different time sections to not same district
Domain pipeline occurs leakage and carries out signal acquisition construction sample of signal library when not leaking, randomly selects from sample of signal library
Signal is as training sample and test sample.Firstly, the feature set using training sample is trained SVM, formed preliminary
Recognize model.Then, it is tested using the SVM model that the feature set of test sample completes training.According to test result pair
SVM is further optimized, and until the accuracy rate of test output is met the requirements, formation is used for the SVM model of pipe leakage identification.
In, the signal that detection system host computer acquires wireless sensor network inputs SVM model, the label exported according to model
Judge whether pipeline leaks.
Formula (26)-(29) theoretical analysis shows that, determine SVM model performance factor be kernel function and penalty factor.
The kernel function of SVM is broadly divided into linear kernel, polynomial kernel, Sigmoid core and radial base core.According to the characteristics of water leakage and
The generalization ability of SVM model, the present invention have selected radial base core as kernel function.The expression formula of Radial basis kernel function are as follows:
K(xi,xj)=exp (- γ | | xi-xj||2) (30)
Therefore, for a SVM based on Radial basis kernel function, performance is determined by parameter (C, γ).In order to make SVM
Model reaches higher recognition effect, the optimization process in Fig. 5 be actually utilize training sample and test sample to parameter C and
γ optimizes adjustment.A large number of studies show that the effect that the fetching Number Sequence of C and γ can achieve in practical application, usually
For C=2-5,2-4,···,215, γ=2-15,2-14,···,25。
The present invention is based on cross-validation's " grid-search " method, SVM parameter is optimized.
Firstly, carrying out value, C=2 to parameter according to the value range of parameterx, x ∈ [- 5,15], γ=2y,y∈[-15,
5].Then, using training sample and test sample to difference 2xWith 2ySVM model under combination is tested, and exports test
Accuracy rate.Finally, choosing the parameter of the C and γ of optimal cross-validation accuracy as leak identification SVM.
Embodiment 1
PVC water supply line has been selected to carry out signal acquisition in experiment.Leakage and non-leakage situation are respectively adopted in different periods
100 groups of data are collected, every group of data length is 5000, for trained and Support Vector Machines Optimized.Meanwhile in the clear of relative silence
Period in morning respectively acquires 100 data to leakage and non-leakage situation, what the method validation present invention by the way that noise is artificially added studied
The validity of leak detection and Time Delay Estimation Algorithms.
Firstly, respectively extracting 50 groups of data from the leakage signal and non-leakage signal that different periods acquire forms one 100
The training sample of group.Then, the test sample of remaining one 100 groups of data of sample architecture is utilized.Support vector machines parameter (C,
γ) the integral number power that value is 2, also, the value range of C is C ∈ [2-5,215], the value range of γ is γ ∈ [2-15,25]。
By grid data service, using (C, the γ) under 21 × 21=441 parameter combination to model training, and test set data are used
Model performance is detected, obtained Detection accuracy is as shown in Figure 6.Fig. 6's the result shows that, the present invention propose algorithm highest
Recognition accuracy is 98%.And it is possible to find out when penalty factor >=22With kernel functional parameter γ≤20, and two parameters
Product 21≤C×γ≤27When the SVM model based on radial base core to pipeline leakage signal have preferable recognition performance.
Fig. 7 is that selection parameter combines (C, γ)=(29,2-4) when testing result, wherein pipeline occur leakage signal label
It is set as -1, the signal label under the non-leakage situation of pipeline is set as 1.As can be seen that method proposed by the present invention from recognition result
The signal under 2 groups of leakage situations is only determined as non-leakage signal, the equal correct judgment of remaining situation.Table 1 is quiet section acquisition
Leakage signal proposes the case where algorithm carries out leakage identification using the present invention after Gaussian noise and impulsive noise manually is added.Knot
Fruit shows that the water supply line leakage detection method proposed by the present invention based on signal multiple features and support vector machines can be examined effectively
The leakage situation in test tube road.
Leakage signal discrimination in 1. Gaussian noise of table and impulse noise environment
Although the embodiment of the present invention is had been presented for herein, it will be appreciated by those of skill in the art that not taking off
In the case where from spirit of that invention, the embodiments herein can be changed.Above-described embodiment is only exemplary, should not be with
Restriction of the embodiments herein as interest field of the present invention.
Claims (9)
1. the water supply line leakage detection method based on signal time-frequency characteristics and support vector machines, which is characterized in that the method
Signal intrinsic mode function, approximate entropy, principal component is based on according to leakage signal and the otherness of non-leakage signal time-frequency characteristic to mention
Three time-frequency characteristics are gone out, and have been inputted using these latent structure eigenmatrixes as support vector machines, support vector machines has been made
Recognition result is identified and exported to signal for classifier, determine signal be leakage signal be also non-leakage signal.
2. the method according to claim 1, wherein the described method includes:
S1: detected signal is inputted;
S2: feature extraction is carried out to the signal of input;
S3: extracted feature set is inputted to the support vector machines optimized, feature is known using support vector machines
Not;
S4: support vector machines according to the signal characteristic of input export recognition result, determine signal be leakage signal be also non-leakage
Signal.
3. according to the method described in claim 2, it is characterized in that, extracted feature includes following three time-frequency spies in S2
Sign: the frequency domain character based on intrinsic mode function, the feature based on approximate entropy and the feature based on principal component analysis.
4. according to the method described in claim 3, it is characterized in that, extracting the frequency domain character step packet based on intrinsic mode function
It includes:
It is handled using signal of the Empirical mode decomposition to input, decomposes and obtain multiple natural modes of the signal of the input
State function;
Obtained intrinsic mode function is handled, the intrinsic mode function power spectrum of signal is obtained;
Calculate the mean value of intrinsic mode function power spectrum, the time-frequency characteristics as leakage signal.
5. according to the method described in claim 3, it is characterized in that, characterization step of the extraction based on approximate entropy includes:
According to acquisition pipe signal N number of sample u (1), u (2) ..., u (N) construct two length be m sequence, x (i)=
[u (i), u (i+1) ..., u (i+m-1)], x (j)=[u (j), u (j+1) ..., u (j+m-1)], wherein i, j≤N-m+1, calculate
The distance between sequence x (i) and x (j),
D [x (i), x (j)]=maxK=1,2 ..., m[|u(i+m-1)-u(j+k-1)|]
A threshold value r is given, the number of d [x (i), x (j)]≤r is counted for each i < N-m+1, and calculate this
Ratio between number and vector number:
For all i values, askAverage value be φm(r),
M is increased by 1, above step is repeated and obtains φm+1(r), according to φm+1(r) and φm(r) value of approximate entropy is obtained are as follows:
ApEn (m, r)=φm(r)-φm+1(r), as the time-frequency characteristics of leakage signal.
6. according to the method described in claim 5, it is characterized in that, the value of r is the 0.1 to 0.2 of detected signal standard deviation
Times.
7. according to the method described in claim 3, it is characterized in that, characterization step of the extraction based on principal component analysis includes:
Acquire m group pipe signal x1, x2..., xm, every group of signal contain n sample, be expressed as xi=(x1i,x2i,…,xni)T, then
Thus n × m rank matrix X=[the x constituted1x2…xm] be
Utilize l characteristic value before the covariance matrix of X, wherein 0 < l≤m, the characteristic value arrange from big to small, corresponding spy
Levy vector αi=(α1i,α2i,…,αmi)T(i=1,2 ..., l) can be new in the hope of l vectors,
yi=X αi, (i=1,2 ..., l)
yiFor the principal component of X;
Utilize the principal component y of signaliConstruct the principal component signal matrix Y=[y of n × l rank1y2…yl], according to principal component signal square
Battle array and original signal matrix inner products gji=[yj,xi] structural matrix G,
G=YT(X-E[X])
Choose gj=[gj1gj2…gjm], time-frequency characteristics of the 0 < j≤l as leakage signal.
8. the method according to the description of claim 7 is characterized in that the value of l is determined by contribution rate and contribution rate of accumulative total.
9. according to the method described in claim 2, it is characterized in that, the support vector machines optimizes in the following ways:
Signal acquisition construction sample of signal is carried out when to different zones pipeline leakage occurs for different time sections and does not leak
Library randomly selects signal as training sample and test sample from sample of signal library;
Support vector machines is trained using the feature set of training sample, support vector machines is formed and recognizes model;
The support vector machines identification model that training is completed is tested using the feature set of test sample;
Support vector machines is further optimized according to test result, until the accuracy rate of test output is met the requirements;
Form the supporting vector machine model for pipe leakage identification.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810988501.3A CN109284777B (en) | 2018-08-28 | 2018-08-28 | Water supply pipeline leakage identification method based on signal time-frequency characteristics and support vector machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810988501.3A CN109284777B (en) | 2018-08-28 | 2018-08-28 | Water supply pipeline leakage identification method based on signal time-frequency characteristics and support vector machine |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109284777A true CN109284777A (en) | 2019-01-29 |
CN109284777B CN109284777B (en) | 2021-09-28 |
Family
ID=65184050
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810988501.3A Active CN109284777B (en) | 2018-08-28 | 2018-08-28 | Water supply pipeline leakage identification method based on signal time-frequency characteristics and support vector machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109284777B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472506A (en) * | 2019-07-11 | 2019-11-19 | 广东工业大学 | A kind of gesture identification method based on support vector machines and Neural Network Optimization |
CN110657355A (en) * | 2019-08-28 | 2020-01-07 | 北京无线电计量测试研究所 | Method for detecting leakage of thermal pipeline |
CN110792928A (en) * | 2019-09-24 | 2020-02-14 | 中国石油化工股份有限公司 | Pipeline leakage diagnosis combined algorithm based on big data |
CN111207926A (en) * | 2019-12-27 | 2020-05-29 | 三明学院 | Fault diagnosis method based on rolling bearing, electronic device and storage medium |
CN112613484A (en) * | 2021-01-06 | 2021-04-06 | 西南石油大学 | Gas transmission pipeline leakage identification method based on singular spectrum analysis and support vector machine |
CN112634945A (en) * | 2020-12-15 | 2021-04-09 | 浙江和达科技股份有限公司 | Intelligent water leakage sound identification method based on cloud platform |
CN112763155A (en) * | 2021-02-02 | 2021-05-07 | 南京信息工程大学 | Pressure container gas leakage detection method |
CN112926384A (en) * | 2021-01-15 | 2021-06-08 | 厦门大学 | Automatic modal identification method based on power spectrum transfer ratio and support vector machine |
CN113486950A (en) * | 2021-07-05 | 2021-10-08 | 华能国际电力股份有限公司上安电厂 | Intelligent pipe network water leakage detection method and system |
CN114034966A (en) * | 2021-10-13 | 2022-02-11 | 国电南瑞科技股份有限公司 | Power transmission line fault identification method and device based on support vector machine |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102323013A (en) * | 2011-05-27 | 2012-01-18 | 长沙理工大学 | Valve leakage detection device |
CN105423772A (en) * | 2015-11-18 | 2016-03-23 | 东北电力大学 | Power station air cooling system adopting combined refrigeration with shaft seal steam leakage of steam turbine and continuous blow-down waste heat of boiler as well as method for predicting heat-transfer coefficient of air-cooling condenser |
CN106017879A (en) * | 2016-05-18 | 2016-10-12 | 河北工业大学 | Universal circuit breaker mechanical fault diagnosis method based on feature fusion of vibration and sound signals |
US20170076563A1 (en) * | 2015-09-15 | 2017-03-16 | General Electric Company | Systems and methods to provide pipeline damage alerts |
CN107480705A (en) * | 2017-07-24 | 2017-12-15 | 厦门快商通科技股份有限公司 | A kind of running water pipe leakage detection method |
-
2018
- 2018-08-28 CN CN201810988501.3A patent/CN109284777B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102323013A (en) * | 2011-05-27 | 2012-01-18 | 长沙理工大学 | Valve leakage detection device |
US20170076563A1 (en) * | 2015-09-15 | 2017-03-16 | General Electric Company | Systems and methods to provide pipeline damage alerts |
CN105423772A (en) * | 2015-11-18 | 2016-03-23 | 东北电力大学 | Power station air cooling system adopting combined refrigeration with shaft seal steam leakage of steam turbine and continuous blow-down waste heat of boiler as well as method for predicting heat-transfer coefficient of air-cooling condenser |
CN106017879A (en) * | 2016-05-18 | 2016-10-12 | 河北工业大学 | Universal circuit breaker mechanical fault diagnosis method based on feature fusion of vibration and sound signals |
CN107480705A (en) * | 2017-07-24 | 2017-12-15 | 厦门快商通科技股份有限公司 | A kind of running water pipe leakage detection method |
Non-Patent Citations (1)
Title |
---|
王昕: "基于 EEMD 和支持向量机的天然气管道泄漏诊断", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472506A (en) * | 2019-07-11 | 2019-11-19 | 广东工业大学 | A kind of gesture identification method based on support vector machines and Neural Network Optimization |
CN110657355B (en) * | 2019-08-28 | 2021-06-01 | 北京无线电计量测试研究所 | Method for detecting leakage of thermal pipeline |
CN110657355A (en) * | 2019-08-28 | 2020-01-07 | 北京无线电计量测试研究所 | Method for detecting leakage of thermal pipeline |
CN110792928A (en) * | 2019-09-24 | 2020-02-14 | 中国石油化工股份有限公司 | Pipeline leakage diagnosis combined algorithm based on big data |
CN110792928B (en) * | 2019-09-24 | 2021-08-10 | 中国石油化工股份有限公司 | Pipeline leakage diagnosis method based on big data combined algorithm |
CN111207926A (en) * | 2019-12-27 | 2020-05-29 | 三明学院 | Fault diagnosis method based on rolling bearing, electronic device and storage medium |
CN111207926B (en) * | 2019-12-27 | 2022-02-01 | 三明学院 | Fault diagnosis method based on rolling bearing, electronic device and storage medium |
CN112634945A (en) * | 2020-12-15 | 2021-04-09 | 浙江和达科技股份有限公司 | Intelligent water leakage sound identification method based on cloud platform |
CN112613484A (en) * | 2021-01-06 | 2021-04-06 | 西南石油大学 | Gas transmission pipeline leakage identification method based on singular spectrum analysis and support vector machine |
CN112926384A (en) * | 2021-01-15 | 2021-06-08 | 厦门大学 | Automatic modal identification method based on power spectrum transfer ratio and support vector machine |
CN112763155A (en) * | 2021-02-02 | 2021-05-07 | 南京信息工程大学 | Pressure container gas leakage detection method |
CN113486950A (en) * | 2021-07-05 | 2021-10-08 | 华能国际电力股份有限公司上安电厂 | Intelligent pipe network water leakage detection method and system |
CN113486950B (en) * | 2021-07-05 | 2023-06-16 | 华能国际电力股份有限公司上安电厂 | Intelligent pipe network water leakage detection method and system |
CN114034966A (en) * | 2021-10-13 | 2022-02-11 | 国电南瑞科技股份有限公司 | Power transmission line fault identification method and device based on support vector machine |
Also Published As
Publication number | Publication date |
---|---|
CN109284777B (en) | 2021-09-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109284777A (en) | Recognition methods is leaked based on signal time-frequency characteristics and the water supply line of support vector machines | |
Meng et al. | An enhancement denoising autoencoder for rolling bearing fault diagnosis | |
Curilem et al. | Pattern recognition applied to seismic signals of the Llaima volcano (Chile): An analysis of the events' features | |
Scitovski et al. | A fast partitioning algorithm and its application to earthquake investigation | |
Sabo et al. | One-dimensional center-based l 1-clustering method | |
Ning et al. | A framework combining acoustic features extraction method and random forest algorithm for gas pipeline leak detection and classification | |
Zhang | Music feature extraction and classification algorithm based on deep learning | |
CN105609116B (en) | A kind of automatic identifying method in speech emotional dimension region | |
Tran et al. | Single and combined fault diagnosis of reciprocating compressor valves using a hybrid deep belief network | |
Jiménez et al. | Machine learning and neural network for maintenance management | |
Singh et al. | Speaker specific feature based clustering and its applications in language independent forensic speaker recognition | |
Wang et al. | Design and optimization of 1d-cnn for spectrum recognition of underwater targets | |
Luo et al. | Virtual autoencoder based recommendation system for individualizing head-related transfer functions | |
Schirmer et al. | Low-frequency energy disaggregation based on active and reactive power signatures | |
CN114582325A (en) | Audio detection method and device, computer equipment and storage medium | |
Gan | Signal processing and image processing for acoustical imaging | |
Song et al. | A novel noise reduction technique for underwater acoustic signals based on dual‐path recurrent neural network | |
Zhao et al. | Leak detection method of liquid-filled pipeline based on VMD and SVM | |
Xie et al. | Multi-label classification of frog species via deep learning | |
CN113723525B (en) | Product recommendation method, device, equipment and storage medium based on genetic algorithm | |
Li et al. | Engine fault diagnosis based on a morphological neural network using a morphological filter as a preprocessor | |
Look et al. | Building Robust Classifiers with Generative Adversarial Networks for Detecting Cavitation in Hydraulic Turbines. | |
Saravanabalaji et al. | Acoustic signal based water leakage detection system using hybrid machine learning model | |
Rohan et al. | Emotion Recognition Through Speech Signal Using Python | |
Hayashi et al. | Fast music information retrieval with indirect matching |
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 |