CN105842588A - Method of correcting supersonic wave partial discharge detection and system thereof - Google Patents

Abstract

The invention provides a method of correcting supersonic wave partial discharge detection. The method comprises the following steps of extracting a plurality of partial discharge signal linearity frequency spectrums from historical data and then processing to acquire a coefficient matrix and a principal component characteristic of each partial discharge signal, and constructing a classification model of the partial discharge signals and each type of a linearity frequency spectrum mean value; acquiring a linearity frequency spectrum mean value of pure noise signals, and according to the coefficient matrix, acquiring a classification model of pure noise supersonic wave signals; adding each type of the linearity frequency spectrum mean value of the partial discharge signals to the linearity frequency spectrum mean value of the pure noise signals and correcting the classification model of the partial discharge signals; acquiring a signal to be detected, according to the coefficient matrix, acquiring a principal component characteristic of the signal to be detected and introducing into the classification model corrected by the pure noise and the partial discharge signals, and screening signal output with a minimum Euclidean distance in the corresponding classification model. By using the method, an artificial factor interference is avoided, an introduced characteristic dimension is low and training data is less so that purposes of labor saving and time saving, and an objective detection result are reached.

Description

A kind of method and system revising ultrasound wave Partial Discharge Detection

Technical field

The present invention relates to local discharge signal detection technique field, particularly relate to a kind of ultrasound wave local of revising and put The method and system of electro-detection.

Background technology

In order to ensure the safe and stable operation of equipment, insulating properties detection and fault diagnosis need to be carried out, and local Electric discharge is the main reason causing apparatus insulated accident to occur, and therefore uses ultrasonic detection method to equipment Carry out shelf depreciation to detect in real time, can grasp equipment real-time operating conditions comprehensively, and can to afterwards one section time State of insulation in length is predicted, and takes suitably to overhaul and maintenance strategy according to its insulation status simultaneously, This has the most positive meaning to the reliable and stable operation ensureing electric power system.

At present, detection equipment based on ultrasound examination shelf depreciation technology has put goods on the market use, conventional Equipment has two kinds: one is by piezoelectric principle, after ultrasonic signal is converted into current signal, through inside Processor is translated into audio signal, thus the presence or absence of local discharge signal can be supervised by high-fidelity headphone The exception listening audio signal judges；Another kind is to realize detection by artificial setting threshold value and warning function, Size by dB value display Processing of Partial Discharge Ultrasonic Signals.

Some following problems all can be run into: at the first equipment during above two equipment is the most actually detected During detection, due to the difference of everyone auditory physiology characteristic, in testing staff's earphone to listening to Audio signal have different judgements, testing result is closely bound up with the subjective initiative of testing staff, hold Easily cause erroneous judgement；In the second device detection procedure, owing to breakdown judge relies on experience in the majority, system is examined Survey reliability the highest.

Therefore, in order to detect ultrasound wave local discharge signal more accurately, some pattern recognitions based on statistics Technology is applied in real system, but these technology there is also following not enough in the application: (1) feature Dimension is more, needs substantial amounts of training data, and local discharge signal is generally difficult to collect；(2) instruction is worked as When white silk environment does not mates with practical service environment, performance can be remarkably decreased.

Summary of the invention

Embodiment of the present invention technical problem to be solved is, it is provided that a kind of ultrasound wave shelf depreciation of revising is examined The method and system surveyed, it is possible to avoiding artifact to disturb, and introduced feature dimension is low, training data is few, Thus reach the time saving and energy saving and objective purpose of testing result.

In order to solve above-mentioned technical problem, embodiments provide a kind of ultrasound wave shelf depreciation of revising and examine The method surveyed, described method includes:

A, from the historical data of equipment high s/n ratio ultrasound wave local discharge signal, extract multiple shelf depreciation Signal, and pass through principal component analysis after multiple local discharge signals of described extraction are converted into linear spectral Method processes, and obtains P and ties up main constituent coefficient matrix, and ties up main constituent coefficient matrix according to the described P obtained, The P obtaining each local discharge signal ties up main constituent feature, further by the described each shelf depreciation obtained The P dimension main constituent feature of signal carries out Unsupervised clustering, obtains the disaggregated model of local discharge signal and right Answer the linear spectral average of each class；Wherein, P is natural number；

B, obtain pure noise ultrasonic signal, and the described pure noise ultrasonic signal got is converted to line Property frequency spectrum after, calculate the linear spectral average of pure noise ultrasonic signal, and tie up according to the described P obtained Main constituent coefficient matrix, is converted to described pure noise by the linear spectral average of described pure noise ultrasonic signal After the P dimension main constituent feature of ultrasonic signal, obtain the disaggregated model of pure noise ultrasonic signal；

C, the linear spectral average of each class in described local discharge signal disaggregated model pure is made an uproar with described respectively The linear spectral average of sound ultrasonic signal disaggregated model is added, each class after being added linear Spectrum mean, and tie up main constituent coefficient matrix according to the described P obtained, every by after the described addition obtained After the linear spectral average of one class is converted to the P dimension main constituent feature of correspondence, the disaggregated model obtained is as office The revised disaggregated model of portion's discharge signal；

D, acquisition suspect signal, and the described suspect signal got is converted to linear spectral, and according to institute State the P obtained and tie up main constituent coefficient matrix, the linear spectral of described suspect signal is converted to described letter to be checked Number P tie up main constituent feature；

E, the P of described suspect signal is tieed up main constituent feature introduce respectively described pure noise ultrasonic signal point In class model and the revised disaggregated model of described local discharge signal, calculate described suspect signal and arrive respectively The disaggregated model of described pure noise ultrasonic signal and the Europe of the revised disaggregated model of described local discharge signal Family name's distance, and according to described two Euclidean distances obtained, filter out minima in said two Euclidean distance Corresponding disaggregated model, further using signal corresponding in institute's sifting sort model as described suspect signal Testing result output.

Wherein, described step a specifically includes:

From the historical data of equipment high s/n ratio ultrasound wave local discharge signal, extract multiple shelf depreciation letter Number, and after multiple local discharge signals of described extraction are all carried out framing in units of certain time length, to institute State each local discharge signal after framing and all carry out Fourier transformation, obtain the line of each local discharge signal Property frequency spectrum；

By PCA, the linear spectral of arbitrary local discharge signal is carried out dimension-reduction treatment, obtain P Dimension main constituent coefficient matrix, and tie up main constituent coefficient matrix, every to described extraction according to the described P obtained One local discharge signal all carries out linear transformation, and the P obtaining each local discharge signal ties up main constituent feature；

Use the k-mean algorithm P simultaneously to the described each local discharge signal obtained to tie up main constituent feature to enter Row Unsupervised clustering, calculates multiple cluster centre and preserves the disaggregated model as local discharge signal；

The principle averaged after being added according to same class linear spectral, the classification to described local discharge signal In model, the linear spectral contained by each class calculates, and obtains in the disaggregated model of described local discharge signal The linear spectral average of each class.

Wherein, described step b specifically includes:

Before obtaining described suspect signal, obtain the pure noise ultrasound wave in the range of actual environment a period of time Signal, and after the described pure noise ultrasonic signal got is carried out framing in units of certain time length, right Pure noise ultrasonic signal after described framing all carries out Fourier transformation, obtains described pure noise ultrasound wave letter Number linear spectral；

Average, as described pure after the linear spectral of the described pure noise ultrasonic signal obtained is added The linear spectral average of noise ultrasonic signal；

Main constituent coefficient matrix is tieed up, by the linear frequency of described pure noise ultrasonic signal according to the described P obtained After spectrum average is converted to the P dimension main constituent feature of described pure noise ultrasonic signal, is only comprised one and gathered The disaggregated model at class center also preserves the disaggregated model as described pure noise ultrasonic signal.

Wherein, described step c specifically includes:

Determine the linear spectral average of each class in described local discharge signal disaggregated model, and pure make an uproar described The linear spectral average of sound ultrasonic signal disaggregated model is equal with the linear spectral of the described each class determined respectively Value is added, the linear spectral average of each class after being added；

Main constituent coefficient matrix is tieed up, by the line of each class after the described addition obtained according to the described P obtained Property spectrum mean be converted to correspondence P dimension main constituent feature after, only comprised the classification of a cluster centre Model also preserves as the revised disaggregated model of described local discharge signal.

Wherein, described step e specifically includes:

The P of described suspect signal is tieed up main constituent feature and introduces the classification of described pure noise ultrasonic signal respectively In model and the revised disaggregated model of described local discharge signal, calculate described suspect signal respectively to institute State disaggregated model and the Euclidean of the revised disaggregated model of described local discharge signal of pure noise ultrasonic signal Distance；

Judge that described suspect signal is the least to the Euclidean distance of the disaggregated model of described pure noise ultrasonic signal Euclidean distance in described suspect signal to the revised disaggregated model of described local discharge signal；

If it is, the disaggregated model that the disaggregated model of described screening is described pure noise ultrasonic signal, and It is that pure noise ultrasonic signal exports as testing result using described suspect signal；

If it is not, then the disaggregated model of described screening is the revised disaggregated model of described local discharge signal, And to be local discharge signal using described suspect signal export as testing result.

The embodiment of the present invention additionally provides a kind of system revising ultrasound wave Partial Discharge Detection, described system bag Include:

Local discharge signal analyzes model acquiring unit, for believing from equipment high s/n ratio ultrasound wave shelf depreciation Number historical data in, extract multiple local discharge signal, and by multiple local discharge signals of described extraction Processed by PCA after being converted into linear spectral, obtain P and tie up main constituent coefficient matrix, and root Tieing up main constituent coefficient matrix according to the described P obtained, the P obtaining each local discharge signal ties up main constituent feature, Further the P of the described each local discharge signal obtained is tieed up main constituent feature and carry out Unsupervised clustering, Disaggregated model and the linear spectral average of corresponding each class thereof to local discharge signal；Wherein, P is natural number；

Pure noise ultrasound signal analyzing model acquiring unit, is used for obtaining pure noise ultrasonic signal, and will After the described pure noise ultrasonic signal got is converted to linear spectral, calculate pure noise ultrasonic signal Linear spectral average, and according to the described P obtained tie up main constituent coefficient matrix, by ultrasonic for described pure noise After the linear spectral average of ripple signal is converted to the P dimension main constituent feature of described pure noise ultrasonic signal, Disaggregated model to pure noise ultrasonic signal；

Local discharge signal analyzes Modifying model unit, for by every in described local discharge signal disaggregated model The linear spectral average of one class linear spectral average with described pure noise ultrasonic signal disaggregated model respectively is entered Row is added, the linear spectral average of each class after being added, and ties up main constituent according to the described P obtained Coefficient matrix, the P dimension that the linear spectral average of each class after the described addition obtained is converted to correspondence is main After composition characteristics, the disaggregated model obtained is as the revised disaggregated model of local discharge signal；

Suspect signal analytic unit, is used for obtaining suspect signal, and the described suspect signal got is changed For linear spectral, and tie up main constituent coefficient matrix according to the described P obtained, linear by described suspect signal Frequency spectrum is converted to the P of described suspect signal and ties up main constituent feature；

Suspect signal recognition unit, introduces described for the P of described suspect signal ties up main constituent feature respectively In the disaggregated model of pure noise ultrasonic signal and the revised disaggregated model of described local discharge signal, calculate Go out described suspect signal and arrive the disaggregated model of described pure noise ultrasonic signal and described local discharge signal respectively The Euclidean distance of revised disaggregated model, and according to described two Euclidean distances obtained, filter out described The disaggregated model that in two Euclidean distances, minima is corresponding, further by corresponding in institute's sifting sort model Signal exports as the testing result of described suspect signal.

Wherein, described local discharge signal analysis model acquiring unit includes:

Local discharge signal linear spectral conversion module, for believing from equipment high s/n ratio ultrasound wave shelf depreciation Number historical data in, extract multiple local discharge signal, and by multiple local discharge signals of described extraction After all carrying out framing in units of certain time length, each local discharge signal after described framing is all carried out Fu In leaf transformation, obtain the linear spectral of each local discharge signal；

Principal component analysis module, for by the PCA linear spectral to arbitrary local discharge signal Carry out dimension-reduction treatment, obtain P and tie up main constituent coefficient matrix, and tie up main constituent coefficient according to the described P obtained Matrix, all carries out linear transformation to each local discharge signal of described extraction, obtains each shelf depreciation letter Number P tie up main constituent feature；

Local discharge signal Clustering Model computing module, for using k-mean algorithm to obtain described simultaneously The P dimension main constituent feature of each local discharge signal carries out Unsupervised clustering, calculates multiple cluster centre also Preserve the disaggregated model as local discharge signal；

Local discharge signal Clustering Model classification mean value computation module, for being added according to same class linear spectral After the principle averaged, linear spectral contained by each class in the disaggregated model to described local discharge signal Calculate, obtain the linear spectral average of each class in the disaggregated model of described local discharge signal.

Wherein, described pure noise ultrasound signal analyzing model acquiring unit includes:

Pure noise ultrasonic signal linear spectral conversion module, for, before obtaining described suspect signal, obtaining Take the pure noise ultrasonic signal in the range of actual environment a period of time, and the described pure noise got is surpassed After acoustic signals carries out framing in units of certain time length, equal to the pure noise ultrasonic signal after described framing Carry out Fourier transformation, obtain the linear spectral of described pure noise ultrasonic signal；

Linear spectral mean value computation module, for by the linear spectral of the described pure noise ultrasonic signal obtained Average after addition, as the linear spectral average of described pure noise ultrasonic signal；

Pure noise ultrasonic signal Clustering Model computing module, ties up main constituent system for the P obtained described in basis Matrix number, is converted to described pure noise ultrasound wave letter by the linear spectral average of described pure noise ultrasonic signal Number P dimension main constituent feature after, only comprised the disaggregated model of a cluster centre and preserved as described The disaggregated model of pure noise ultrasonic signal.

Wherein, described local discharge signal analysis Modifying model unit includes:

Linear spectral accumulator module, for determining the linear of each class in described local discharge signal disaggregated model Spectrum mean, and by the linear spectral average of described pure noise ultrasonic signal disaggregated model respectively with described really The linear spectral average of fixed each class is added, the linear spectral average of each class after being added；

Local discharge signal Clustering Model correcting module, ties up main constituent coefficient square for the P obtained described in basis Battle array, is converted to the P Wei Zhuchengfente of correspondence by the linear spectral average of each class after the described addition obtained After levying, only comprised the disaggregated model of a cluster centre and preserved as described local discharge signal correction After disaggregated model.

Wherein, described suspect signal recognition unit includes:

Oldham distance calculating module, introduces described for the P of described suspect signal ties up main constituent feature respectively In the disaggregated model of pure noise ultrasonic signal and the revised disaggregated model of described local discharge signal, calculate Go out described suspect signal and arrive the disaggregated model of described pure noise ultrasonic signal and described local discharge signal respectively The Euclidean distance of revised disaggregated model；

Judge module, for judging the described suspect signal disaggregated model to described pure noise ultrasonic signal Whether Euclidean distance is less than the described suspect signal Euclidean to the revised disaggregated model of described local discharge signal Distance；

First result output module, the disaggregated model for described screening is described pure noise ultrasonic signal Disaggregated model, and be that pure noise ultrasonic signal exports as testing result using described suspect signal；

Second result output module, after the disaggregated model of described screening is described local discharge signal correction Disaggregated model, and to be local discharge signal using described suspect signal export as testing result.

Implement the embodiment of the present invention, have the advantages that

In embodiments of the present invention, owing to using principal component analytical method can be effectively reduced training data Dimension reduces data complexity, uses being modified disaggregated model of environment noise average, Ke Yiyou simultaneously Effect improves detection method performance under all kinds of noise circumstances such that it is able to avoids artifact to disturb, and draws Entering intrinsic dimensionality low, training data is few, thus reaches the time saving and energy saving and objective purpose of testing result.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to enforcement In example or description of the prior art, the required accompanying drawing used is briefly described, it should be apparent that, describe below In accompanying drawing be only some embodiments of the present invention, for those of ordinary skill in the art, do not paying On the premise of going out creative work, the accompanying drawing obtaining other according to these accompanying drawings still falls within scope of the invention.

The flow chart of a kind of method revising ultrasound wave Partial Discharge Detection that Fig. 1 provides for the embodiment of the present invention；

Fig. 2 is the method flow diagram of step S1 in Fig. 1；

Fig. 3 is the method flow diagram of step S2 in Fig. 1；

Fig. 4 is the method flow diagram of step S3 in Fig. 1；

Fig. 5 is the method flow diagram of step S5 in Fig. 1；

Fig. 6 shows for the structure of a kind of system revising ultrasound wave Partial Discharge Detection that the embodiment of the present invention provides It is intended to.

Detailed description of the invention

For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing to the present invention It is described in further detail.

As it is shown in figure 1, be in the embodiment of the present invention, it is provided that a kind of ultrasound wave Partial Discharge Detection of revising Method, described method includes:

Step S1, from the historical data of equipment high s/n ratio ultrasound wave local discharge signal, extract multiple office Portion's discharge signal, and multiple local discharge signals of described extraction are converted into after linear spectral by main one-tenth Divide analytic process to process, obtain P and tie up main constituent coefficient matrix, and tie up main constituent coefficient according to the described P obtained Matrix, the P obtaining each local discharge signal ties up main constituent feature, further by the described each office obtained The P dimension main constituent feature of portion's discharge signal carries out Unsupervised clustering, obtains the disaggregated model of local discharge signal And the linear spectral average of corresponding each class；Wherein, P is natural number；

Detailed process is, step S11, from the historical data of equipment high s/n ratio ultrasound wave local discharge signal, Extract multiple local discharge signal, and the multiple local discharge signals extracted all are entered in units of certain time length After row framing, each local discharge signal after framing is all carried out Fourier transformation, obtain each local and put The linear spectral of the signal of telecommunication；

Step S12, by PCA, the linear spectral of arbitrary local discharge signal is carried out at dimensionality reduction Reason, obtains P and ties up main constituent coefficient matrix, and tie up main constituent coefficient matrix according to the P obtained, to extract Each local discharge signal all carries out linear transformation, and the P obtaining each local discharge signal ties up main constituent feature；

The P of each local discharge signal obtained is tieed up main constituent by step S13, employing k-mean algorithm simultaneously Feature carries out Unsupervised clustering, calculates multiple cluster centre and preserves the classification mould as local discharge signal Type；

Step S14, be added according to same class linear spectral after the principle averaged, to local discharge signal In disaggregated model, the linear spectral contained by each class calculates, and obtains in the disaggregated model of local discharge signal The linear spectral average of each class.

In one embodiment, the first step, the extraction signal to noise ratio ultrasound wave local discharge signal more than 20dB, After the multiple local discharge signals extracted all frame lengths in units of 1s are divided, carry out fast Fourier transform Calculate the DFT of signal, obtain the linear spectral of multiple local discharge signal；

Second step, linearly frequency spectrum is carried out principal component analysis, signal is dropped into P dimension, obtain P dimension main Composition coefficient matrix.In principal component analysis, for a sample data, observe p variable, n sample Data information battle array, concrete as shown in formula (1):

$X=\left(\begin{array}{cccc}{x}_{11}& x_{12}& \mathrm{...}& x_{1p}\\ x_{21}& x_{22}& \mathrm{...}& x_{2p}\\ .& .& .& .\\ .& .& .& .\\ .& .& .& .\\ x_{n1}& x_{n2}& \mathrm{...}& x_{np}\end{array}\right)=(x_1,x_2,\mathrm{...}{x}_p)---\left(1\right)$

Variable in formula (1) is the linear spectral of audio fragment.

Now, p observational variable is comprehensively become the aggregate variable that p is new by principal component analysis, variable turns to Shown in formula (2):

$\left\{\begin{array}{c}{F}_1=a_{11}{x}_1+a_{12}{x}_2+\mathrm{...}+a_{1p}{x}_p\\ F_2=a_{21}{x}_1+a_{22}{x}_2+\mathrm{...}+a_{2p}{x}_p\\ \mathrm{...}\\ F_p=a_{p1}{x}_1+a_{p2}{x}_2+\mathrm{...}+a_{pp}{x}_p\end{array}\right.---\left(2\right)$

In formula (2), F₁-F_nNew aggregate variable, the main constituent of i.e. original p variable of signal；F₁Side Difference maximum is referred to as first principal component use, F₂The big Second principal component, that is referred to as of variance time, by that analogy.

The coefficient of aforesaid equation (2) is calculated by principal component analysis (PCA) method, it may be assumed that

A, initial data is standardized process, as shown in formula (3)；

$\begin{array}{cc}{x}_{ij}^{*}=\frac{x_{ij}-{\stackrel{\‾}{x}}_j}{\sqrt{\mathrm{var}\left(x_j\right)}}& (i=1,2,\mathrm{...},n;j=1,2,\mathrm{...},p)\end{array}---\left(3\right)$

In formula (3),

B, calculating sample correlation coefficient matrix R, as shown in formula (4)；

$R=\left[\begin{array}{cccc}{r}_{11}& r_{12}& \mathrm{...}& r_{1p}\\ r_{21}& r_{22}& \mathrm{...}& r_{2p}\\ .& .& & .\\ .& .& \mathrm{...}& .\\ .& .& & .\\ r_{p1}& r_{p2}& \mathrm{...}& r_{pp}\end{array}\right]---\left(4\right)$

For convenience, it is assumed that still represent with X after initial data standardization, the data after the most normalized process Correlation coefficient is:

C, seek the eigenvalue (λ of correlation matrix R with Jacobian technique₁,λ₂…λ_p) and corresponding feature to Amount a_i=(a_i1, a_i2... a_ip), i=1,2 ... p；Thus obtain high s/n ratio ultrasound wave local discharge signal Main constituent coefficient matrix, i.e. as shown in formula (5):

$A=\left(\begin{array}{cccc}{a}_{11}& a_{12}& \mathrm{...}& a_{1p}\\ a_{21}& a_{22}& \mathrm{...}& a_{2p}\\ .& .& .& .\\ .& .& .& .\\ .& .& .& .\\ a_{p1}& a_{p2}& \mathrm{...}& a_{pp}\end{array}\right)=\left(\begin{array}{c}{a}_1\\ a_2\\ .\\ .\\ .\\ a_p\end{array}\right)---\left(5\right)$

3rd step, by main constituent coefficient matrices A, all signals are carried out linear transformation and obtain P Wei Zhuchengfente Levy, be characterized as inputting Unsupervised clustering for main constituent with the P of all signals, preserve its cluster centre and obtain M The disaggregated model of class ultrasound wave local discharge signal.All audio signal main constituent coefficient matrixes are become Changing, obtain P and tie up main constituent feature, the linear spectral of such as certain section ultrasound wave discharge signal is X={x₁x₂x₃...x₄x_p, dimensionality reduction is i.e. taken advantage of in main constituent coefficient matrix X_p=Ax, the P dimension obtaining this segment signal is main Composition characteristics X_p.Using the P dimensional feature of all signals as inlet flow Unsupervised clustering, use k-mean algorithm It is M class by ultrasound wave local discharge signal cluster.Wherein, M=10；

4th step, calculate the linear spectral average of each class respectively.The i.e. disaggregated model to local discharge signal In the linear spectral that comprised of each class sue for peace and do average, obtain the linear spectral average of such signal.

Step S2, obtain pure noise ultrasonic signal, and the described pure noise ultrasonic signal got is turned After being changed to linear spectral, calculate the linear spectral average of pure noise ultrasonic signal, and obtain according to described P tie up main constituent coefficient matrix, the linear spectral average of described pure noise ultrasonic signal is converted to described After the P dimension main constituent feature of pure noise ultrasonic signal, obtain the disaggregated model of pure noise ultrasonic signal；

Detailed process is, step S21, is obtaining before suspect signal, obtains actual environment a period of time scope Interior pure noise ultrasonic signal, and the pure noise ultrasonic signal got is entered in units of certain time length After row framing, the pure noise ultrasonic signal after framing is all carried out Fourier transformation, obtains pure noise ultrasonic The linear spectral of ripple signal；

Step S22, the linear spectral of the pure noise ultrasonic signal obtained is added after average, as pure The linear spectral average of noise ultrasonic signal；

The P that step S23, basis obtain ties up main constituent coefficient matrix, by the linear frequency of pure noise ultrasonic signal After spectrum average is converted to the P dimension main constituent feature of pure noise ultrasonic signal, only comprised in a cluster The disaggregated model of the heart also preserves the disaggregated model as pure noise ultrasonic signal.

In one embodiment, before actually detected, enroll the ultrasonic signal of one section of pure noise in scene, will This section of ultrasonic signal carries out framing with 1s for frame length, all frames is asked linear spectral and does average, obtaining pure The linear spectral average of noise signal.With P dimension linear by pure noise signal of main constituent coefficient matrix obtained Spectrum mean is converted to P and ties up main constituent feature, due to the linear spectral average only one of which of pure noise signal, P dimension main constituent feature after conversion is equivalent to a cluster centre, thus is only comprised in a cluster The disaggregated model of the pure noise ultrasonic signal of the heart.If linear spectral average is N, then the main constituent of P dimension is N_p=AN.

Step S3, by the linear spectral average of each class in described local discharge signal disaggregated model respectively with institute The linear spectral average stating pure noise ultrasonic signal disaggregated model is added, each class after being added Linear spectral average, and according to the described P obtained tie up main constituent coefficient matrix, by the described addition obtained After the linear spectral average of each class be converted to the P dimension main constituent feature of correspondence after, the disaggregated model obtained As the revised disaggregated model of local discharge signal；

Detailed process is, step S31, determines that in local discharge signal disaggregated model, the linear spectral of each class is equal Value, and by the linear spectral average of pure noise ultrasonic signal disaggregated model respectively with the line of each class determined Property spectrum mean be added, the linear spectral average of each class after being added；

The P that step S32, basis obtain ties up main constituent coefficient matrix, the line of each class after the addition that will obtain Property spectrum mean be converted to correspondence P dimension main constituent feature after, only comprised the classification of a cluster centre Model also preserves as the revised disaggregated model of described local discharge signal.

Step S4, acquisition suspect signal, and the described suspect signal got is converted to linear spectral, and Tie up main constituent coefficient matrix according to the described P obtained, the linear spectral of described suspect signal is converted to described The P of suspect signal ties up main constituent feature；

Step S5, the P of described suspect signal is tieed up main constituent feature introduce described pure noise ultrasound wave letter respectively Number disaggregated model and the revised disaggregated model of described local discharge signal in, calculate described suspect signal Arrive disaggregated model and the described local discharge signal revised classification mould of described pure noise ultrasonic signal respectively The Euclidean distance of type, and according to described two Euclidean distances obtained, filter out in said two Euclidean distance The disaggregated model that minima is corresponding, treats signal corresponding in institute's sifting sort model as described further The testing result output of inspection signal.

Detailed process is, step S51, the P of suspect signal ties up main constituent feature, and to introduce pure noise respectively ultrasonic In the disaggregated model of ripple signal and the revised disaggregated model of local discharge signal, calculate suspect signal respectively Disaggregated model and the Euclidean of the revised disaggregated model of local discharge signal to described pure noise ultrasonic signal Distance；

Step S52, judge that suspect signal is the least to the Euclidean distance of the disaggregated model of pure noise ultrasonic signal Euclidean distance in the revised disaggregated model of suspect signal to local discharge signal；If it is, under Zhi Hanging One step S53；If it is not, then redirect execution step S54；

Step S53, the disaggregated model of screening are the disaggregated model of pure noise ultrasonic signal, and by suspect signal Export as testing result for pure noise ultrasonic signal；

Step S54, the disaggregated model of screening are the revised disaggregated model of local discharge signal, and by letter to be checked Number it is that local discharge signal exports as testing result.

It should be noted that, k-mean clustering algorithm and Euclidean distance algorithm belong to the conventional calculation of art technology Method, does not repeats at this.

As shown in Figure 6, in the embodiment of the present invention, it is provided that a kind of ultrasound wave Partial Discharge Detection revised System, described system includes:

Local discharge signal analyzes model acquiring unit 610, for from equipment high s/n ratio ultrasound wave shelf depreciation In the historical data of signal, extract multiple local discharge signal, and multiple shelf depreciations of described extraction are believed Processed by PCA after number being converted into linear spectral, obtain P and tie up main constituent coefficient matrix, and Tie up main constituent coefficient matrix according to the described P obtained, obtain the P Wei Zhuchengfente of each local discharge signal Levy, further the P of the described each local discharge signal obtained tieed up main constituent feature and carry out Unsupervised clustering, Obtain disaggregated model and the linear spectral average of corresponding each class thereof of local discharge signal；Wherein, P is nature Number；

Pure noise ultrasound signal analyzing model acquiring unit 620, is used for obtaining pure noise ultrasonic signal, and After the described pure noise ultrasonic signal got is converted to linear spectral, calculate pure noise ultrasound wave letter Number linear spectral average, and according to the described P obtained tie up main constituent coefficient matrix, described pure noise is surpassed After the linear spectral average of acoustic signals is converted to the P dimension main constituent feature of described pure noise ultrasonic signal, Obtain the disaggregated model of pure noise ultrasonic signal；

Local discharge signal analyzes Modifying model unit 630, for by described local discharge signal disaggregated model The linear spectral average of each class respectively with the linear spectral average of described pure noise ultrasonic signal disaggregated model It is added, the linear spectral average of each class after being added, and ties up main one-tenth according to the described P obtained Divide coefficient matrix, the linear spectral average of each class after the described addition obtained is converted to the P dimension of correspondence After main constituent feature, the disaggregated model obtained is as the revised disaggregated model of local discharge signal；

Suspect signal analytic unit 640, is used for obtaining suspect signal, and the described suspect signal got is turned It is changed to linear spectral, and ties up main constituent coefficient matrix according to the described P obtained, by the line of described suspect signal Property frequency spectrum be converted to described suspect signal P tie up main constituent feature；

Suspect signal recognition unit 650, introduces institute respectively for the P of described suspect signal is tieed up main constituent feature State in disaggregated model and the revised disaggregated model of described local discharge signal of pure noise ultrasonic signal, meter Calculate described suspect signal and arrive disaggregated model and the described shelf depreciation letter of described pure noise ultrasonic signal respectively The Euclidean distance of number revised disaggregated model, and according to described two Euclidean distances obtained, filter out institute State the disaggregated model that in two Euclidean distances, minima is corresponding, further by corresponding in institute's sifting sort model Signal as described suspect signal testing result export.

Wherein, described local discharge signal analysis model acquiring unit 610 includes:

Local discharge signal linear spectral conversion module 6101, for putting from equipment high s/n ratio ultrasound wave local In the historical data of the signal of telecommunication, extract multiple local discharge signal, and by multiple shelf depreciations of described extraction After signal all carries out framing in units of certain time length, each local discharge signal after described framing is all entered Row Fourier transformation, obtains the linear spectral of each local discharge signal；

Principal component analysis module 6102, for by linear to arbitrary local discharge signal of PCA Frequency spectrum carries out dimension-reduction treatment, obtains P and ties up main constituent coefficient matrix, and ties up main constituent according to the described P obtained Coefficient matrix, all carries out linear transformation to each local discharge signal of described extraction, obtains each local and put The P of the signal of telecommunication ties up main constituent feature；

Local discharge signal Clustering Model computing module 6103, is used for using k-mean algorithm to obtain described simultaneously To each local discharge signal P dimension main constituent feature carry out Unsupervised clustering, calculate in multiple cluster The heart also preserves the disaggregated model as local discharge signal；

Local discharge signal Clustering Model classification mean value computation module 6104, for according to same class linear spectral The principle averaged after addition, linear contained by each class in the disaggregated model to described local discharge signal Frequency spectrum calculates, and obtains the linear spectral average of each class in the disaggregated model of described local discharge signal.

Wherein, described pure noise ultrasound signal analyzing model acquiring unit 620 includes:

Pure noise ultrasonic signal linear spectral conversion module 6201, is used for before obtaining described suspect signal, Obtain the pure noise ultrasonic signal in the range of actual environment a period of time, and by the described pure noise got After ultrasonic signal carries out framing in units of certain time length, to the pure noise ultrasonic signal after described framing All carry out Fourier transformation, obtain the linear spectral of described pure noise ultrasonic signal；

Linear spectral mean value computation module 6202, linear for by the described pure noise ultrasonic signal obtained Frequency spectrum is averaged, as the linear spectral average of described pure noise ultrasonic signal after being added；

Pure noise ultrasonic signal Clustering Model computing module 6203, ties up main one-tenth for the P obtained described in basis Divide coefficient matrix, the linear spectral average of described pure noise ultrasonic signal is converted to described pure noise ultrasonic After the P dimension main constituent feature of ripple signal, only comprised the disaggregated model of a cluster centre and preserved conduct The disaggregated model of described pure noise ultrasonic signal.

Wherein, described local discharge signal analysis Modifying model unit 630 includes:

Linear spectral accumulator module 6301, for determining each class in described local discharge signal disaggregated model Linear spectral average, and by the linear spectral average of described pure noise ultrasonic signal disaggregated model respectively with institute The linear spectral average stating each class determined is added, the linear spectral average of each class after being added；

Local discharge signal Clustering Model correcting module 6302, ties up main constituent system for the P obtained described in basis Matrix number, the P that the linear spectral average of each class after the described addition obtained is converted to correspondence ties up main one-tenth After dtex is levied, only comprised the disaggregated model of a cluster centre and preserved as described local discharge signal Revised disaggregated model.

Wherein, described suspect signal recognition unit 650 includes:

Oldham distance calculating module 6501, introduces respectively for the P of described suspect signal is tieed up main constituent feature In the disaggregated model of described pure noise ultrasonic signal and the revised disaggregated model of described local discharge signal, Calculate described suspect signal and arrive the disaggregated model of described pure noise ultrasonic signal and described shelf depreciation respectively The Euclidean distance of the disaggregated model after signal correction；

Judge module 6502, for judging the described suspect signal classification mould to described pure noise ultrasonic signal Whether the Euclidean distance of type is less than described suspect signal to the revised disaggregated model of described local discharge signal Euclidean distance；

First result output module 6503, the disaggregated model for described screening is described pure noise ultrasound wave letter Number disaggregated model, and to be pure noise ultrasonic signal using described suspect signal export as testing result；

Second result output module 6504, the disaggregated model for described screening is that described local discharge signal is repaiied Disaggregated model after just, and be that local discharge signal exports as testing result using described suspect signal.

Implement the embodiment of the present invention, have the advantages that

In embodiments of the present invention, owing to using principal component analytical method can be effectively reduced training data Dimension reduces data complexity, uses being modified disaggregated model of environment noise average, Ke Yiyou simultaneously Effect improves detection method performance under all kinds of noise circumstances such that it is able to avoids artifact to disturb, and draws Entering intrinsic dimensionality low, training data is few, thus reaches the time saving and energy saving and objective purpose of testing result.

It should be noted that in said system embodiment, each included system unit is according to function Logic carries out dividing, but is not limited to above-mentioned division, as long as being capable of corresponding function； It addition, the specific name of each functional unit is also only to facilitate mutually distinguish, it is not limited to the present invention Protection domain.

One of ordinary skill in the art will appreciate that all or part of step realizing in above-described embodiment method is Can instruct relevant hardware by program to complete, described program can be stored in a computer-readable Taking in storage medium, described storage medium, such as ROM/RAM, disk, CD etc..

Above disclosed be only present pre-ferred embodiments, certainly can not with this limit the present invention it Interest field, the equivalent variations therefore made according to the claims in the present invention, still belong to the scope that the present invention is contained.

Claims (10)

1. the method revising ultrasound wave Partial Discharge Detection, it is characterised in that described method includes:

A, from the historical data of equipment high s/n ratio ultrasound wave local discharge signal, extract multiple shelf depreciation Signal, and pass through principal component analysis after multiple local discharge signals of described extraction are converted into linear spectral Method processes, and obtains P and ties up main constituent coefficient matrix, and ties up main constituent coefficient matrix according to the described P obtained, The P obtaining each local discharge signal ties up main constituent feature, further by the described each shelf depreciation obtained The P dimension main constituent feature of signal carries out Unsupervised clustering, obtains the disaggregated model of local discharge signal and right Answer the linear spectral average of each class；Wherein, P is natural number；

B, obtain pure noise ultrasonic signal, and the described pure noise ultrasonic signal got is converted to line Property frequency spectrum after, calculate the linear spectral average of pure noise ultrasonic signal, and tie up according to the described P obtained Main constituent coefficient matrix, is converted to described pure noise by the linear spectral average of described pure noise ultrasonic signal After the P dimension main constituent feature of ultrasonic signal, obtain the disaggregated model of pure noise ultrasonic signal；

C, the linear spectral average of each class in described local discharge signal disaggregated model pure is made an uproar with described respectively The linear spectral average of sound ultrasonic signal disaggregated model is added, each class after being added linear Spectrum mean, and tie up main constituent coefficient matrix according to the described P obtained, every by after the described addition obtained After the linear spectral average of one class is converted to the P dimension main constituent feature of correspondence, the disaggregated model obtained is as office The revised disaggregated model of portion's discharge signal；

D, acquisition suspect signal, and the described suspect signal got is converted to linear spectral, and according to institute State the P obtained and tie up main constituent coefficient matrix, the linear spectral of described suspect signal is converted to described letter to be checked Number P tie up main constituent feature；

E, the P of described suspect signal is tieed up main constituent feature introduce respectively described pure noise ultrasonic signal point In class model and the revised disaggregated model of described local discharge signal, calculate described suspect signal and arrive respectively The disaggregated model of described pure noise ultrasonic signal and the Europe of the revised disaggregated model of described local discharge signal Family name's distance, and according to described two Euclidean distances obtained, filter out minima in said two Euclidean distance Corresponding disaggregated model, further using signal corresponding in institute's sifting sort model as described suspect signal Testing result output.

2. the method for claim 1, it is characterised in that described step a specifically includes:

From the historical data of equipment high s/n ratio ultrasound wave local discharge signal, extract multiple shelf depreciation letter Number, and after multiple local discharge signals of described extraction are all carried out framing in units of certain time length, to institute State each local discharge signal after framing and all carry out Fourier transformation, obtain the line of each local discharge signal Property frequency spectrum；

By PCA, the linear spectral of arbitrary local discharge signal is carried out dimension-reduction treatment, obtain P Dimension main constituent coefficient matrix, and tie up main constituent coefficient matrix, every to described extraction according to the described P obtained One local discharge signal all carries out linear transformation, and the P obtaining each local discharge signal ties up main constituent feature；

Use the k-mean algorithm P simultaneously to the described each local discharge signal obtained to tie up main constituent feature to enter Row Unsupervised clustering, calculates multiple cluster centre and preserves the disaggregated model as local discharge signal；

The principle averaged after being added according to same class linear spectral, the classification to described local discharge signal In model, the linear spectral contained by each class calculates, and obtains in the disaggregated model of described local discharge signal The linear spectral average of each class.

3. the method for claim 1, it is characterised in that described step b specifically includes:

Before obtaining described suspect signal, obtain the pure noise ultrasound wave in the range of actual environment a period of time Signal, and after the described pure noise ultrasonic signal got is carried out framing in units of certain time length, right Pure noise ultrasonic signal after described framing all carries out Fourier transformation, obtains described pure noise ultrasound wave letter Number linear spectral；

Average, as described pure after the linear spectral of the described pure noise ultrasonic signal obtained is added The linear spectral average of noise ultrasonic signal；

Main constituent coefficient matrix is tieed up, by the linear frequency of described pure noise ultrasonic signal according to the described P obtained After spectrum average is converted to the P dimension main constituent feature of described pure noise ultrasonic signal, is only comprised one and gathered The disaggregated model at class center also preserves the disaggregated model as described pure noise ultrasonic signal.

4. the method for claim 1, it is characterised in that described step c specifically includes:

Determine the linear spectral average of each class in described local discharge signal disaggregated model, and pure make an uproar described The linear spectral average of sound ultrasonic signal disaggregated model is equal with the linear spectral of the described each class determined respectively Value is added, the linear spectral average of each class after being added；

Main constituent coefficient matrix is tieed up, by the line of each class after the described addition obtained according to the described P obtained Property spectrum mean be converted to correspondence P dimension main constituent feature after, only comprised the classification of a cluster centre Model also preserves as the revised disaggregated model of described local discharge signal.

5. the method for claim 1, it is characterised in that described step e specifically includes:

The P of described suspect signal is tieed up main constituent feature and introduces the classification of described pure noise ultrasonic signal respectively In model and the revised disaggregated model of described local discharge signal, calculate described suspect signal respectively to institute State disaggregated model and the Euclidean of the revised disaggregated model of described local discharge signal of pure noise ultrasonic signal Distance；

Judge that described suspect signal is the least to the Euclidean distance of the disaggregated model of described pure noise ultrasonic signal Euclidean distance in described suspect signal to the revised disaggregated model of described local discharge signal；

If it is, the disaggregated model that the disaggregated model of described screening is described pure noise ultrasonic signal, and It is that pure noise ultrasonic signal exports as testing result using described suspect signal；

If it is not, then the disaggregated model of described screening is the revised disaggregated model of described local discharge signal, And to be local discharge signal using described suspect signal export as testing result.

6. the system revising ultrasound wave Partial Discharge Detection, it is characterised in that described system includes:

Local discharge signal analyzes model acquiring unit, for believing from equipment high s/n ratio ultrasound wave shelf depreciation Number historical data in, extract multiple local discharge signal, and by multiple local discharge signals of described extraction Processed by PCA after being converted into linear spectral, obtain P and tie up main constituent coefficient matrix, and root Tieing up main constituent coefficient matrix according to the described P obtained, the P obtaining each local discharge signal ties up main constituent feature, Further the P of the described each local discharge signal obtained is tieed up main constituent feature and carry out Unsupervised clustering, Disaggregated model and the linear spectral average of corresponding each class thereof to local discharge signal；Wherein, P is natural number；

Pure noise ultrasound signal analyzing model acquiring unit, is used for obtaining pure noise ultrasonic signal, and will After the described pure noise ultrasonic signal got is converted to linear spectral, calculate pure noise ultrasonic signal Linear spectral average, and according to the described P obtained tie up main constituent coefficient matrix, by ultrasonic for described pure noise After the linear spectral average of ripple signal is converted to the P dimension main constituent feature of described pure noise ultrasonic signal, Disaggregated model to pure noise ultrasonic signal；

Local discharge signal analyzes Modifying model unit, for by every in described local discharge signal disaggregated model The linear spectral average of one class linear spectral average with described pure noise ultrasonic signal disaggregated model respectively is entered Row is added, the linear spectral average of each class after being added, and ties up main constituent according to the described P obtained Coefficient matrix, the P dimension that the linear spectral average of each class after the described addition obtained is converted to correspondence is main After composition characteristics, the disaggregated model obtained is as the revised disaggregated model of local discharge signal；

Suspect signal analytic unit, is used for obtaining suspect signal, and the described suspect signal got is changed For linear spectral, and tie up main constituent coefficient matrix according to the described P obtained, linear by described suspect signal Frequency spectrum is converted to the P of described suspect signal and ties up main constituent feature；

Suspect signal recognition unit, introduces described for the P of described suspect signal ties up main constituent feature respectively In the disaggregated model of pure noise ultrasonic signal and the revised disaggregated model of described local discharge signal, calculate Go out described suspect signal and arrive the disaggregated model of described pure noise ultrasonic signal and described local discharge signal respectively The Euclidean distance of revised disaggregated model, and according to described two Euclidean distances obtained, filter out described The disaggregated model that in two Euclidean distances, minima is corresponding, further by corresponding in institute's sifting sort model Signal exports as the testing result of described suspect signal.

7. system as claimed in claim 6, it is characterised in that described local discharge signal is analyzed model and obtained Take unit to include:

Local discharge signal linear spectral conversion module, for believing from equipment high s/n ratio ultrasound wave shelf depreciation Number historical data in, extract multiple local discharge signal, and by multiple local discharge signals of described extraction After all carrying out framing in units of certain time length, each local discharge signal after described framing is all carried out Fu In leaf transformation, obtain the linear spectral of each local discharge signal；

Principal component analysis module, for by the PCA linear spectral to arbitrary local discharge signal Carry out dimension-reduction treatment, obtain P and tie up main constituent coefficient matrix, and tie up main constituent coefficient according to the described P obtained Matrix, all carries out linear transformation to each local discharge signal of described extraction, obtains each shelf depreciation letter Number P tie up main constituent feature；

Local discharge signal Clustering Model computing module, for using k-mean algorithm to obtain described simultaneously The P dimension main constituent feature of each local discharge signal carries out Unsupervised clustering, calculates multiple cluster centre also Preserve the disaggregated model as local discharge signal；

Local discharge signal Clustering Model classification mean value computation module, for being added according to same class linear spectral After the principle averaged, linear spectral contained by each class in the disaggregated model to described local discharge signal Calculate, obtain the linear spectral average of each class in the disaggregated model of described local discharge signal.

8. system as claimed in claim 6, it is characterised in that described pure noise ultrasound signal analyzing mould Type acquiring unit includes:

Pure noise ultrasonic signal linear spectral conversion module, for, before obtaining described suspect signal, obtaining Take the pure noise ultrasonic signal in the range of actual environment a period of time, and the described pure noise got is surpassed After acoustic signals carries out framing in units of certain time length, equal to the pure noise ultrasonic signal after described framing Carry out Fourier transformation, obtain the linear spectral of described pure noise ultrasonic signal；

Linear spectral mean value computation module, for by the linear spectral of the described pure noise ultrasonic signal obtained Average after addition, as the linear spectral average of described pure noise ultrasonic signal；

Pure noise ultrasonic signal Clustering Model computing module, ties up main constituent system for the P obtained described in basis Matrix number, is converted to described pure noise ultrasound wave letter by the linear spectral average of described pure noise ultrasonic signal Number P dimension main constituent feature after, only comprised the disaggregated model of a cluster centre and preserved as described The disaggregated model of pure noise ultrasonic signal.

9. system as claimed in claim 6, it is characterised in that described local discharge signal is analyzed model and repaiied Positive unit includes:

Linear spectral accumulator module, for determining the linear of each class in described local discharge signal disaggregated model Spectrum mean, and by the linear spectral average of described pure noise ultrasonic signal disaggregated model respectively with described really The linear spectral average of fixed each class is added, the linear spectral average of each class after being added；

Local discharge signal Clustering Model correcting module, ties up main constituent coefficient square for the P obtained described in basis Battle array, is converted to the P Wei Zhuchengfente of correspondence by the linear spectral average of each class after the described addition obtained After levying, only comprised the disaggregated model of a cluster centre and preserved as described local discharge signal correction After disaggregated model.

10. system as claimed in claim 6, it is characterised in that described suspect signal recognition unit includes:

Oldham distance calculating module, introduces described for the P of described suspect signal ties up main constituent feature respectively In the disaggregated model of pure noise ultrasonic signal and the revised disaggregated model of described local discharge signal, calculate Go out described suspect signal and arrive the disaggregated model of described pure noise ultrasonic signal and described local discharge signal respectively The Euclidean distance of revised disaggregated model；

Judge module, for judging the described suspect signal disaggregated model to described pure noise ultrasonic signal Whether Euclidean distance is less than the described suspect signal Euclidean to the revised disaggregated model of described local discharge signal Distance；

First result output module, the disaggregated model for described screening is described pure noise ultrasonic signal Disaggregated model, and be that pure noise ultrasonic signal exports as testing result using described suspect signal；

Second result output module, after the disaggregated model of described screening is described local discharge signal correction Disaggregated model, and to be local discharge signal using described suspect signal export as testing result.