CN103018537B - The Classification of Transient Power Quality Disturbances recognition methods of kurtosis is composed based on CWD - Google Patents

Abstract

The present invention proposes and a kind ofly compose the Classification of Transient Power Quality Disturbances recognition methods of kurtosis based on CWD, and it is combined with effective value be applied to five kinds of single identifications with the transient power quality disturbance of compound.First this recognition methods is fallen by voltage swell, temporarily and is interrupted this three kinds of amplitude classes disturbance and regard a class as, by CWD compose kurtosis value by power office's data acquisition system to data be divided into transient state pulse, transient oscillation and amplitude class disturbance three class, data are divided into voltage swell, voltage dip and voltage interruption by the effective value again by calculating the disturbance of amplitude class, and export required classification results to subsequent processing device.The disturbance type that the present invention directly utilizes the size discrimination of numerical value different, does not need to use any sorter, greatly simplify the flow process of identification and the time of identification.The compound disturbance that the inventive method accurately can be distinguished transient state pulse, transient oscillation and rise temporarily, falls temporarily and interrupt these five kinds of single transient disturbance and combine, also has good noise immunity.

Description

The Classification of Transient Power Quality Disturbances recognition methods of kurtosis is composed based on CWD

Technical field

The present invention relates to Power System Intelligent monitoring, especially based on the transient power quality Classification and Identification technical field of high-order statistic and signal transacting.

Background technology

Along with a large amount of inputs of the continuous expansion of electric system scale and various power electronic equipment, nonlinear-load, impact load, the various disturbance events in electric system seriously have impact on quality and the daily life of industrial products.Because the influence degree of dissimilar electrical energy power quality disturbance is different, therefore, identification is carried out to electrical energy power quality disturbance extremely important.

The identification of electrical energy power quality disturbance comprises feature extraction and Classification and Identification two steps.Conventional feature extracting method has Short Time Fourier Transform, wavelet transformation, S-transformation, HHT to convert, and wherein the measuring accuracy of Fourier transform is subject to the impact of spectrum leakage and fence effect, is not suitable for the Power Disturbance signal analyzing non-stationary; S-transformation accurately cannot measure the characteristic parameter of fundamental frequency fluctuation and m-Acetyl chlorophosphonazo; Hilbert-Huang transform (HHT:Hilbert-HuangTransform) empirical mode decomposition thoroughly not there will be the problem of chaff component and modal overlap; Multiple continuous wavelet transform (CWT:ContinuousWaveletTransform) exists overlapping due to the wavelet function frequency domain window that its centre frequency is close, have impact on the measurement of harmonic wave or m-Acetyl chlorophosphonazo component characterization parameter, be unfavorable for the accurate differentiation of quality of power supply event type.Conventional sorter has artificial neural network (ANN:ArtificialNeuralNetwork), support vector machine (SVM:SupportVectorMachine), expert system (ES:ExpertSystem) etc.The sorter training speed of conventional ANN is comparatively slow, and cannot provide the local of signal trickle situation, its accuracy also has much room for improvement; The sorter training time of SVM is short, recognition accuracy is high, and to insensitive for noise, but the method is comparatively difficult in identification hybrid perturbation; The sorter of ES easily produces shot array problem when quality of power supply kind of event increases.And above-mentioned sorter all needs mass data to carry out training and testing, and the power quality data in reality is not easy to obtain, and this just makes algorithm be difficult in practice realize.

The concept of spectrum kurtosis is proposed by Dwyer the earliest, is used for detecting the transient state composition in signals and associated noises.Subsequently, V.Vrabie definition spectrum kurtosis be a process apart from Gaussian tolerance, and to apply it in rolling bearing fault diagnosis.J.Antoni system definition spectrum kurtosis, and propose the spectrum kurtosis based on STFT (Short-TimeFourierTransform), demonstrate it and there is the ability detecting non-stationary signal in additive noise.N.Sawalhi proposes the spectrum kurtosis method based on wavelet transformation and WVD (Wigner-VilleDistribution, Wigner-Ville distribute), and applies it in mechanical fault diagnosis.

Spectrum kurtosis can non-stationary in characterization signal and non-gauss component, and can automatically suppress white noise to disturb.The present invention proposes a kind of spectrum kurtosis algorithm newly namely based on the spectrum kurtosis algorithm of CWD distribution (Choi-WilliamsDistribution), and it is combined with effective value be applied in transient power quality disturbance identification, simulate signal is identified, demonstrates feasibility and the validity of the method.

Summary of the invention

The object of this invention is to provide a kind of transient power quality classifying identification method be combined with effective value based on CWD spectrum kurtosis.Make it rising temporarily, fall temporarily, interrupt, feature difference that pulse and vibration five class transient disturbance extract obviously, and has stronger noiseproof feature.

The technical solution adopted in the present invention is: the transient power quality classifying identification method be combined with effective value based on CWD spectrum kurtosis, and identify transient disturbance in electric system, its specific practice is:

A, extraction perturbation features signal

If u (n) is the input voltage signal containing disturbing signal, n=1,2 ..., N, N are data length.U (n) is carried out wavelet transformation, extracts perturbation features signal u _r(n).

B, calculation perturbation characteristic signal spectrum kurtosis

First perturbation features signal u is obtained _rn the Choi-Williams of () is distributed as Cu _r(t, f), then according to Cu _rthe instantaneous spectral moment in 2n rank of (t, f) draws 2 rank and 4 rank instantaneous spectrum distances, finally according to the definition of spectrum kurtosis, can try to achieve u _rthe spectrum kurtosis of (n)

Specific practice is:

B1, perturbation features signal u _rn (), calculate its Choi-Williams and distribute, result is Cu _r(t, f).

${\mathrm{Cu}}_r(t,f)=\underset{-\∞}{\overset{\∞}{\∫}}\sqrt{\frac{\mathrm{\σ}}{4{\mathrm{\π\τ}}^2}}{e}^{(-\frac{{\mathrm{\σt}}^2}{{4\mathrm{\τ}}^2})}{u}_r(t+f+\frac{\mathrm{\τ}}{2}){u_r}^{*}(t+f-\frac{\mathrm{\τ}}{2})e^{-j2\mathrm{\πf\τ}}\mathrm{d\τ}---\left(3\right)$

B2, according to Choi-Williams distribute Cu _rthe 2 rank instantaneous spectrum distances that (t, f) asks for with 4 rank instantaneous spectrum distances

${\hat{S}}_2\left(f\right){=<{\left|{\mathrm{Cu}}_r(t,f)\right|}^2>}_k$

$\left(4\right)$

${\hat{S}}_4\left(f\right)={<{\left|{\mathrm{Cu}}_r(t,f)\right|}^4>}_k$

Wherein, <> _krepresent be k rank time average.

B3, according to spectrum kurtosis definition, try to achieve the spectrum kurtosis based on CWD.

${\hat{K}}_x\left(f\right)=\frac{{\hat{S}}_{4x}\left(f\right)}{{\hat{S}}_{2x}^2\left(f\right)}-2,f\&NotEqual;0---\left(5\right)$

C, according to threshold classification identification

According to the maximal value K of spectrum kurtosis _maxwith the little value K of spectrum kurtosis _minsize, suitable threshold value is set, judge input signal belong to the disturbance of amplitude class, transient state pulse or transient oscillation;

D, effective value method distinguish the disturbance of amplitude class

By the amplitude class class disturbance do not distinguished, calculate its effective value curve, judge it is voltage swell, voltage dip or voltage interruption by the size of amplitude, draw and export required classification results to subsequent processing device.

The specific practice that above-mentioned A step extracts perturbation features signal is:

If u (n) is the input voltage signal containing disturbing signal, n=1,2 ..., N, N are data length.U (n) is carried out wavelet transformation.

$u\left(n\right)=\underset{j=1}{\overset{J}{\mathrm{\&Sigma;}}}{D}_j\left(n\right)+A_j\left(n\right)---\left(1\right)$

A in formula _jn () represents the approximation component of yardstick j, D _jn () represents the details coefficients of yardstick j, J is maximum decomposition scale.As j=J, A _jn () is the approximation component under maximum decomposition scale, can think that this approximation component only comprises power frequency component, then disturbing signal can be expressed from the next

u _r(n)＝u(n)-A _J(n)(2)

The specific practice that above-mentioned B walks calculation perturbation characteristic signal spectrum kurtosis is:

B1, perturbation features signal u _rn (), calculate its Choi-Williams and distribute, result is Cu _r(t, f).

${\mathrm{Cu}}_r(t,f)=\underset{-\&infin;}{\overset{\&infin;}{\&Integral;}}\sqrt{\frac{\mathrm{\&sigma;}}{4{\mathrm{\&pi;\&tau;}}^2}}{e}^{(-\frac{{\mathrm{\&sigma;t}}^2}{{4\mathrm{\&tau;}}^2})}{u}_r(t+f+\frac{\mathrm{\&tau;}}{2}){u_r}^{*}(t+f-\frac{\mathrm{\&tau;}}{2})e^{-j2\mathrm{\&pi;f\&tau;}}\mathrm{d\&tau;}---\left(3\right)$

B2, according to Choi-Williams distribute Cu _rthe 2 rank instantaneous spectrum distances that (t, f) asks for with 4 rank instantaneous spectrum distances

${\hat{S}}_2\left(f\right){=<{\left|{\mathrm{Cu}}_r(t,f)\right|}^2>}_k$

$\left(4\right)$

${\hat{S}}_4\left(f\right)={<{\left|{\mathrm{Cu}}_r(t,f)\right|}^4>}_k$

Wherein, <> _krepresent be k rank time average.

B3, according to spectrum kurtosis definition, try to achieve the spectrum kurtosis based on CWD.

${\hat{K}}_x\left(f\right)=\frac{{\hat{S}}_{4x}\left(f\right)}{{\hat{S}}_{2x}^2\left(f\right)}-2,f\&NotEqual;0---\left(5\right)$

The choosing of above-mentioned C step according to the specific practice of threshold classification identification is:

According to the maximal value K of spectrum kurtosis _maxwith the little value K of spectrum kurtosis _minsize, suitable threshold value is set, judge input signal belong to the disturbance of amplitude class, transient state pulse or transient oscillation;

The specific practice that above-mentioned D walks the disturbance of effective value method identification amplitude class is:

By the amplitude class disturbance do not distinguished, calculate its effective value curve, judge it is voltage swell, voltage dip or voltage interruption by the size of amplitude, draw and export required classification results to subsequent processing device.

According to definition, try to achieve effective value:

$U_{\mathrm{RMS}}=\sqrt{\frac{1}{N}\underset{n=k-N+1}{\overset{n=k}{\mathrm{\&Sigma;}}}{u}_i^2}---\left(6\right)$

The invention has the advantages that:

1, spectrum kurtosis method of the present invention belongs to the category of higher-order statistical method, and spectrum kurtosis is a fourth order cumulant, can suppress white noise completely in theory, the non-stationary in characterization signal and non-Gaussian signal, and can determine its position on frequency band.Spectrum kurtosis method self has good character.

2, of the present invention the adopted spectrum kurtosis computing method based on CWD, a kind of spectrum kurtosis computing method newly, not only inherit the ability of the higher time-frequency focusing of CWD and suppressing crossterms interference, and meet most of time-frequency distributions character of Cohen class time-frequency distributions, can non-stationary that is more clear, that comprise in characterization signal more accurately and non-gauss component.

3, the perturbation features amount that the present invention extracts makes different classes of disturbance mutually can be distinguished fully, to rising temporarily, falling temporarily and interrupt the method that three class disturbances have employed effective value, the combination of spectrum kurtosis and these two kinds of methods of effective value is simply effective, anti-noise is effective, to extracted characteristic quantity, suitable threshold value is set and is distinguished, do not need to add sorter, improve classification speed and accuracy, be also easy in actual applications realize.

Accompanying drawing explanation

Fig. 1 is overhaul flow chart of the present invention.

Fig. 2 is the key step block diagram of detection module of the present invention.

Fig. 3 is 220KV transmission line of electricity simplified model figure in the embodiment of the present invention one.

Fig. 4 is the lower five kinds of noisy input disturbance signal graphs of 30dB in the embodiment of the present invention one.

Fig. 5 is the disturbance component figure of five kinds of noisy disturbing signals in the embodiment of the present invention one.

Fig. 6 is five kinds of disturbance component spectrum kurtosis figure in the embodiment of the present invention one.

Fig. 7 is the effective value figure of effective value figure: Fig. 7 a voltage swell of amplitude class disturbance in the embodiment of the present invention one, the effective value figure of the effective value figure of Fig. 7 b voltage dip, Fig. 7 c voltage interruption.

Fig. 8 is the design sketch in the embodiment of the present invention one under different SNR (signaltonoiseratio), is the Transient Disturbance Signal recognition result under different noise.

Fig. 9 is the design sketch in the embodiment of the present invention one under different amplitude.

Figure 10 is the recognition effect figure of the compound disturbance of transient state pulse and voltage swell in the embodiment of the present invention one.

Embodiment

Below in conjunction with accompanying drawing and concrete embodiment, the present invention is further detailed explanation.

Embodiment one

Shown in Fig. 2, a kind of embodiment of the present invention is: the transient power quality classifying identification method be combined with effective value based on CWD spectrum kurtosis, and its concrete practice is:

A, extraction perturbation features signal

Utilize PSCAD/EMTDC to establish a model of power transmission system, schematic diagram is Fig. 3, and power supply E1, E2 are 220kV, and phase angle is zero; A1, A2, A3 are bus; B1, B2, B3, B4, B5, B6 are isolating switch; C1, C2 are ground capacitance; R1 is resistance to earth.

(1) impulse transients is produced.Add the controlled current source simulation thunder-strike phenomenon that a control source is lightning current at a M place, obtain impulse transients signal.

(2) vibration transient state is produced.Drop into the direct earth capacitance group C3 of 1uF at bus A3 place, obtain vibration transient signal.

(3) disturbance of amplitude class is produced.At bus A2 place input stake resistance R1, a short trouble can be added at N place, obtain the simulate signal of amplitude class disturbance.

Produce simulate signal with PSCAD/EMTDC, but be often mingled with noise in the disturbing signal of actual acquisition, therefore, need to add noise.

Fig. 4 illustrates the white noise simulation actual environment adding signal to noise ratio (S/N ratio) (signaltonoiseratio, SNR) 30dB, and sample frequency is set to 10kHz, original five kinds of substantially noisy disturbing signals of collection.

Fig. 5 illustrates that employing utilizes wavelet transformation to extract transient oscillation disturbance component separately.Its specific practice is:

Wavelet transformation is carried out to original noisy disturbing signal u (n).

$u\left(n\right)=\underset{j=1}{\overset{J}{\mathrm{\&Sigma;}}}{D}_j\left(n\right)+A_j\left(n\right)---\left(1\right)$

A in formula _jn () represents the approximation component of yardstick j, D _jn () represents the details coefficients of yardstick j, J is maximum decomposition scale.As j=J, A _jn () is the approximation component under maximum decomposition scale, can think that this approximation component only comprises power frequency component, then disturbing signal can be expressed from the next

B、u _r(n)＝u(n)-A _J(n)(2)

C, calculation perturbation characteristic signal spectrum kurtosis

Fig. 6 illustrates the spectrum kurtosis of the disturbance component based on the voltage swell of CWD, voltage dip, voltage interruption, transient state pulse and transient oscillation.

Fig. 7 illustrates the effective value of voltage swell, voltage dip and voltage interruption.

Fig. 8 illustrates the recognition effect figure under different signal to noise ratio (S/N ratio) condition.

Fig. 9 illustrates the recognition effect figure under different amplitude.

Figure 10 illustrates the recognition effect figure of compound disturbance.

First perturbation features signal u is obtained _rn the Choi-Williams of () is distributed as Cu _r(t, f), then according to Cu _rthe instantaneous spectral moment in 2n rank of (t, f) draws 2 rank and 4 rank instantaneous spectrum distances, finally according to the definition of spectrum kurtosis, can try to achieve u _rthe spectrum kurtosis of (n)

The specific practice of above-mentioned calculation perturbation characteristic signal spectrum kurtosis step is:

B1, perturbation features signal u _rn (), calculate its Choi-Williams and distribute, result is Cu _r(t, f).

${\mathrm{Cu}}_r(t,f)=\underset{-\&infin;}{\overset{\&infin;}{\&Integral;}}\sqrt{\frac{\mathrm{\&sigma;}}{4{\mathrm{\&pi;\&tau;}}^2}}{e}^{\left(\frac{{\mathrm{\&sigma;t}}^2}{{4\mathrm{\&tau;}}^2}\right)}{u}_r(t+f+\frac{\mathrm{\&tau;}}{2}){u_r}^{*}(t+f-\frac{\mathrm{\&tau;}}{2})e^{-j2\mathrm{\&pi;f\&tau;}}\mathrm{d\&tau;}---\left(3\right)$

B2, according to Butterworth distribute Cu _rthe 2 rank instantaneous spectrum distances that (t, f) asks for with 4 rank instantaneous spectrum distances

${\hat{S}}_2\left(f\right){=<{\left|{\mathrm{Cu}}_r(t,f)\right|}^2>}_k$

$\left(4\right)$

${\hat{S}}_4\left(f\right)={<{\left|{\mathrm{Cu}}_r(t,f)\right|}^4>}_k$

Wherein, <> _krepresent be k rank time average.

B3, according to spectrum kurtosis definition, try to achieve the spectrum kurtosis based on CWD.

${\hat{K}}_x\left(f\right)=\frac{{\hat{S}}_{4x}\left(f\right)}{{\hat{S}}_{2x}^2\left(f\right)}-2,f\&NotEqual;0---\left(5\right)$

D, according to threshold classification identification

According to the maximal value K of spectrum kurtosis _maxwith the little value K of spectrum kurtosis _minsize, suitable threshold value is set, judge input signal belong to the disturbance of amplitude class, transient state pulse or transient oscillation;

E, effective value method distinguish the disturbance of amplitude class

By the amplitude class class disturbance do not distinguished, calculate its effective value curve, judge it is voltage swell, voltage dip or voltage interruption by the size of amplitude, draw and export required classification results to subsequent processing device.

The algorithm of effective value as shown in the formula:

$U_{\mathrm{RMS}}=\sqrt{\frac{1}{N}\underset{n=k-N+1}{\overset{n=k}{\mathrm{\&Sigma;}}}{u}_i^2}---\left(6\right)$

In order to carry out simulating, verifying, circuit shown in Fig. 2 is utilized to produce each 200 groups of all kinds of disturbing signal at random herein, totally 1000 samples.Sample frequency is 10kHz, and the optimum configurations of various disturbance is as shown in table 1.Classified by the algorithm composing kurtosis and effective value combination based on CWD, classification results is as shown in table 2.

Table 1 transient power quality disturbance setting parameter

The classification results of the single disturbance of table 2

As can be seen from Table 2, categorizing system is very high to the discrimination of single disturbance, discrimination minimum for short interruptions disturbance (discrimination is 97%), analyze known its and be all divided into by mistake and fall disturbance temporarily, mainly because both temporal signatures are very similar, be not easily distinguishable.In addition, individual cases due to perturbation amplitude less, and be subject to noise effect, causing cannot Accurate classification.

In order to make a concrete analysis of the discrimination of the method to compound disturbance, the 6 kinds of compound disturbances combined by these five kinds of disturbances (disturbance of amplitude class can not exist simultaneously, and transient state pulse and transient oscillation also can not exist simultaneously) are analyzed.200 groups, sample is produced respectively at random with PSCAD/EMTDC.Utilize method herein to carry out Classification and Identification, result is as shown in table 3.

Table 3 compound disturbance classification results

Known by table 3, this algorithm is applicable equally to compound disturbance, and classification accuracy rate is higher.Centering dislocation mark is more, this is because interrupt and fall this temporarily just significantly not distinguishing, and boundary is fuzzyyer.

Claims (2)

1., based on the Classification of Transient Power Quality Disturbances recognition methods of CWD (Choi-WilliamsDistribution, Choi-Williams distribute) spectrum kurtosis, identify transient disturbance in electric system, its key step is:

A, extraction perturbation features signal

Voltage signal u (n) of input containing disturbing signal, n=1,2 ..., N, N are data length, and u (n) is carried out wavelet transformation, extract perturbation features signal u _r(n);

B, calculation perturbation characteristic signal spectrum kurtosis

Calculate the perturbation features signal u obtained by A _rthe Choi-Williams distribution Cu of (n) _r(t, f), then according to Cu _rthe instantaneous spectral moment in 2n rank of (t, f) draws 2 rank and 4 rank instantaneous spectrum distances, finally according to the definition of spectrum kurtosis, tries to achieve u _rthe spectrum kurtosis of (n)

C, according to threshold classification identification

The maximal value K of the spectrum kurtosis calculated by B _maxwith the minimum value K of spectrum kurtosis _minsize, suitable threshold value is set, judge input signal belong to the disturbance of amplitude class, transient state pulse or transient oscillation;

D, effective value method distinguish the disturbance of amplitude class

By the amplitude class disturbance do not distinguished, calculate its effective value curve, judge it is voltage swell, voltage dip or voltage interruption by the magnitude range of amplitude, draw and export required classification results to subsequent processing device.

2. compose the Classification of Transient Power Quality Disturbances recognition methods of kurtosis as claimed in claim 1 based on CWD, it is characterized in that: the concrete practice that described B walks calculation perturbation characteristic signal spectrum kurtosis is:

B1, perturbation features signal u _rn (), calculate its Choi-Williams and distribute, result is Cu _r(t, f),

${\mathrm{Cu}}_r(t,f)=\underset{-\mathrm{\&infin;}}{\overset{\mathrm{\&infin;}}{\&Integral;}}\sqrt{\frac{\mathrm{\&sigma;}}{4{\mathrm{\&pi;\&tau;}}^2}}{e}^{(-\frac{{\mathrm{\&sigma;t}}^2}{4{\mathrm{\&tau;}}^2})}{u}_r(t+f+\frac{\mathrm{\&tau;}}{2}){u_r}^{*}(t+f-\frac{\mathrm{\&tau;}}{2})e^{-j2\mathrm{\&pi;}f\mathrm{\&tau;}}d\mathrm{\&tau;}---\left(1\right)$

B2, according to Choi-Williams distribute Cu _rthe 2 rank instantaneous spectrum distances that (t, f) asks for with 4 rank instantaneous spectrum distances

${\hat{S}}_2\left(f\right)=<|{\mathrm{Cu}}_r(t,f){|}^2{>}_k$ (2)

${\hat{S}}_4\left(f\right)=<|{\mathrm{Cu}}_r(t,f){|}^4{>}_k$

Wherein, < > _krepresent be k rank time average; σ represents time scale parameter, represent u _rthe conjugation of (n);

B3, according to spectrum kurtosis definition, try to achieve the spectrum kurtosis based on CWD,

${\hat{K}}_{u_r}\left(f\right)=\frac{{\hat{S}}_4\left(f\right)}{{\hat{S}}_2^2\left(f\right)}-2,f\&NotEqual;0---\left(3\right).$