CN104537227B - Transformer station's noise separation method - Google Patents

Abstract

A kind of transformer station's noise separation method, is related to audio signal processing technique field, and what is solved is three kinds of technical problems of noise of independent measurement high voltage substation.The method obtains six groups of wavelet coefficients to the noise implementation multi-level Wavelet Transform conversion of high voltage substation first；Secondly low-frequency wavelet coefficients are isolated into this bulk noise by doing inverse wavelet transform after passband comb filter；Low-frequency wavelet coefficients are done into inverse wavelet transform by after stopband comb filter with another five groups of wavelet coefficients simultaneously, the mixed signal comprising corona noise and fan noise is obtained；Recycle high-frequency wavelet coefficient detection corona noise, estimate fan noise power spectrum in the time period without corona noise, estimate corona noise power spectrum in the time period for having corona noise, the fan noise and corona noise Power spectrum reconstruction time-domain signal that will finally estimate, so as to isolate fan noise and corona noise.The method that the present invention is provided, is applicable treatment Large-scale High-Pressure alternating current and direct current transformer station noise signal.

Description

Transformer station's noise separation method

Technical field

The present invention relates to audio signal processing technique, more particularly to a kind of technology of transformer station's noise separation method.

Background technology

In electrical power transmission system, under same nominal power condition, transmission voltage is higher, and corresponding transmission current is got over It is small, so that the energy loss of circuit is smaller.Therefore, high pressure, extra high voltage network are current and the weight of future electrical energy construction Point developing direction.Superelevation electric field can be produced to cause air ionization nearby in high voltage power transmisson system, near wire, discharged, Atmospherical discharges can produce " Dth Dth " sound, form corona noise.Simultaneously because voltage cycle changes, the band electrochondria near wire , in the motion of electric field action periodical, the period of motion is identical with ac period, and in China, alternating current work frequency is for son 50Hz.Charged particle collides with air molecule, forms sound wave to external radiation, and this periodicity sound pressure variations constitute body and make an uproar Sound, due to nonlinear effect, this bulk noise also each harmonic signal including work frequency.High voltage power transmission transformer can be produced greatly Calorimetric amount so that become electrical equipment temperature and raise, influences transformer service life, and general high voltage becomes electrical equipment all with cooling facility, main If cooling fan, cooling fan can produce noise, referred to as fan noise.

In Large-scale High-Pressure transformer station, due to transformer, hv transmission line Relatively centralized, the noise in transformer station can be to week Enclose resident's productive life produce harmful effect, according to more than analysis understand high voltage substation in noise mainly have corona noise, Three kinds of this bulk noise and fan noise.Three kinds of noises have different noise sources respectively, and are exist simultaneously and power transformation In standing, the method there is presently no that can realize carrying out independent measurement to three kinds of noises.

The content of the invention

For defect present in above-mentioned prior art, the technical problems to be solved by the invention are to provide one kind can be effective Corona noise, this bulk noise, the fan noise of high voltage substation are separated, so as to realize carrying out independent measurement to three kinds of noises Transformer station's noise separation method.

In order to solve the above-mentioned technical problem, a kind of transformer station's noise separation method provided by the present invention, it is characterised in that Comprise the following steps that：

1) noise of high voltage substation is picked up using microphone, the noise of high voltage substation includes that corona noise, body are made an uproar Sound, fan noise, its signal model is：

S (n)=c (n)+b (n)+e (n)

w₀=2 π f₀/f_s

Wherein, s (n) is high voltage substation noise, and c (n) is corona noise, and b (n) is this bulk noise, and e (n) makes an uproar for fan Sound, w₀It is numeric field frequency, f₀It is body fundamental vibration frequency frequency, f₀=50Hz, f_sIt is signal sampling frequencies, P is harmonic wave number；B_k It is harmonic amplitude, φ_kIt is harmonic phase, n is discrete sampling time index values；

2) implement multi-level Wavelet Transform conversion to s (n), obtain six groups of wavelet coefficients and be respectively：

s_i(n), i=1,2 ... 6；

Wherein, s₁N () is low-frequency wavelet coefficients, s₁N () corresponding frequency band range is [0, f_s/ 64], another five groups of wavelet coefficients s_i(n), i=2 ... 6 corresponding frequency ranges are respectively [f_s/2^8-i,f_s/2^7-i]；

Wherein, the wavelet transform filter group employed in wavelet transformation is Daubechies wave filter groups, wavelet coefficient Calculated using Mallat fast algorithms；

3) by s₁N () obtains the estimation signal of this bulk noise by doing inverse wavelet transform after passband comb filterAnd willUsed as this bulk noise separated, the band connection frequency of passband comb filter is the fundamental frequency of this bulk noise Frequency f₀And each harmonics；

4) by s₁N () does inverse wavelet transform by after stopband comb filter with another five groups of wavelet coefficients, comprising Mixed signal y (n) of corona noise and fan noise；

5) by s₅(n) and s₆N () does inverse wavelet transform, obtain corona detection signal s_d(n)；

6) s is utilized_dN () detection whether has corona noise, and according to testing result, in the time period pair without corona noise Fan noise power spectrum estimated, in the time period for having corona noise by the power spectrum of y (n) and the fan noise power estimated Spectrum subtraction, draws the corona noise power spectrum of estimation, and spectra calculation is realized using Fast Fourier Transform (FFT)；

6.1) s is utilized_dN the step of () detection corona noise, is as follows：

6.1.1) to s_dN (), y (n) implement framing, obtain s_dN the frame signal of each frame is in (), y (n)：

S_q=[s_d(qL-L+1),s_d(qL-L+2),...s_d(qL)]

Y_q=[y (qL-L+1), y (qL-L+2) ... y (qL)]

Wherein, S_qIt is s_dThe frame signal of q frames in (n), Y_qIt is the frame signal of q frames in y (n), L is frame length；

6.1.2 a decision threshold) is set, by s_dN the maximum amplitude standard deviation and decision threshold of each frame are entered in (), y (n) Row compares；

IfThen judge that the q frames in y (n) have corona noise, otherwise then judge that the q frames in y (n) do not have There is corona noise；

Wherein, T_hIt is decision threshold, T_h=10,max{S_qRefer to take S_qMaximum,E{s_d(n) } refer to take s_dThe variance of (n)；

6.2) method of estimation of fan noise power spectrum is as follows：

IfThen makeOtherwise then make

IfThen makeOtherwise then make

If the q frames in y (n) have corona noise, andThenInstead Then

Wherein, the power spectrum of q frames is in y (n) | Y_q(k)|²,It is the minimum spectrum of q frames in y (n),In q frames maximum spectrum,It is the power spectrum of the fan noise q frames of estimation, α, β, γ, κ It is adjustable parameter；

6.3) method of estimation of corona noise power spectrum is as follows：

If the q frames in y (n) have corona noise, makeIt is therein | C_q(k)|² It is the corona noise power spectrum of q frames in y (n)；

7) time-domain signal is reconstructed using Fourier inversion to the fan noise power spectrum estimated, and will be obtained after reconstruct SignalAs the fan noise separated；

Corona noise power spectrum to estimating reconstructs time-domain signal, and the letter that will be obtained after reconstruct using Fourier inversion NumberAs the corona noise separated；

Because spectra calculation is by y (n) sub-frame processings, signal reconstruction is also required to framing reconstruct, and reconstructing method is as follows：

When corona free noise occurs, fan noise frame signal is：

When there is corona noise to occur, the frame signal of fan noise and corona noise is respectively：

Wherein,It is the frame signal of q frames in fan noise reconstruction signal,It is q in corona noise reconstruction signal The frame signal of frame, L is frame length, and IFFT is Fourier inversion.

Further, step 6.2) in, α ∈ [0.01,0.03], β ∈ [0.4,0.65], γ ∈ [4,6], κ ∈ [0.1, 0.3]。

The present invention provide transformer station's noise separation method, using corona noise, this bulk noise, fan noise when, frequency Domain characteristic, the line spectrum characteristic according to this bulk noise designs the separation that corresponding passband comb filter realizes this bulk noise, using electricity Dizzy noise and fan noise when-frequency domain on separability, using wavelet transformation, realize corona noise and fan noise point From, so as to come reach separate these three noises purpose, corona noise, this bulk noise, the wind of high voltage substation can be efficiently separated Fan noise, so as to realize carrying out independent measurement to three kinds of noises.

Brief description of the drawings

Fig. 1 is the separation process figure of transformer station's noise separation method of the embodiment of the present invention；

Fig. 2 is the time domain waveform measurement figure of direct current electrical noise；

Fig. 3 is the time domain waveform measurement figure after direct current electrical noise amplifies in time domain；

Fig. 4 is the time domain waveform measurement figure for exchanging electrical noise；

Fig. 5 is to exchange the time domain waveform measurement figure after electrical noise amplifies in time domain.

Fig. 6 is the separating effect figure after being separated to direct current electrical noise using the method for the embodiment of the present invention；

Fig. 7 is the separating effect figure after being separated to exchange electrical noise using the method for the embodiment of the present invention.

Specific embodiment

Embodiments of the invention are described in further detail below in conjunction with brief description of the drawings, but the present embodiment is not used to limit The system present invention, it is every to use similar structure of the invention and its similar change, protection scope of the present invention all should be listed in.

As shown in figure 1, a kind of transformer station's noise separation method that the embodiment of the present invention is provided, it is characterised in that specific Step is as follows：

1) microphone is utilized, the noise of high voltage substation, the noise bag of high voltage substation is picked up with the sample rate of 65536Hz Corona noise, this bulk noise, fan noise are included, its signal model is：

S (n)=c (n)+b (n)+e (n)

w₀=2 π f₀/f_s

Wherein, s (n) is high voltage substation noise, and c (n) is corona noise, and b (n) is this bulk noise, and e (n) makes an uproar for fan Sound, w₀It is numeric field frequency, f₀It is body fundamental vibration frequency frequency, f₀=50Hz, f_sIt is signal sampling frequencies, P is harmonic wave number, B_k It is harmonic amplitude, φ_kIt is harmonic phase, n is discrete sampling data time index value, typically in 100Hz, 200Hz, 300Hz etc. Body noise energy is larger in harmonic frequency, and corona noise c (n) shows as the time upper random pulse signal for occurring, transmission electricity Pressure is higher, and corona noise is more intensive, while pulse interval is significantly larger than the pulse duration, fan noise e (n) is broadband Noise；

2) implement multi-level Wavelet Transform conversion (being five layers in this example) to s (n), obtain six groups of wavelet coefficients and be respectively：

s_i(n), i=1,2 ... 6；

Wherein, s₁N () is low-frequency wavelet coefficients, s₁N () corresponding frequency band range is [0, f_s/ 64], another five groups of wavelet coefficients s_i(n), i=2 ... 6 corresponding frequency ranges are respectively [f_s/2^8-i,f_s/2^7-i]；

Wherein, the wavelet transform filter group employed in wavelet transformation is many Bei Xi (Daubechies) wave filter group, Wavelet coefficient is calculated using horse traction spy's (Mallat) fast algorithm；

3) by s₁N () obtains the estimation signal of this bulk noise by doing inverse wavelet transform after passband comb filterAnd willAs this bulk noise separated,Sample rate be f_s, the passband of passband comb filter is frequently Rate is the fundamental frequency f of this bulk noise₀And each harmonics；

Wherein, s₁N the sample rate of () is f_s/ 32, passband comb filtering is realized using Fast Fourier Transform (FFT), passband pectination Amplitude response of the wave filter at work frequency (50Hz) and each harmonic frequency is 1, the amplitude response at other frequencies It is 0, can also be using the passband comb filter of other modes in other embodiments of the present invention；

4) by s₁N () does inverse wavelet transform by after stopband comb filter with another five groups of wavelet coefficients, comprising Mixed signal y (n) of corona noise and fan noise, realizes the separation of this bulk noise, and the sample rate of y (n) is f_s；

Wherein, stopband comb filter realizes that stopband comb filter is in work frequency using Fast Fourier Transform (FFT) Amplitude response at (50Hz) and each harmonic frequency is 0, and the amplitude response at other frequencies is 1, other realities of the invention Applying can also be using the stopband comb filter of other modes in example；

5) by s₅(n) and s₆N () does inverse wavelet transform, obtain corona detection signal s_d(n)；

6) s is utilized_dN () detection whether has corona noise, and according to testing result, in the time period pair without corona noise Fan noise power spectrum estimated, in the time period for having corona noise by the power spectrum of y (n) and the fan noise power estimated Spectrum subtraction, draws the corona noise power spectrum of estimation, and spectra calculation is realized using Fast Fourier Transform (FFT)；

s₅(n) and s₆N () is high-frequency wavelet coefficient, fan noise energy can decline in high band, and corona noise energy exists Full frequency band is uniformly distributed, thus corona noise is higher than fan noise in the signal to noise ratio of high band, it is possible to increase corona noise is examined The accuracy rate of survey；

6.1) s is utilized_dN the step of () detection corona noise, is as follows：

6.1.1) to s_dN (), y (n) implement framing, obtain s_dN the frame signal of each frame is in (), y (n)：

S_q=[s_d(qL-L+1),s_d(qL-L+2),...s_d(qL)]

Y_q=[y (qL-L+1), y (qL-L+2) ... y (qL)]

Wherein, S_qIt is s_dThe frame signal of q frames in (n), Y_qIt is the frame signal of q frames in y (n), L is frame length；

6.1.2 a decision threshold) is set, by s_dN the maximum amplitude standard deviation and decision threshold of each frame are entered in (), y (n) Row compares；

IfThen judge that the q frames in y (n) have corona noise, otherwise then judge the q frames in y (n) There is no corona noise；

Wherein, T_hIt is decision threshold, T_h=10,max{S_qRefer to take S_qMaximum,E{s_d(n) } refer to take s_dThe variance of (n)；

6.2) method of estimation of fan noise power spectrum is as follows：

IfThen makeOtherwise then make

IfThen makeOtherwise then make

If the q frames in y (n) have corona noise, andThenOtherwise Then

Wherein, the power spectrum of q frames is in y (n) | Y_q(k)|²,It is the minimum spectrum of q frames in y (n),It is the maximum spectrum of q frames in y (n),It is the power spectrum of the fan noise q frames of estimation, α, β, γ, κ are Adjustable parameter, α ∈ [0.01,0.03], β ∈ [0.4,0.65], γ ∈ [4,6], κ ∈ [0.1,0.3]；

When q frames in y (n) have corona noise to occur,Increase rapidly on corona noise spectral range, andIt is slow to rise so thatSet up, the now estimation to fan noise does not update, because fan is made an uproar Sound has the stationarity of long period, and when having corona noise to occur, the fan noise power spectrum of estimation remains to represent real wind Fan noise power spectrum；

6.3) method of estimation of corona noise power spectrum is as follows：

Assuming that fan noise is uncorrelated to corona noise, if the q frames in y (n) have corona noise, makeIt is therein | C_q(k)|²It is the corona noise power spectrum of q frames in y (n)；

7) time-domain signal is reconstructed using Fourier inversion to the fan noise power spectrum estimated, and will be obtained after reconstruct SignalAs the fan noise separated；

Corona noise power spectrum to estimating reconstructs time-domain signal, and the letter that will be obtained after reconstruct using Fourier inversion NumberAs the corona noise separated；

Because spectra calculation is by y (n) sub-frame processings, signal reconstruction is also required to framing reconstruct, and reconstructing method is as follows：

When corona free noise occurs, fan noise frame signal is：

When there is corona noise to occur, the frame signal of fan noise and corona noise is respectively：

Wherein,It is the frame signal of q frames in fan noise reconstruction signal,It is q in corona noise reconstruction signal The frame signal of frame, L is frame length, and IFFT is Fourier inversion.

Fig. 2 for direct current electrical noise time domain waveform measurement figure, Fig. 4 be exchange electrical noise time domain waveform measurement figure, Fig. 2, Axis of abscissas in Fig. 4 is time shaft, and axis of ordinates is amplitude axis, can be seen that direct current electrical noise and hands over from Fig. 2, Fig. 4 Stream electrical noise tool is very different, and galvanic corona noise occurs at random on a timeline, and the corona noise of alternating current Cluster is formed, alternating current occurs multiple corona pulse waveforms, the cluster cycle of the corona noise formation of alternating current when having corona noise Occur, its cycle time domain work frequency cycle phase is same；

Fig. 3 is that the time domain waveform after direct current electrical noise amplifies in time domain measures figure, and Fig. 5 is exchange electrical noise in time domain Time domain waveform measurement figure after amplification, the axis of abscissas in Fig. 3, Fig. 5 is time shaft, and axis of ordinates is amplitude axis, from figure 3rd, Fig. 5 can be seen that direct current due in the absence of periodically variable electric field, the time period without corona noise show as compared with It is stable fan noise, and alternating current is in the time period without corona noise, noise has stronger periodicity, and this mainly comes Come from each harmonic signal in this bulk noise；

From Fig. 2, Fig. 4 it can also be seen that direct current electrical noise also has identical point with electrical noise is exchanged, i.e.,：Made an uproar there is corona When sound occurs, the amplitude of noise signal becomes big；This is some pulse trains mainly due to corona noise, and amplitude is larger, corona The spectral range of noise is wide, is distributed in efficiently sampling frequency range self-energy；

The embodiment of the present invention using corona noise, this bulk noise, fan noise when, frequency domain characteristic come reach separate this Three kinds of purposes of noise；Wherein, corona noise due to the duration it is short, signal amplitude is big, pulse characteristic is shown, in frequency The characteristics of bandwidth has without limit for width；This bulk noise has line spectrum structure, larger in higher frequency scope line spectrum energy attenuation, can be with Do not consider；Fan noise energy increases with frequency, and energy is reduced.

The embodiment of the present invention is applied to treatment Large-scale High-Pressure alternating current and direct current transformer station noise signal, can efficiently separate height Three kinds of noises in buckling power station；

Fig. 6 is the separating effect figure after being separated to direct current electrical noise using the method for the embodiment of the present invention, in Fig. 6 S (n) is original direct current electrical noise, and c (n) is corona noise, and b (n) is this bulk noise, and e (n) is fan noise；

Fig. 7 is the separating effect figure after being separated to exchange electrical noise using the method for the embodiment of the present invention, in Fig. 7 S (n) is original exchange electrical noise, and c (n) is corona noise, and b (n) is this bulk noise, and e (n) is fan noise.

Claims (2)

1. a kind of transformer station's noise separation method, it is characterised in that comprise the following steps that：

1) noise of high voltage substation is picked up using microphone, the noise of high voltage substation includes corona noise, this bulk noise, wind Noise is fanned, its signal model is：

S (n)=c (n)+b (n)+e (n)

$b\left(n\right)=\underset{k=1}{\overset{P}{\mathrm{\Σ}}}{B}_k\cos (k{w}_{0}n+{\mathrm{\φ}}_k)$

w₀=2 π f₀/f_s

Wherein, s (n) is high voltage substation noise, and c (n) is corona noise, and b (n) is this bulk noise, and e (n) is fan noise, w₀ It is numeric field frequency, f₀It is body fundamental vibration frequency frequency, f₀=50Hz, f_sIt is signal sampling frequencies, P is harmonic wave number；B_kFor humorous Wave amplitude, φ_kIt is harmonic phase, n is discrete sampling time index values；

2) implement multi-level Wavelet Transform conversion to s (n), obtain six groups of wavelet coefficients and be respectively：

s_i(n), i=1,2 ... 6；

Wherein, s₁N () is low-frequency wavelet coefficients, s₁N () corresponding frequency band range is [0, f_s/ 64], another five groups of wavelet coefficient s_i (n), i=2 ... 6 corresponding frequency ranges are respectively [f_s/2^8-i,f_s/2^7-i]；

Wherein, the wavelet transform filter group employed in wavelet transformation is Daubechies wave filter groups, and wavelet coefficient is used Mallat fast algorithms are calculated；

3) by s₁N () obtains the estimation signal of this bulk noise by doing inverse wavelet transform after passband comb filterAnd WillUsed as this bulk noise separated, the band connection frequency of passband comb filter is the fundamental frequency f of this bulk noise₀With And each harmonics；

4) by s₁N () does inverse wavelet transform by after stopband comb filter with another five groups of wavelet coefficients, obtain being made an uproar comprising corona Mixed signal y (n) of sound and fan noise；

5) by s₅(n) and s₆N () does inverse wavelet transform, obtain corona detection signal s_d(n)；

6) s is utilized_dN () detection whether has corona noise, and according to testing result, in the time period without corona noise to fan Noise power spectrum estimated, in the time period for having corona noise by the power spectrum of y (n) and the fan noise power spectrum phase estimated Subtract, draw the corona noise power spectrum of estimation, spectra calculation is realized using Fast Fourier Transform (FFT)；

6.1) s is utilized_dN the step of () detection corona noise, is as follows：

6.1.1) to s_dN (), y (n) implement framing, obtain s_dN the frame signal of each frame is in (), y (n)：

S_q=[s_d(qL-L+1),s_d(qL-L+2),...s_d(qL)]

Y_q=[y (qL-L+1), y (qL-L+2) ... y (qL)]

Wherein, S_qIt is s_dThe frame signal of q frames in (n), Y_qIt is the frame signal of q frames in y (n), L is frame length；

6.1.2 a decision threshold) is set, by s_dN the maximum amplitude standard deviation and decision threshold of each frame are compared in (), y (n) Compared with；

IfThen judge that the q frames in y (n) have corona noise, otherwise then judge that the q frames in y (n) do not have Corona noise；

Wherein, T_hIt is decision threshold, T_h=10,max{S_qRefer to take S_qMaximum, ${\mathrm{\σ}}_{s_d}^2=E\left\{{(s_d\left(n\right)-E\{s_d\left(n\right)\left\}\right)}^2\right\},$ E{s_d(n) } refer to take s_dThe variance of (n)；

6.2) method of estimation of fan noise power spectrum is as follows：

IfThen makeOtherwise then make

IfThen makeOtherwise then make

If the q frames in y (n) have corona noise, andThen ${\hat{E}}_q\left(k\right)={\hat{E}}_{q-1}\left(k\right),$ Otherwise then ${\hat{E}}_q\left(k\right)={\hat{E}}_{q-1}\left(k\right)+\mathrm{\κ}\·\left(\right|Y_q\left(k\right)|-{\hat{E}}_{q-1}\left(k\right));$

Wherein, the power spectrum of q frames is in y (n) | Y_q(k)|²,It is the minimum spectrum of q frames in y (n),For The maximum spectrum of q frames in y (n),It is the power spectrum of the fan noise q frames of estimation, α, β, γ, κ are adjustable ginseng Number；

6.3) method of estimation of corona noise power spectrum is as follows：

If the q frames in y (n) have corona noise, makeIt is therein | C_q(k)|²It is y The corona noise power spectrum of q frames in (n)；

7) time-domain signal, and the signal that will be obtained after reconstruct are reconstructed using Fourier inversion to the fan noise power spectrum estimatedAs the fan noise separated；

Corona noise power spectrum to estimating reconstructs time-domain signal, and the signal that will be obtained after reconstruct using Fourier inversionAs the corona noise separated；

Because spectra calculation is by y (n) sub-frame processings, signal reconstruction is also required to framing reconstruct, and reconstructing method is as follows：

When corona free noise occurs, fan noise frame signal is：

${\hat{E}}_q=[\hat{e}(\mathrm{qL}-L+1),\hat{e}(\mathrm{qL}-L+2),...\hat{e}\left(\mathrm{qL}\right)]=Y_q$

When there is corona noise to occur, the frame signal of fan noise and corona noise is respectively：

$\left\{\begin{array}{c}{\hat{E}}_q=\mathrm{IFFT}\left\{\right|{\hat{E}}_q\left(k\right)\left|e^{j\∠Y_q\left(k\right)}\right\}\\ {\hat{C}}_q=\mathrm{IFFT}\left\{\right|{\hat{C}}_q\left(k\right)\left|e^{j\∠Y_q\left(k\right)}\right\}\end{array}\right.$

Wherein,It is the frame signal of q frames in fan noise reconstruction signal,It is q frames in corona noise reconstruction signal Frame signal, L is frame length, and IFFT is Fourier inversion.

2. transformer station's noise separation method according to claim 1, it is characterised in that：Step 6.2) in, α ∈ [0.01, 0.03], β ∈ [0.4,0.65], γ ∈ [4,6], κ ∈ [0.1,0.3].