Wavelet Transform with Noise-SuppressionMethod for Detection and Improvement of

Power Quality Disturbances 1K.Vijay Kumar

1Associate Professor and H.O.D Department of EEE, Dadi Institute of Engineering and Technology,Visakhapatnam

1Email : hodeee@dietakp.com

Abstract— Recently power quality has become oneof the most important issues in modern powerindustry. The origin of events which affectquality of power is mainly electromagnetictransients, harmonic distortion in addition tovoltage sag, swell, flicker, and other powerquality disturbances that proliferate inmodern electric power grids. The WT isoffering a large variety of potential bases,where optimization can be further performed.In the WT applications the process oftenconfuses the transient signals and the noisesride on the signals. Consequently, thethreshold is difficult to give in detectingthe existence of transient signals. Toeliminate the difficulty of distinguishing thesignals from the noises existing in theWavelet Transform Coefficients (WTCs) of thesignals, a denoising algorithm is presented tosuppress the noise riding on the signals. Tofurther improve the disturbance detection rateby more effectively discriminating the signalsfrom the noises after the DWT, a spatial-correlation-based noise-suppression algorithmis proposed in this paper.

This work describes the techniques ofcorrecting the supply voltage sag, swell in adistributed system. At present, a wide rangeof very flexible controllers, which capitalizeon newly available power electronicscomponents, are emerging for custom powerapplications. Among these, the distributionstatic compensator is the most effectivedevices. A D-STATCOM injects a current intothe system to correct the voltage sag, swell.Comprehensive results are presented to assessthe performance of device as a potentialcustom power solution.

Keywords—Noise measurement, power quality,wavelet transform, D-STATCOM , Voltage dips,swells, interruption, and power systemmonitoring.

I. INTRODUCTION In the deregulated powermarket, the quality of power has a directeconomic impact on utilities andindustrial customers. With the increasein the use of more modern equipment and

more automation, there is a growingconcern about power quality (PQ) and itsautomatic monitoring and analysis hasbecome an important challenging issue forpower engineers. To upgrade the PQ,monitoring the power signals in a widerange and a long run is evitable to findout the real causes of the PQ problem.While monitoring fault recorder shouldcapture only the waveform cycles coveringthe disturbance in order to reduce thevolume of data to be recorded andanalyzed. Therefore, only the portion ofthe

signals covering the disturbances shouldbe detected and saved for furtherinvestigation of the PQ problems. In general, there arethree steps to address the PQ problems,which are detection, localization, andclassification. The frequency and timedomain information is needed to

complete the three steps. To analyze thePQ, feature extraction is the key point.The extraction of the features of thedisturbances requires its time domainduration estimation, suitabletransformation, and observation of thesignals after transformation. Varioustransformation methods have been proposedfor detecting and localizing the PQevents, which include the discreteFourier transforms (DFTs), the short-timeFourier transform (STFT) and the wavelettransform (WT). Traditionally, byextracting the frequency contents of therecorded signals, the DFT has been usedfor the steady-state analysis of thepower harmonics. According to thefrequency contents of the signals, someof the PQ problems related to harmonicscan be detected. Nevertheless, with itsconstant bandwidth, the DFT is notsuitable for the detection of the shortterm transient signals. Moreover, the

time-evolving effects of the frequency innonstationary signals are not consideredin the DFT techniques. If the amplitude of aharmonic signal fluctuates with time, theFT cannot be used without modification. Amethod commonly used is to window thesignal into a sequence of sufficientlysmall that the waveform approximates astationary waveform. This approach isknown as a short-time FT (STFT). Thisapproach has the effect of giving sometime resolution to the measurement at theexpense of frequency resolution. Thelimitation of a fixed window width, infact, is inadequate for the analysis ofthe transient nonstationary signals. TheSTFT will work well provided the windowis short enough compared with thefluctuation rate. High rates offluctuation can give rise to significanterrors. Wavelet transforms (WTs)have emerged as fast and effective toolsfor automated detection and moreefficient characterization of the PQdisturbances, in power system protection.Discrete wavelet transform (DWT) basedmulti resolution analysis (MRA), which isalso referred to as multi resolutionsignal decomposition (MSD) in PQliterature, analyzes the signal atdifferent frequencies with differentresolutions. Since the WT is offering alarge variety of potential bases, whereoptimization can be further performed.Multiresolutions of both time andfrequency are prepared in the WTapproach: a greater resolution in time isprovided for the high-frequencycomponents of a signal, and a greaterresolution in frequency is provided forthe low-frequency components. With themultiresolution characteristics, the WTis considered very adequate for theanalysis of the power system transientsdue to various disturbances.

The presence of noise notonly degrades the detection capability ofwavelet and other higher time resolution-based PQ monitoring systems but alsohinders the recovery of importantinformation from the captured waveformfor time localization and classificationof the disturbances. In actualapplications, the process often confusesthe transient signals and the noisesriding on the signals. Consequently, thethreshold is difficult to give indetecting the existence of transientsignals. To further improve thedisturbance detection rate by moreeffectively discriminating the signalsfrom the noises after the DWT, a spatial-correlation-based noise-suppressionalgorithm is proposed in this paper. Bymaking use of the spatially selectivenoise-filtration technique, the proposedalgorithm calculates the correspondingcorrelation coefficients from the WTCs ofthe signals at several adjacent scales.The time points with higher correlatedWTCs at the adjacent scales indicate thefact that a significant signal variationor a transient event exists. Accordingly,the approach can avoid the effects of thenoises on the detection of the transientsignals.

Utilities often focus ondisturbances from end-user equipment asthe main power quality problems. This iscorrect for many disturbances, flicker,harmonics, etc., but voltage dips mainlyhave their origin in the higher voltagelevels. If the economical losses due tovoltage dips are significant, mitigationactions can be profitable for thecustomer. There are different ways tomitigate voltage dips, swell intransmission and distribution systems. Atpresent, a wide range of very flexiblecontrollers, which capitalize on newlyavailable power electronics components,are emerging for custom power

applications. Among these, thedistribution static compensator is mosteffective devices, which is based on theVSC principle. A new PWM-based controlscheme has been implemented to controlthe electronic valves in the two-levelVSC used in the D-STATCOM.

II. WAVELET EXPLANATION Due to its ability toextract time and frequency information ofsignal simultaneously, the WT is anattractive technique for analyzing PQwaveform. WT can be continuous ordiscrete. Discrete WT (DWT) can be viewedas a subset of Continuous WT (CWT). Basedon the characteristics of the bandpassfilters, the WT has been proposed toinvestigate the transient phenomena ofthe power signals from different scalesof the WTCs. In practical applications,the DWT is commonly used. The DWT usesthe low-pass h(k) and the high-pass g(k)filters to divide the frequency-band ofthe input signal f(k) in respective low-and high-frequency components into octavebands. The low-pass filter h(k) isdetermined from the scaling function. Thehigh-pass filter g(k) is determined fromboth the wavelet and scaling functions.The wavelet and scaling functions are,respectively, given as,

(1) where n is integers and represent thenumber of samples. While the low-passfiltering produces the approximations Aj,the high-pass filtering produces thedetails Dj of the decomposition. Therelationship of the approximationcoefficients and detail coefficientsbetween two adjacent levels are given as,

(2)

where j is frequency band level. In particular, the transientproblems with high-frequency componentscan be detected and located through anadequate scale of WTCs. The WMRA is basedon decomposition of the original signalinto different signals at various levelsof resolution. In essence, it isrepresented by one set of scalingcoefficients, and one or several sets ofwavelet coefficients,

(3)

There are many wavelet functions named asmother wavelets. The choice of motherwavelet is important because differenttypes of mother wavelets have differentproperties. Several popular waveletfunctions are Haar, Morlet, Coiflet,Symlet and Daubechies wavelets. A dyadic DWT of discrete time sequencef(k) of length N is essentially adecomposition of the spectrum of f(k);F(w) into orthogonal subbands defined by

m = 1,2……J

Where T is the sampling period associatedwith f(k) and represents the totalnumber of resolution levels.

Let ψ(t) be a mother wavelet, and its FT can be shown as

(4)

After the operation of dilation and translation, (4) can be written as a babywavelet shown in the following:

(5)

Fig. 1. Mother wavelet D4.

where a is a scaling factor and b is atranslating factor. The FT of (5) can bedescribed in

= (6)The continuous WT (CWT) of a time-continuous signal xt is defined as

Fxtcwt(a, b) = |a|−1/2

(7)where Fxtcwt(a, b) is the CWT of signal xt withwavelet ψ(t)and parameters a and b. The asterisk in(7) denotes a complex conjugateoperation. In this paper, theDaubechies’s wavelets with a four-coefficient filter (D4 wavelet) are usedas a mother wavelet. Shown in Fig. 1 isan example scaling function generatedusing the D4 filter. For a mother waveletψ(t) given in (7), different scaling and

translating factors can lead to scales offrequency bands at levels of resolutions. In practical applications, thecontinuous wavelet function can betransferred to a discrete form via asampling way. In (8), the factors a and bare replaced by am

0 and nam0b0,

respectively, where m, n ∈ Z and Z is a setof integers. Then, (8) can be writteninto a general discrete formψmn(t) = a0

−m)2ψ (a0m− t nb− 0), a0 > 1, b0 ≠� 0.

(8)

The DWT of the sampled signals xn isemployed to replacethe CWT operation of xt in (4)

DWTx(m, n) =

(9)

where k is an operating index; m is ascaling number; n is asampling time point, n = 1, 2, . . .,N; and N isthe number of sampling points. Thefactors a and b in (7) are expressed usinginteger parameters m and n through a = 2mand b = n (thereby, a dyadictransformation), respectively, in theDWT. Based on (9), the DWT of xn, n = 1,2, . . .,N, proceeds successively with variousbandpass filtering effects. At the firststage, the time-stamped data (m = 1 forscale 1) out of the filtering operationextract the highest frequency componentsfrom the signal. Moreover, the outputdata for the scale 2 WTCs denote thesecond higher frequency band. Theprocedure continues until the highestscale (m = M) is reached. As described earlier,because diverse frequency components,especially higher frequency, often existin the signals of disturbance events,thus examining scales of the WTCs wouldhelp determine the occurrence of thedisturbance events as well as theiroccurring time. In determining the

occurrence of a disturbance event,thresholding the WTCs is one of the mostcommon approaches adopted in theliterature. The WTCs that are larger thanthe given threshold values indicate thestart or end points of the disturbanceevents. Counting on this, the DWTtechniques have been widely used toanalyze the disturbance events in thepower systems.

III.NOISE EXPLANATIONA. SNR Ratio To test the effectiveness ofthe proposed technique, actual andsimulated PQ disturbances have beenconsidered. The simulated PQ disturbancesinclude pure sine wave, voltage sag,voltage swell, oscillatory transients andsustained interruptions. All simulatedwaveforms are obtained for a powerfrequency, and are normalized with anormal sinusoidal voltage without anydisturbance before they are processedthrough the proposed denoising module fortesting. An another important issuerelated to the proposed method is noisewhich is generally omnipresent in anelectrical power distribution network.Therefore, the performance of proposedalgorithm was tested in different noiseenvironments. Gaussian white noise iswidely used in the researches on the PQissues. The value of SNR is defined asfollows: SNR = 10 log (PS/PN)dB

where Ps is the power (variance) of thesignal and Pn is that of the noise. Thenoise was added with three phase signalsof the events. Event signals were testedby adding 35, 40, 45 and 50 dB of noise.

B. Noise Suppression Algorithm Although the DWT exhibits itsgreat power in the detection andlocalization of the disturbance, itsability is often degraded due to thenoises riding on the signal. To overcomethe difficulties of capturing thedisturbances out of the background noisesin a low-SNR environment, a noise-suppression algorithm is proposed to beintegrated with the DWT described earlierThe steps of the noise-suppressionalgorithm adapted from are described asfollows.

Step 1: Calculate the correlations WTRk(m, n) of adjacentScales

WTRk(m, n) = DWTx(m, n) •DWTx(m + 1, n) • . . . •DWTx(m + k − 1, n) =

(10) s.t. m < M – k + 1where k is the number of scaleintermultiplication andM is the number ofall scales. The effectiveness of the stepis to augment the disturbance phenomenain the scales of the WT. The correlationfunction WTR2 = DWTx(1, n) ∗DWTx(2, n), (k =2), is enhanced at the pointscorresponding to the main WTC pulses ofthe disturbances, but simultaneously, itis suppressed at the points correspondingto the background noises of the signals.Nevertheless, the background noises stillexist in the current WTR2 functionsobtained in the step. In the next step, ascheme is proposed to further strengthenthe correlated features of thedisturbances to distinguish thesignificant disturbance signals from thenoises.

Step 2: Calculate the new values of WTRnew,k(m, n)

where PWTR (m) = ∑nWTR2k(m, n) and

PDWT (m) =∑nDWTx2(m, n).

(11)

WTRnew,k is adaptively strengthened at thesharp pulses of the mth scale WTCsthrough the WTRk and the factor of

.

Step 3: Extract the disturbances from the noise-riding signals at scale m through the following procedure: MASK(m, n) = 0; For m = 1 : M For n = 1 : N If |WTRnew,k(m, n)| > |DWTx(m, n)| Then{WTRnew,k(m, n) = 0; DWTx(m, n) = 0; MASK(m, n) = 1; } End End. (12) As described earlier, the main pulsesare extracted by comparing the absolutevalues of WTR2(1, n) and DWTx(1, n) bysetting k = 2 and m = 1. The pulses arenow masked with the symbol MASK(1, n) = 1.At fine scales, except for disturbances,WTCs only have noises. By comparing thevalues of |WTRnew,k(m, n)| and |DWTx(m, n)|directly, many noises may be extractedas significant disturbances. To avoidthis situation, |DWTx(m, n)| canmultiply by a weight of λ(m), λ(m) ≥ 1.Through this scheme, only when |WTRnew,k(m,n)| λ≥ (m)|DWTx(m, n)| is to come intoexistence, DWTx(m, n) is just regarded asthe main PQ disturbance features. In asimulated test, setting the weights λ(m)equal to [1.15, 1.06, 1, 1, . . ., 1], theaccuracy of the PQ monitoring results are

reasonable. The results will be discussedin the next section.

Step 4: Create a new DWTxnew(m, n) of allscales

If PDWT (m) < THR(m) Then DWTxnew(m, n) = MASK(m, n) ∗DWTx(m, n) Else DWTx(m, n) = DWTxnew(m, n), go to Step 1 (13) Where THR(m) is a preset threshold value for scale m.For instance MASK(1, n) and DWTx,new(1, n)after the first run are obtained toextract the significant disturbancesignals by referring to WTR2(1, n) andDWTx(1, n). This procedure of signalcompression and disturbance informationextraction is iterated until the sum ofthe power of the unextracted points atDWTx(1, n) is less than some referencenoise power at the first scale.

Step 5: Stopping condition. Repeat theprocedure until all of the scales havebeen calculated. The flowchart of theproposed noise-suppression algorithm isshown in Fig. 2. However, in theprocedure described earlier, the averagepower THR(m) of the noises at each scaleis not given in Step 4. The average powerof the noises can be used as a thresholdby estimation as stated in the nextsection.

Estimating the Average Power of the Noises

It is well known that thestandard deviation σ of a sequence X ∼ N(0,σ2) can be estimated as , where

PDWT = n x(n)2 and N is the sampling points.Assume that K points had been extractedfrom the sequence X, and then, thediscrete wavelet coefficients DWT (m, n)can be written as DWT' (m, n). In the

meantime, DWT' (m, n) can be regarded asall produced from noises. Currently, thevariance σ2

m can be estimated as P'DWT (m)/(N K− ), and hence, the average power of the

noises is (N K− ) σ2m.

In fact, DWT' (m, n) = DWTs ' (m,n) + DWTn' (m, n),where DWTs' (m, n) and DWT' n (m, n) representthe WTCsof the power signals and noises,respectively. Therefore,P' DWT (m) can be written as follows:

P' DWT (m) = (N K− ) · E {DWT' (m, n)2}

= (N K− ) · E {DWT' s (m,

n)2 +DWTn' (m, n)2

+2DWTs ' (m, n)DWTn' (m, n)}

= (N K− ) · E {DWT' s(m,

n)2}+ (N K− ) · σ2m..

(14) E {DWTs' (m, n)2} is alwayspositive. However, (N K− )σ2

m cannot all betreated as noises in the coarse scales.To overcome this problem, (N K− )σ2

m canmultiply a factor th(m), th(m) > 1.Nevertheless, E {DWTs ' (m, n)2} changes withvarious power signals, so th(m) is avariation. Luckily, setting the thresholdvalue is not sensitive to extracting thefeatures from the WTCs when the powersystem occurs in the PQ events.

Fig. 2. Flowchart of the proposed noise-suppression algorithm.

It is not hard to realizethat, as E{DWT's (m, n)2} increases, then, σ2

m

decreases with scale. Hence, σ2m is

donating P'DWT (m) in fine scale and viceversa. The th(m) can be given as follows:th(1) = 1.1−1.2 th(2) = 1.2−1.4 th(3) = 1.4−1.6

When m > 3, th(m) = 1.6−1.8.

The threshold THR(m) in Step 4 can beadjusted to be THR(m) = (N K− )σ2

m∗ th(m).Due to noise donating at fine scales, σ2

m

can be estimated from the first twoscales of the DWT coefficients. If |WTRnew,2(1, n)| > λ(1)|DWT2(1,n)|, then set thecorresponding n points in DWT2(1, n) to bezero. Therefore, assume that k points havebeen removed from the sampling points,and can be regarded as thatproduced by noises. Hence, the power of

Input

DWT

Scale1

Scale2

Strengthen thefeatures

Pure features atScale 1 without

Is condition (12) true

Is condition (13) true

Extract the features

Yes

Yes

No

No

Step 1

Step 2

Step 3

Step 4

Create New Scales1 and 2

can be written as DWT (1), and the

variance is σ21 = DWT (1) / (N K− ). In a

similar way, the variances of all DWTscales can be shown to be the same withscale 1, i.e.,σ2

1 = σ22 = · · · = σ2

M.

IV. PROPOSED SYSTEM.A. Explanation of Layout

Fig.3. Diagram of the circuit for thesimulations

The proposed system is as shownin fig.3. The circuit contains powergenerating station, transmission line,circuit breakers (for sag and swell) andgenerator with cyclo convertor. Thedisturbances which are to be analysed arepresented in system caused by faults,sudden change in steady state conditionand malfunction of protective device. Simulated 10% voltage-sagsignal caused by the single-line groundfault (SLGF) that occurred on phase A andmonitored on phase B for 0.1 s, and then,the system recovered to be normal again.The signal simulating the voltage-swelldisturbance on a 3.3-kV bus and thesignal with the peak value is larger thanthe normal, which is caused by the SLGFon phase A and monitored on phase B asshown in circuit.

A transient disturbance is asudden nonpower frequency change in thesteady-state condition of the voltage,current, or both. Lighting or capacitorenergization usually causes the impulsiveor oscillatory transients, whose durationlasts for a few microseconds tomilliseconds. The impulses riding on thepower-frequency signal due to capacitorswitching are simulated at the capacitorbank as shown in circuit. A sustained interruption isdefined as that when the supply voltagedecreases to less than 10% of the normalfor a duration exceeding 1 min. Theinterruption disturbance is usuallycaused by a fault in the power systems,failure of the equipment, or malfunctionof the protective device. Simulatedsustained interruption signal, with thevoltage dropping to zero due to a faultin the generator. As explained in previoussections DWT is applied to thesedisturbance signals. Scales 1 and 2 ofthe WTCs are occurred. In order toimprove the performance of WT techniquein processing the noise riding signals,Spatial noise-suppression algorithmintegrates with DWT as explained insection III.B (Noise SuppressionAlgorithm).Using (10)–(11) of the noise-suppressionscheme, the WTR2occurred. However, it still remainsdifficult, as done by the computer, whenthe WTC magnitudes of the disturbance andthe noises are variable. Through thenoise-suppression scheme proposed in thispaper, the WTCs are reduced. The resultof the noise suppression scheme iscleaning up the noises. However, it still remainsdifficult, as done by setting a suitablethreshold to remove noises in differentvoltage levels, when the WTC magnitudesof the disturbance and the noises arevariable. As we know, it is not a good

idea to set a fix threshold in differentvoltage levels, but according todifferent levels, to put variousthresholds is impracticable in real powersystems. The noise-suppression schemeproposed in this paper has the ability toadjust an appropriate thresholdautomatically according to differentvoltage levels.

B. Simulink model

Fig.4. Simulink model for proposedcircuit diagram

The simulink model forproposed circuit is as shown in fig.4.The disturbance signals from circuit aresaved for analysis. These signals containnoise of 45db which is added foranalysis. Application of DWT and noise-suppression algorithm is as explainedearlier. Scale 1, scale2, Correlation ofscales and detection of disturbances aregiven below.

C. Wave forms Shown in Fig.5 the event with sag

of the voltage waveform at about 0.1 s is

related to sampling points. The recordedsignal was then transformed into scales 1and 2 through the DWT. Through (10)–(13),the results are shownas correlation in fig.5. Using the noise-suppression algorithm, the WTCs at thefirst scale after trimming the noises offareshown as disturbance in signal whichpinpoints out clearly the occurring timeof the voltage sag by a single impulse ofthe WTCs.

Fig.5. Noise-suppression process for thesimulated sag in relation to sampling

points.

Detection of swell using suppressionalgorithm is as shown in fig.6. Voltagewaveform is related to sampling pointscorresponds to time. The WTCs pinpointouts the occurrence of swell atparticular sampling points.

Fig.6. Noise-suppression process for thesimulated swell in relation to sampling

points. The result of the noise

suppression scheme in cleaning up thenoises is shown in fig.7, fig.8 foroscillatory transients and sustainedinterruption respectively. The WTCs againprovide a concise for the occurrence ofthe impulses.

Fig.7. Noise-suppression process for thesimulated oscillatory transient in

relation to sampling points.

Fig.8. Noise-suppression process for thesimulated sustained interruption in

relation to sampling points.

Table 1 shows the WTCs ofscale1after processing through

suppression algorithm. For eachindividual disturbance respectivesampling points(corresponding to time),detailed coefficient of scale 1 afterDWT, correlation values of scale 1andscale 2, denoised coefficients (which areindicates the occurrence of disturbancein scale 1) are tabulated at noise levelof 45 dB. As per the (12) in step 3correlation values become zero whencondition obeys and scale 1 coefficientvalues are masked with symbolMASK(1,n)=1. The remaining coefficientsare treated as noise in the signal andthose are nullified. The iterativeprocess had been taken for accurateidentification of disturbances. In table 1 coefficients ofdisturbance occurred are detected at4000th sampling point for sag and swell,which are shown at 2000th sampling point infigures 5 and 6 respectively. That means that for applying denoisingalgorithm the data of signal is takenbetween 2000-10000 sampling points whichcovers the disturbance. Same for

Oscillatory transient and Sustainedinterruption also.

Disturbance

Sampling

points

Detailed coefficients

of scale1

Correlation

Denoised coefficients

Sag

2000-

792.55002

-1075.577

40

3000-

286.377537

14209.2428 0

3999-

6583.9436

0-

5806.82230

4000 12844.27794 0 12678.95

048

6000 105.7211754

50733.6139 0

8000 150.789336

-151232.8

180

10000-

42.1436001

236491.3774 0

Swell

2000-

16.7719396

-13.28138

020

3000 1.24540639

27.88747842 0

3999 36.85370276 0 40.77888

652

4000-

80.2463140

0-

84.3350288

6000 1.62886086

-6.092938

510

8000-

7.06424946

3.754759185 0

10000-

3.17737272

-110.5082

870

Oscillatory

Transient

2000-

34.0850897

-26713.05

820

3000-

15.0159288

-7.631281

380

4000-

15.9591444

23.32750893 0

5999 -472.9717

50

0 -496.8255

16

6000 1009.602559

181744.1118 0

8000 7.395846894

23.85089418 0

10000-

8.93368705

-19.25991

520

Sustained

Interrupt

2000-

276.255806

-276.2558

060

3000-

139.169444

15968.87264 0

4000 87.1354346

-12637.63

870

5999-

6375.72658

0-

6375.72658

6000 13180.74197 0 13180.74

197

8000 73.12142263

1778.254734 0

10000-

149.954373

-52349.77

040

Table 1. Denoised coefficients ofdisturbance signals at noise of 45 dB.

Table 2 shows the WTCs of scale1For each individual disturbance afterprocessing through suppression algorithm.Denoised coefficients (which areindicates the occurrence of disturbancein scale 1) are tabulated at noise levelof 50 dB, 40 dB, 35 dB. As noted, the tolerance to noise

declines as the SNR becomes smaller. Thetable also reveals that the tolerance tonoise of the PQ monitoring system isheavily related to the types ofdisturbances. For the transient andinterruption disturbances, the system canhave a 100% detection rate for the SNRthat is as low as 35 dB. Nevertheless,for that SNR value, the detection rate ofthe sag disturbances still has 98% andfor swell detection rate is 99%.Generally, in all of the testing cases,the PQ monitoring system with the noise-suppression scheme shows the essentialcapability of tolerating the potential

noises contained on the signals underinvestigation.

Disturbance

Sampling

points

Denoisedcoeffici

ents for 50

dB

Denoised coeffici

ents for 40dB

Denoised coefficients for 35

dB

Sag

2000 0 0 03000 0 0 0

3999-

4856.5341

-2992.160

07

-7278.353

43

4000 9608.97932

9177.87545

11341.6228

6000 0 0 08000 0 0 010000 0 0 0

Swell

2000 0 0 03000 0 0 0

3999 63.4031867

65.57753218

188.8896637

4000-

110.01984

-151.1878

83

-114.7289

726000 0 0 08000 0 0 010000 0 0 0

Oscillatory

Transient

2000 0 0 03000 0 0 04000 0 0 0

5999-

480.35089

-487.2611

59

-460.3019

52

6000 0 1012.38880

1004.58276

8000 0 0 010000 0 0 0

Sustained

Interrupt

2000 0 0 03000 0 0 04000 0 0 0

5999-

6106.0346

-5040.509

51

-5674.893

15

6000 12685.1724

12317.26022

13991.10941

8000 0 0 010000 0 0 0

Table 2. Denoised coefficients ofdisturbance signals at noise levels of 50

dB, 40 dB, 35 dB.

D. D-STATCOM

A D-STATCOM (Distribution StaticCompensator), which is schematically

depicted in Figure 9, consists of a two-level Voltage Source Converter (VSC), adc energy storage device, a couplingtransformer connected in shunt to thedistribution network through a couplingtransformer. The VSC converts the dcvoltage across the storage device into aset of three-phase ac output voltages.These voltages are in phase and coupledwith the ac system through the reactanceof the coupling transformer. Suitableadjustment of the phase and magnitude ofthe D-STATCOM output voltages allowseffective control of active and reactivepower exchanges between the D-STATCOM andthe ac system. Such configuration allowsthe device to absorb or generatecontrollable active and reactive power.

The D-STACOM employs aninverter to convert the DC link voltageVdc on the capacitor to a voltage sourceof adjustable magnitude and phase.Therefore the D-STATCOM can be treated asa voltage-controlled source. Capacitorsizing is referred to the fault current inthe system. The difference in currentbetween the current before and after thefault is considered as current faults. Incapacitor sizing, a suitable range of DCcapacitor is needed to store the energyto mitigate the voltage sag. The DCcapacitor, CDC is used to inject reactivepower to the D-STATCOM when the voltageis in sag condition. The VSC connected inshunt with the ac system provides amultifunctional topology which can beused for up to three quite distinctpurposes:

1. Voltage regulation and compensation ofreactive power

2. Correction of power factor

3. Elimination of current harmonics.

Distribution STATCOM (DSTATCOM)exhibits high speed control of reactivepower to provide voltage stabilization,

flicker suppression, and other types ofsystem control. such device is employedto provide continuous voltage regulationusing an indirectly controlled converter.

Fig.9. Schematic diagram of a D-STATCOM.

Distribution Static Compensatorinjected current Ish corrects the voltagesag by adjusting the voltage drop acrossthe system impedance Zth. The value of Ish

can be controlled by adjusting the outputvoltage of the converter. The shunt injected current Ish can be

written as,

(15)

(16)

The complex power injection of the D-STATCOM can be expressed as,

(17)

It may be mentioned that theeffectiveness of the D-STATCOM incorrecting voltage sag depends on thevalue of Zth or fault level of the loadbus. When the shunt injected current Ish iskept in quadrature with VL, the desiredvoltage correction can be achievedwithout injecting any active power intothe system. On the other hand, when thevalue of Ish is minimized, the same voltagecorrection can be achieved with minimumapparent power injection into the system.

Following figure shows the simulink modelof D-STATCOMwhich is proposed for mitigation ofvoltage dips.

Fig.10. Simulink model of D-STATCOM.

The simulation of Vrms containsno D-STATCOM is as shown in fig.11. Sagin voltage caused by single line toground fault during the period 0.5–0.9s. The voltage sag at the load point is54% with respect to the referencevoltage.

Fig.11. Voltage sag Vrms at the load point:WithoutD-STATCOM.

A new simulations was carriedout with the D-STATCOM connected to thesystem as shown in Figure 4 where thevery effective voltage regulationprovided by the D-STATCOM can be clearlyappreciated.

Fig.12. Voltage sag Vrms at the load point:With

D-STATCOM.

Fig. 13 illustrates thesimulation contains no D-STATCOM andsingle line to ground fault occurred forthe period 0.5–0.9 s. The voltage swellat the load point is 58% with respect tothe reference voltage. The test systemfor the simulation of D-STATCOM for swellis shown in Figure 14.

Fig.13. Voltage swell Vrms at the loadpoint: Without

D-STATCOM.

Fig.14. Voltage swell Vrms at the loadpoint: WithD-STATCOM.

v. CONCLUSION To improve the abilities ofthe PQ monitoring spatial correlationbased noise-suppression algorithm havebeen presented in this paper to denoisePQ waveforms. The noises mixed with themonitored signals were automaticallyremoved according to the correlations atadjacent scales through the proposedapproach. The user control in theperformance of these techniques is one ofthe prominent advantages and providesgreater flexibility compared to otherdenoising techniques. The results of testing theproposed approach on the data for actualtransient events showed that the PQmonitoring approach could restore thecapabilities of the DWT techniques. Thespatial correlation based noise-suppression algorithm, when integratedwith a DWT-based PQ monitoring system,not only restores the detection andlocalization capability of the DWT-basedmonitoring system but also reduces thevolume of data to be recorded. In thispaper, the performance of proposedapproach was tested using event signalsunder the conditions with noisy andnoiseless. Moreover, denoising of thewaveform data restores the wavelet-domainenergy distribution pattern of thedisturbance and will certainly increasethe monitoring accuracy of the DWT-basedmonitoring. It has the great potential toimprove the performance of the automaticPQ monitoring.

This paper has presented the powerquality problems such as voltage dips,swells and interruptions, consequences,and mitigation techniques of custom powerelectronic device D-STATCOM. The designand applications of D-STATCOM for voltagesags, swells and comprehensive resultsare presented.

vI. REFFERANCES

[1] Chiung-Chou Liao, Hong-Tzer Yang and Hsueh-HsienChang “Denoising Techniques With a SpatialNoise-Suppression Method for Wavelet-Based PowerQuality Monitoring,” IEEE Trans. Instrum. Meas., Vol.60, No. 6, July 2011.

[2] U. D. Dwivedi and S. N. Singh, “Enhanceddetection of power quality events using intraand interscale dependencies of waveletcoefficients,” IEEE Trans. Power Del., vol. 25, no. 1,pp. 358–366,Jan. 2010.

[3] U. D. Dwivedi and S. N. Singh, “Denoisingtechniques with change point approach forwavelet-based power-quality monitoring,” IEEETrans. Power Del., vol. 24, no. 3, pp. 1719–1727,Jul. 2009.

[4] T. Tarasiuk, “Comparative study of variousmethods of DFT calculation in the wake of IECstandard 61000-4-7,” IEEE Trans. Instrum. Meas.,vol.58, no. 10, pp. 3666–3677, Oct.

[5] W. G. Morsi and M. E. El-Hawary, “Wavelet packettransform based power quality indices forbalanced and unbalanced three-phase systemsunder stationary or nonstationary operatingconditions,” IEEE Trans. Power Del., vol. 24, no. 4,pp. 2300–2310,Oct. 2009.

[6] P. S. Wright, “Short-time Fourier transformsand Wigner–Ville distributions applied to thecalibration of power frequency harmonicanalyzers,”IEEE Trans. Instrum. Meas., vol. 48, no. 2,pp. 475–478, Apr. 1999.

[7] M. Han, Y. Liu, J. Xi, and W. Guo, “Noisesmoothing for nonlinear time series usingwavelet soft threshold,” IEEE Trans. Signal Process.Lett., vol. 14, no. 1, pp. 62–65, Jan. 2007.

[8] Julio Barros, Senior Member, IEEE, and Ramón I.Diego “Analysis of Harmonics in Power SystemsUsing the Wavelet-Packet Transform” IEEE Trans.instrum. meas., vol. 57, no. 1, Jan. 2008.

[9] Y. Xu, J. B. Weaver, D. M. Healy, and J. Lu,“Wavelet transform domain filters: Aspatially selective noise filtration technique,”IEEE Trans. Image Process., vol. 3, no. 6, pp. 747–758, Nov. 1994.

*****