IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO.

8, AUGUST 2012 3321

Virtual Multifunction Power Quality Analyzer Based

on Adaptive Linear Neural Network
Cheng-I Chen, Member, IEEE

Abstract—Monitoring of electrical quantities is an important the dynamic harmonic/interharmonic estimation [8]. A digi-
task for the evaluation of power quality (PQ). However, analysis tal harmonic analyzer is proposed by integrating developed
methods for PQ disturbances are quite different. This circum- frequency estimation method and finite-impulse-response fil-
stance would make the design of a multifunction PQ-measuring
instrument difficult. In this paper, the design and implementation ter [9]. The discrete Fourier transformation is implemented
of a virtual multifunction PQ analyzer based on the adaptive in a small programmable-logic-device chip for the harmonic
linear neural network are discussed. The main advantages of the measurement [10]. For the evaluation of voltage fluctuations,
realized analyzer are the simplification and integration for the a digital implementation of an IEC flickermeter in the hybrid
harmonic/interharmonic analyzer, flickermeter, and voltage event time–frequency domain is adopted [11]. A virtual flickermeter
identifier by adopting the same computational mechanism. Finally,
some tests are made to verify the performance of the proposed followed by the model in IEC Std. 61000-4-15 is developed
virtual multifunction analyzer. [12]. To condition the PQ of distribution systems, a model-
based controller combined with multivariable regulator theory
Index Terms—Adaptive linear neural network (ADALINE),
flickermeter, harmonic/interharmonic analyzer, power quality and Kalman filter is provided [13]. The synchronized rms
(PQ), voltage event identifier. calculation is utilized for the voltage dip measurement [14].
A phase-locked-loop-based mechanism is proposed in [15] to
I. I NTRODUCTION perform the synchronization of PQ instruments for stationary
and transient disturbances. A synchronizer based on chirp z

P OWER QUALITY (PQ) is always of great concern for

the control and protection of the supply system and sen-
sitive equipment. Effective evaluation of electrical quantities
transform is proposed for the PQ measurement and verified
with the testing states defined in IEC Std. 61000-4-30 [16].
For monitoring PQ integrated indexes, a real-time multifunc-
is the first step to maintain the PQ [1], [2]. In this paper, tion instrument is implemented on a digital signal processing
the measurement of PQ is mainly focused on power system board [17].
frequency deviation, harmonics/interharmonics, voltage fluctu- By observing these research results, it is easily found that
ations (flickers), voltage sags (dips), swells, and interruptions, the analysis methods are quite different. If one would like to
which are defined in International Electrotechnical Commis- establish a digital multifunction analyzer, the integration of
sion (IEC) Std. 61000-4-30 [3]. In particular, the proposed distinct computational algorithms for PQ disturbances is the
multifunction PQ analyzer mainly refers to IEC Std. 61000- critical issue. To provide a solution to this situation, a virtual
4-7 [4] for the harmonic and interharmonic analysis and to multifunction PQ analyzer with the ADALINE is proposed.
IEC Std. 61000-4-15 for the implementation of a flickermeter With the developed virtual multifunction PQ analyzer, several
[5]. In the monitoring of voltage events (voltage sags, swells, benefits can be obtained as follows.
and interruptions), IEC Std. 61000-4-30 [3] and IEEE Std.
1159-1995 [6] provide the required guidelines and related test 1) By integrating different kinds of PQ analyzers with the
procedure. ADALINE-based method, the setup of the measuring
Although the aforementioned standards suggest design re- system and the analysis algorithm are easy to implement.
quirements for PQ analyzers, analysis methods are not lim- 2) Using Prony’s spectral estimation method to estimate
ited for providing the flexibility of instrument implementation. slowly varying power system frequency in the analysis
Therefore, an advanced spectrum estimation based on Prony’s model, it is not necessary to perform the resampling
method is proposed to provide high resolutions for harmonic process for synchronization in the harmonic analysis.
analysis [7]. A two-stage method based on the adaptive lin- 3) The proposed ADALINE-based indirect demodulation
ear neural network (ADALINE) is displayed to implement method can extract the accurate voltage envelope with the
help of high frequency resolution.
4) Compared with commonly seen algorithms, the proposed
Manuscript received March 30, 2010; revised January 23, 2011 and hybrid identifier can perform detection and identification
April 3, 2011; accepted April 23, 2011. Date of publication May 5, 2011; date of voltage events without training process.
of current version March 30, 2012. This work was supported by the National
Science Council of Taiwan under Grants NSC99-2218-E-468-002 and NSC99- 5) The virtual PQ analyzer is a flexible analytical and mea-
2632-B-468-001-MY3. suring platform without taking hardware requirements
The author is with the Department of Computer Science and Information into account. Therefore, the analysis algorithm can be
Engineering, Asia University, Taichung 41354, Taiwan (e-mail: cichen@
asia.edu.tw). easily adjusted according to the revisions of international
Digital Object Identifier 10.1109/TIE.2011.2151811 standards.
0278-0046/$26.00 © 2011 IEEE
3322 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO. 8, AUGUST 2012

6) The designed interface could facilitate the complicated can be obtained via weights wm sin and wm cos according to
task to monitor different PQ states of systems. [8], [21]

The organization of this paper is as follows. In Section II, 2 2
Am = wm cos + wm sin (3)
the analysis methods based on the ADALINE for power system
frequency deviation, harmonics/interharmonics, flickers, volt-  
wm sin
age sags, swells, and interruptions are introduced. To provide φm = tan−1 . (4)
wm cos
a multifunction PQ analyzer, analysis structures in Section II
are integrated in Section III. Comprehensive case studies are Since the analysis model of conventional ADALINE is based
designed and analyzed in Section IV to verify the performance on a known power system frequency f1 and makes fm =
of the proposed virtual instrument. mf1 (m is an integer), the presence of system frequency
deviation and interharmonics would deteriorate the analysis
results. Prony’s method, which calculates the linear prediction
II. PQ A NALYSIS W ITH ADALINE-BASED A LGORITHMS parameters that represent the observed data sequence, can
To characterize PQ disturbances, sinusoidal models are the provide the high resolution for frequency estimation to solve
most suitable ones. For a measured signal y, its discrete-time the aforementioned problems. The concept of this solution
form y(n) of finite length N sampled at the time interval Δt process is to find the frequency information fm in (2) first.
can be represented by H sinusoidal components with noises as Then, the correct frequencies of fundamental, harmonic, and
interharmonic components will be fed into the analysis model

H of conventional ADALINE. Once all the components are con-
y(n) = Ah cos(ωh nΔt + φh ) + W N (n) (1) sidered in the ADALINE model, the estimation error can be
h=1 effectively reduced. The key technique in Prony’s method is to
introduce a polynomial P (z) with exponential terms containing
where Ah is the amplitude, φh is the initial phase angle, ωh = frequency information being its roots, as shown in [8], [22]
2πfh is the radian harmonic or interharmonic frequency, and
W N (n) represents the additive white noise. 
M 
M

The ADALINE is a multiple-input, single-output, and single- P (z) = (z − zm ) = am z M −m (5)

m=1 m=0
layer linear neural element used for noise canceling or sig-
nal extraction. In recent years, this method has been widely where zm = ej2πfm Δt , a0 = 1, and M is the estimation order.
used in the area of PQ studies [18]–[20]. The advantages of The sinusoidal frequencies fm then can be obtained by
ADALINE over other neural networks are the online adjust-
ment mechanism and easy-to-implement structure. The solving fm = tan−1 [Im{zm }/Re{zm }] /2πΔt (6)
processes for different PQ disturbances with the ADALINE-
based method are illustrated in the following. where Im{} and Re{} represent the imaginary and real parts
of zm , respectively. However, this may introduce a set of new
unknown polynomial coefficients am . By observing Prony’s
A. Power System Frequency Deviation and solution process, it is found that the estimated signal and
Harmonics/Interharmonics polynomial coefficients can be represented as an autoregressive
For harmonic analysis, the discrete form of the estimated (AR) model in
signal y in (1) then can be expressed by
y(n) = −a1 y(n − 1) − a2 y(n − 2) − · · · − aM y(n − M )

M 
M
y(n) = Am cos(2πfm nΔt + φm ) =− am y(n − m). (7)
m=1 m=1


M By observing (2) and (7), it is found that the mathematical
= (Am cos φm cos 2πfm nΔt forms of these equations are similar. Therefore, it is convenient
m=1 to apply ADALINE to implement the solution procedure for
all am in Prony’s method to solve the frequency problem in
− Am sin φm sin 2πfm nΔt)
ADALINE. More details can be obtained in [8].

M After obtaining the expansion coefficients and solving (5),
= (wm cos cos θm − wm sin sin θm ) (2) the estimated frequencies would be used in all θm in (2).
m=1 Then, the power system frequency deviation and information
about interharmonics would be calculated by the conventional
where M is the estimation order representing the number of ADALINE. The advantage of such two-stage ADALINE struc-
considered harmonics, θm = 2πfm nΔt, wm cos = Am cos φm , ture over the frame-based techniques is that the time-varying
and wm sin = Am sin φm . characteristics of power system frequency, harmonics, and in-
By minimizing the estimation error with the gradient descent terharmonics can be accurately tracked [22]. Since the effect of
method, the amplitude and phase angle of the mth harmonic power system frequency deviation is considered in the model of
CHEN: VIRTUAL MULTIFUNCTION POWER QUALITY ANALYZER BASED ON ADAPTIVE LINEAR NEURAL NETWORK 3323

ADALINE, the resampling process for harmonic measurement Section II-A performs the analysis by tracking the time trajec-
in [16] and [17] is not necessary. tories of signal parameters, (3) can represent the variation of
system voltage. However, the adjustment of weights with in-
troducing a learning parameter would take some computational
B. Voltage Fluctuations (Flickers)
steps to achieve numerical convergence. This is the common
In general, the voltage fluctuations can be expressed as the problem for the recursive methods [22]. In this solution mecha-
amplitude-modulated signal without harmonics yf (n) given in nism, the time locations of voltage events obtained directly with
  ADALINE are not precise enough, as shown in Fig. 1(b).
 I
yf (n) = A1 + Ai cos(ωf i nΔt + φf i ) In recent years, wavelet transformation has been frequently
i=1
applied in the study of PQ assessment [24]–[26]. When a
voltage event starts or ends, the singular-signal samples would
× cos(ω1 nΔt + φ1 ) cause significant estimation errors in the analysis model. The
= AEn (n) cos(ω1 nΔt + φ1 ) (8) wavelet analysis providing multiresolution analysis is a power-
ful tool for the localization of singular signals. With the dyadic
where A1 , ω1 , φ1 , Ai , ωf i , and φf i are the amplitudes, the analysis structure of wavelet, the singular signals can be sepa-
angular frequencies, and the phase angles of the fundamental rated from those normal ones. Although the residues of wavelet
and flicker components, respectively, and I is the number of coefficients can accurately represent occurrences of events,
flicker components. AEn (n) represents the envelope signal the identification of events becomes a problem, as shown in
in (8). Fig. 1(c). Therefore, many energy indexes and postprocessing
Based on the analog functional diagram of a flickermeter in methods based on artificial intelligence are proposed [24]–[26].
IEC Std. 61000-4-15, the flicker signal is converted into an in- The common drawbacks of these techniques are increase of
dex to express the severity of the voltage fluctuation. Therefore, computational burden and different self-defined parameters of
the accurate envelope extraction is the crucial step for the flicker identification based on experimentation.
evaluation. In this paper, the proposed solution process with In this paper, a hybrid voltage event identifier with ADA-
ADALINE for the harmonic analysis can be directly adopted to LINE and wavelet transformation is developed. If the voltage
obtain AEn (n) with (3). In general, the commonly used indirect events are present, the detection of wavelet analysis will be
demodulation method to extract a voltage envelope is based on activated to perform the time locating of disturbances. Then, the
squaring multiplier and specified bandpass filter, as addressed analyzed magnitudes of ADALINE Aevent within the duration
in [5] and [12]. To solve the desynchronization of power system of events Devent will be extracted. Finally, the classification can
frequency in the squaring demodulation method, frequency- be performed by comparing the extracted information with the
domain interpolation is proposed in [11]. However, the esti- threshold T defined in IEEE Std. 1159-1995 by using
mation accuracy of frequency-domain interpolation would be
limited due to the frequency resolution in fast Fourier transform Tsag > max {Aevent (X), X ∈ sample set within Devent }
(FFT). For the signal in (1), the resolution of FFT is (11)
1
ΔfFFT = . (9) Tinter. > max {Aevent (X), X ∈ sample set within Devent }
N Δt
(12)
Since Prony’s method can be solved by building the AR
model, the resolution of Prony’s method is approximately eval- Tswell < min {Aevent (X), X ∈ sample set within Devent } .
uated based on the AR analysis and is given by
(13)
1.03
ΔfAR = (10) Therefore, since the magnitude within the duration that lasted
Δt · M · [SN R(M + 1)]0.31
from 0.140 to 0.253 s (6.78 cycles for a 60-Hz system) is
where SN R represents the signal-to-noise ratio of the signal. 0.85 p.u., the voltage event in Fig. 1(a) can be easily identified
By comparing with the resolution of FFT in (9), it is found that as instantaneous sag. The advantages of the built mechanism
the resolution of the AR model is much finer [23]. Therefore, are the convenience of identification with related coefficients in
Prony’s method can effectively prevent the leakage effect in the IEEE Std. 1159-1995 and simplicity of computational complex-
FFT-based method. It is also clearly seen that the resolution of ity via online adjustment capability of ADALINE. In addition,
the AR model is not dependent on the length of the measured the estimation error can be reduced with the adoption of the
signal. As a result, the proposed ADALINE-based indirect AR model in (7) to modify the original ADALINE model if the
demodulation method can obtain more accurate envelope than power system frequency deviation is present.
that in [11].
III. S ETUP OF THE V IRTUAL M ULTIFUNCTION
C. Voltage Sags, Swells, and Interruptions PQ A NALYZER
In general, the accurate analysis of voltage events is relied In most PQ analyzers, implemented algorithms for assess-
on the detection of time locations when event happens and ment of harmonics/interharmonics, flickers, and voltage events
the category of event. Since the proposed solution process in are different. To facilitate the implementation of integration
3324 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO. 8, AUGUST 2012

Fig. 1. Detection of voltage events: (a) Waveform of sag; (b) directly with ADALINE; and (c) with wavelet transformation.

for various instruments, a multifunction virtual PQ analyzer tained voltage envelope could be further processed through the
developed with LabVIEW is introduced. IEC flickermeter developed in [11] to provide PQ assessment
(block I). For voltage event analysis mode, a hybrid voltage
event identifier with blocks J and H is developed. In block J,
A. Structure of the Proposed Virtual PQ Analyzer the time locations of voltage event starts and ends will be calcu-
The virtual instrument is a measuring device combining lated. By checking the rms value of the voltage (block H) within
the analog-to-digital conversion and related signal conditioning the event duration, the classification of event can be effectively
modules with the personal computer (PC). In such measuring performed in block K according to the event definition of IEEE
structure, the analysis functions are determined by the system Std. 1159-1995 [6]. Analysis results, along with PQ assessment
designer. Since the conversion and preprocessing circuits are indexes obtained in block L, will be collected in block M. Then,
realized in the modules, the designers could easily develop the the text and graphical analysis results can be viewed on the GUI
analysis algorithms based on the hardware abstraction. In such in Fig. 3 through block N.
measuring mechanism, many complicated settings are deter-
mined automatically by the hardware drivers, and designers
only need to specify the required parameters to operate the B. Hardware Description for the Laboratory Setup
related hardware. The laboratory setup of the proposed virtual instrumen-
In the proposed PQ analyzer, several analysis techniques tation is shown in Fig. 4. By obtaining the testing signal
described in Section II are integrated. A block diagram of the generated from the arbitrary waveform generator [(AWG);
implemented instrument is shown in Fig. 2, where the gray Agilent Technologies 33220A] with the I/O connector and DAQ
areas represent the corresponding function blocks observable card (National Instruments PCI-6036E), the analysis program
on the graphical user interface (GUI) in Fig. 3. According performed on the PC (Dual Core Intel Pentium D 3.4-GHz
to IEC Std. 61000-4-7, the analysis frequency range of in- processor and 2-GB RAM) can be easily used for PQ assess-
terest for harmonic measurement is within 0–9 kHz. Thus, ments. In this way, system designers can conveniently modify
the analyzed signal (block A) sampled with specified para- the analysis algorithms according to the revisions of interna-
meters is first processed through a low-pass antialiasing filter tional standards. A digital phosphor oscilloscope (Tektronix
performed in the module data acquisition (DAQ) device with TDS5034) is used for observing the waveform of the measured
a cutoff frequency of 10 kHz (block B). The waveform of signal. By comparing the referenced signal generated from the
the sampled signal can be observed in block D. By selecting AWG with the estimation results, users can easily verify the
different analysis mode in block C, users can easily observe accuracy and practicality of the proposed multifunction PQ
the analyzed results for harmonics/interharmonics, flickers, and analyzer.
voltage events. Under harmonic and interharmonic analysis
mode, the analyzed signal is first processed in block E to
obtain required parameters am of Prony’s polynomial in (5) by IV. E XPERIMENTAL R ESULTS
implementing with ADALINE for (7). Through the root-finding
To verify the performance of the described virtual analyzer
procedure, the calculated frequency components in (6) will be
for the estimation of PQ disturbances, a series of test signals
fed into the original ADALINE (block F) to perform the har-
is generated via the AWG in Fig. 4. For all ADALINE-based
monic analysis. Because the power system frequency deviation
functional blocks, the initial weights are set to be zero, and the
and interharmonics are taken into account in the ADALINE
estimation order M is set to be 128. The sampling rate of DAQ
model, the estimated errors can be effectively mitigated.
is 20 kHz.
Since the harmonic components usually are not considered
in the assessment of flickers and voltage events, the extrac-
tion of fundamental components is performed in block G.
A. Testing for Estimation of Power System Frequency
Under voltage fluctuation (flicker) analysis mode, the pro-
Deviation and Harmonics/Interharmonics
posed ADALINE-based indirect demodulation with funda-
mental components and the root-finding procedure in block To examine the accuracy and robustness of PQ instruments,
H are used to obtain the correct voltage envelope. The ob- IEC Std. 61000-4-7 and Std. 61000-4-30 provide a complete
CHEN: VIRTUAL MULTIFUNCTION POWER QUALITY ANALYZER BASED ON ADAPTIVE LINEAR NEURAL NETWORK 3325

Fig. 2. Block diagram of the virtual multifunction PQ analyzer.

Fig. 3. GUI of the integrated virtual instrument with respect to the function blocks in Fig. 2 under harmonic and interharmonic analysis mode.

view of both testing techniques and permissible limits of produce the same readings within the accuracy of the in-
estimation errors [3], [4]. IEC Std. 61000-4-7 also indicates struments) by several testing states. The uncertainty of an
requirements of the synchronous sampling of the power signal instrument shall be tested for each measured quantity as
to limit errors. Therefore, the accurate estimation of the power follows [3].
system frequency is required even in the presence of nonsta-
tionary signals. 1) Select a measured quantity (for example, rms voltage
The measurement of a specific characteristic can be ad- magnitude).
versely affected by the application of a disturbing influ- 2) Holding all other quantities in testing state 1, verify the
ence (i.e., influence quantity) on the electrical input signal. uncertainty of the measured quantity to be tested at five
Therefore, Table 3 of IEC Std. 61000-4-30 recommends equally spaced points throughout the range of influence
the standard uncertainty testing process for class A instru- quantity (for example, 0% through 200% of Udin , where
ments (i.e., reference instrument for contracts and disputes; Udin is the declared input voltage obtained from the
two class A instruments connected to the same signal will declared supply voltage by a transducer ratio).
3326 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO. 8, AUGUST 2012

Fig. 4. Hardware configuration for the laboratory setup of the

instrumentation.

3) Holding all other quantities in testing state 2, repeat the

test.
4) Holding all other quantities in testing state 3, repeat the
test.

In this case, a test signal in (14) is generated via the AWG,

where the nominal power system frequency f is 60 Hz Fig. 5. Relative errors of the (a) power system frequency and (b) amplitudes
for the fundamental and harmonic components.
◦ ◦
v(t) = sin(2πf t + 15 ) + 0.3 sin(3 × 2πf t + 34 )
permissible limit of 5%. However, there is no specific indication
+ 0.2 sin(5 × 2πf t + 52◦ )+ 0.1 sin(7 × 2πf t + 67◦ ).
for testing state 1 concerning the orders of harmonic and
(14) interharmonic components. A particular condition is selected
Since the measurement in this section is related to the harmon- by specifying a third harmonic and two equal interharmonics at
ics, only the power system frequency and the magnitude of 1.5ω1 and 2.5ω1 . By observing Fig. 5(b), it can be seen that the
the declared input voltage are selected as influence quantities. estimation error is increased as the harmonic frequency goes up.
According to IEC Std. 61000-4-30, the ranges of influence The reason is that the actual amplitudes of the high-frequency
quantities for the power system frequency and the voltage harmonics are usually smaller than those of the low-frequency
magnitude are 51–69 Hz and 0%–200% of Udin in the 60-Hz ones, as shown in testing states 2 and 3. Such situation would
system, respectively. Thus, the estimation of the power system make the calculations of the relative error in (15) being larger
frequency and harmonic amplitudes is verified at five equally for the high-order harmonics. Although the state of the power
spaced points in the specific ranges. To perform the verification system frequency is usually well maintained by the automatic
of estimation for the ADALINE-based mechanism, the effects generation control, the proposed estimation mechanism still
of transducers are not taken into account, as declared in IEC could satisfy the measurement requirements set by the IEC
Std. 61000-4-30. This means that the extracted digital signal standard, where the variation of the fundamental frequency over
can completely represent all the characteristics of a true power a wide range is considered a standard testing state.
signal. The maximum estimation errors among the three testing
states for the power system frequency and harmonics are shown B. Testing for Assessment of Flickers
in Fig. 5(a) and (b). By observing Fig. 5(a), it is found that a
large deviation from the nominal fundamental frequency may According to the performance tests listed in IEC 61000-4-15,
lead to an increasing relative estimation error, as defined in the corresponding test signals denoting the flicker voltages are
produced from the AWG, and the results are shown in Table I
Relative Error(%) and Fig. 6. From Fig. 6 and Table I, it is found that there
are good agreements of the analyzed results in both sinusoidal
|Actual V alue − Estimated V alue|
= × 100%. (15) modulated signal and rectangular modulated signal between
Actual V alue
IEC 61000-4-15 and the proposed mechanism.
Due to the high resolution of Prony’s method for the fre-
quency estimation, the corresponding estimation error is able
C. Testing for Detection and Classification of Voltage Events
to meet the maximum permissible limit of 0.03%, which is
required by IEC Std. 61000-4-7 [4]. About the estimation of The categories and related characteristics of identification for
harmonic amplitude, IEC Std. 61000-4-7 requires a maximum voltage events are clearly illustrated in IEEE Std. 1159-1995
CHEN: VIRTUAL MULTIFUNCTION POWER QUALITY ANALYZER BASED ON ADAPTIVE LINEAR NEURAL NETWORK 3327

TABLE I TABLE II
T EST S PECIFICATION C OMPARED B ETWEEN THE P ROPOSED M ECHANISM C LASSIFICATION R ESULTS FOR VOLTAGE E VENTS IN IEEE S TD .
AND IEC 61000-4-15 1159-1995 [6]

TABLE III
M AXIMUM AND M EAN R ELATIVE E RRORS OF A MPLITUDES FOR E ACH
H ARMONIC IN C ASE D

Fig. 6. Comparison of the proposed mechanism and IEC 61000-4-15 for the
flicker perceptibility threshold curve.

[6]. To verify the correctness of voltage event classification,

1000 patterns for each voltage event are generated. It is found
that the correctness of classification shown in Table II can
be maintained in a permissible range. Compared with other
traditional neural-network-based methods, an important benefit
in the proposed mechanism is that it is needless to train the
analysis system. It is found that the correctness of sustained
interruption is lower than others. The reasons may come from
the sensitivity of wavelet transformation to noises. Moreover, Fig. 7. Measured fundamental rms voltage of the EAF.
when the amplitudes and duration of the automatically gener-
ated pattern events are close to the thresholds defined in IEEE
Std. 1159-1995, the accuracy of event identification may be E. Actual Measurements From AC EAF
affected. To show the usefulness of the proposed approach, the same
system in [11] including a 50-ton three-phase ac electric arc
furnace (EAF) connected to an 11.4-kV/460-V transformer
D. Actual Measurements From UPS
rated at 33 MVA is investigated. Fig. 7 shows the fundamental
A current signal from an uninterrupted power supply [(UPS); rms voltage of the EAF. Table IV shows the relative errors of
Phoenixtec 3C3-60000] is extracted. Because the harmonic short-term flicker severities for frequency-domain interpolation
components of the measured waveform are unknown, results in [11], conventional ADALINE, and the proposed method,
of the FFT calculated by 12 cycles of the signal are used where estimation results of the IEC flickermeter are considered
as reference [4], where the four dominant components, the as reference for comparison.
fundamental, the third, the fifth, and the seventh harmonics, are
adopted for analysis. The maximum and mean relative errors
F. Actual Measurements From a Substation
of the amplitude for each harmonic obtained by the FFT, the
conventional ADALINE, and the proposed method are shown For examining the detection of voltage events, an actual sag
in Table III. waveform extracted from a voltage event recorder in Fig. 8
3328 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 59, NO. 8, AUGUST 2012

TABLE IV G. Discussions
R ELATIVE E RRORS OF S HORT-T ERM F LICKER S EVERITY (PST ) IN
C ASE E O BTAINED BY S EVERAL M ETHODS For the harmonic and interharmonic analysis, the proposed
solution mechanism can meet the requirements of IEC Std.
61000-4-7 and IEC Std. 61000-4-30. In addition, the estimation
of UPS current shows the superior accuracy of the proposed
method. It is worth noting that the current signal is time varying,
so the reference results obtained from 12-cycle FFT are only the
average values.
From Table I and Fig. 6, it is known that the designed
flickermeter with the proposed demodulation method meets the
test specification in IEC Std. 61000-4-15. Similar to case D,
the estimation results of the IEC flickermeter are regarded as
references since it is not able to realize the standard flicker
severity for the actually measured signal. The estimation results
of the conventional ADALINE demodulation method would
be deteriorated due to the power system frequency deviation,
and the frequency-domain interpolation demodulation method
in [11] would be influenced if the frequency resolution of FFT
Fig. 8. Simplified circuit diagram for voltage event monitoring.
is not sufficient. With the high resolution of Prony’s method,
the envelope of voltage flicker can be accurately obtained by
the proposed method.
In the identification of voltage events, the correctness of esti-
mation can be maintained in a permissible range. By observing
the analysis results for a sag signal extracted from a substation,
the commonly used rms method would cause detection delay.
For the conventional ADALINE method, the estimation results
are deteriorated because of desynchronization of power system
frequency. This problem can be improved with the proposed so-
lution process by adjusting the ADALINE model from (5)–(7).
According to the hybrid mechanism of wavelet transformation
in Fig. 9(d) and the proposed ADALINE-based method in
Fig. 9(e), the power signal in Fig. 9(a) can be identified as the
instantaneous sag.

V. C ONCLUSION
In this paper, a virtual multifunction PQ analyzer based
on the ADALINE has been developed. Since the ADALINE-
based method is an easy-to-implement approach, the proposed
measuring system is convenient to build up and integrate. More-
over, the analysis algorithms can be flexibly adjusted under this
virtual platform according to the revisions of international stan-
dards. From the experiments, it is displayed that testing results
obtained by the proposed mechanism meet the requirements in
specific standards.

R EFERENCES
[1] M. H. J. Bollen and I. Y. H. Gu, Signal Processing of Power Quality
Disturbances. Hoboken, NJ: Wiley, 2006.
[2] P. Caramia, G. Carpinelli, and P. Verde, Power Quality Indices in Liberal-
Fig. 9. (a) Waveform of the captured voltage sag. (b) Detection magni- ized Markets. Hoboken, NJ: Wiley, 2009.
tude of the rms method. (c) Detected magnitude of conventional ADALINE. [3] Electromagnetic Compatibility (EMC). Part4: Testing and Measurement
(d) Detected result of wavelet transformation. (e) Detected magnitude of the Techniques—Power Quality Measurement Methods, IEC Std. 61 000-4-
proposed ADALINE-based method. 30, 2003.
[4] Testing and Measurement Techniques—General Guide on Harmonics and
Interharmonics Measurements and Instrumentation, for Power Supply
is investigated. Fig. 9 shows the estimation results obtained Systems and Equipment Connected Thereto, IEC Std. 61 000-4-7, 2002.
[5] Electromagnetic Compatibility (EMC). Part4: Testing and Measurement
by the rms method, the conventional ADALINE, the wavelet Techniques. Section 15: Flickermeter. Functional and Design Specifica-
transformation, and the proposed ADALINE-based method. tions, IEC Std. 61 000-4-15, 2003.
CHEN: VIRTUAL MULTIFUNCTION POWER QUALITY ANALYZER BASED ON ADAPTIVE LINEAR NEURAL NETWORK 3329

[6] IEEE Recommended Practice for Monitoring Electric Power Quality, [19] B. Singh, V. Verma, and J. Solanki, “Neural network-based selective
IEEE Std. 1159-1995, Nov. 1995. compensation of current quality problems in distribution system,” IEEE
[7] Z. Leonowicz, T. Lobos, and J. Rezmer, “Advanced spectrum estimation Trans. Ind. Electron., vol. 54, no. 1, pp. 53–60, Feb. 2007.
methods for signal analysis in power electronics,” IEEE Trans. Ind. Elec- [20] M. Cirrincione, M. Pucci, and G. Vitale, “A single-phase DG generation
tron., vol. 50, no. 3, pp. 514–519, Jun. 2003. unit with shunt active power filter capability by adaptive neural filtering,”
[8] G. W. Chang, C. I. Chen, and Q. W. Liang, “A two-stage ADALINE for IEEE Trans. Ind. Electron., vol. 55, no. 5, pp. 2093–2110, May 2008.
harmonics and interharmonics measurement,” IEEE Trans. Ind. Electron., [21] H. C. Lin, “Fast tracking of time-varying power system frequency and
vol. 56, no. 6, pp. 2220–2228, Jun. 2009. harmonics using iterative-loop approaching algorithm,” IEEE Trans. Ind.
[9] J. J. Tomic, M. D. Kusljevic, and V. V. Vujicic, “A new power system dig- Electron., vol. 54, no. 2, pp. 974–983, Apr. 2007.
ital harmonic analyzer,” IEEE Trans. Power Del., vol. 22, no. 2, pp. 772– [22] C. I. Chen and G. W. Chang, “Virtual instrumentation and educational
780, Apr. 2007. platform for time-varying harmonics and interharmonics detection,” IEEE
[10] V. Pjevalica and V. Vujicic, “Further generalization of the low-frequency Trans. Ind. Electron., vol. 57, no. 10, pp. 3334–3342, Oct. 2010.
true-rms instrument,” IEEE Trans. Instrum. Meas., vol. 59, no. 3, pp. 736– [23] G. W. Chang and C. I. Chen, “An accurate time-domain procedure for har-
744, Mar. 2010. monics and interharmonics detection,” IEEE Trans. Power Del., vol. 25,
[11] G. W. Chang, C. I. Chen, and Y. L. Huang, “A digital implementation no. 3, pp. 1787–1795, Jul. 2010.
of flickermeter in the hybrid time and frequency domains,” IEEE Trans. [24] I. Monedero, C. Leon, J. Ropero, A. Garcia, J. M. Elena, and
Power Del., vol. 24, no. 3, pp. 1475–1482, Jul. 2009. J. C. Montano, “Classification of electrical disturbances in real time using
[12] L. W. White and S. Bhattacharya, “A discrete Matlab–Simulink flickerme- neural networks,” IEEE Trans. Power Del., vol. 22, no. 3, pp. 1288–1296,
ter model for power quality studies,” IEEE Trans. Instrum. Meas., vol. 59, Jul. 2007.
no. 3, pp. 527–533, Mar. 2010. [25] B. Biswal, P. K. Dash, and B. K. Panigrahi, “Power quality disturbance
[13] K. H. Kwan, Y. C. Chu, and P. L. So, “Model-based H∞ control of a classification using fuzzy C-means algorithm and adaptive particle swarm
unified power quality conditioner,” IEEE Trans. Ind. Electron., vol. 56, optimization,” IEEE Trans. Ind. Electron., vol. 55, no. 1, pp. 212–220,
no. 7, pp. 2493–2504, Jul. 2009. Jan. 2009.
[14] D. Gallo, C. Landi, and M. Luiso, “Accuracy analysis of algorithms [26] C. C. Liao and H. T. Yang, “Recognizing noise-influenced power quality
adopted in voltage dip measurements,” IEEE Trans. Instrum. Meas., events with integrated feature extraction and neuro-fuzzy network,” IEEE
vol. 59, no. 10, pp. 2652–2659, Oct. 2010. Trans. Power Del., vol. 24, no. 4, pp. 2132–2141, Oct. 2009.
[15] A. Cataliotti, V. Cosentino, and S. Nuccio, “A phase-locked loop for the
synchronization of power quality instruments in the presence of stationary
and transient disturbances,” IEEE Trans. Instrum. Meas., vol. 56, no. 6,
Cheng-I Chen (S’07–M’09) received the Ph.D. de-
pp. 2232–2239, Dec. 2007.
[16] M. Aiello, A. Cataliotti, V. Cosentino, and S. Nuccio, “Synchronization gree from National Chung Cheng University, Chiayi,
Taiwan, in 2009.
techniques for power quality instruments,” IEEE Trans. Instrum. Meas.,
He is currently an Assistant Professor with the
vol. 56, no. 5, pp. 1511–1519, Oct. 2007.
Department of Computer Science and Information
[17] D. Gallo, C. Landi, and N. Rignano, “Real-time digital multifunction in-
strument for power quality integrated indexes measurement,” IEEE Trans. Engineering, Asia University, Taichung, Taiwan. His
areas of research interests include energy informa-
Instrum. Meas., vol. 57, no. 12, pp. 2769–2776, Dec. 2008.
tion and communication technology, instrumentation
[18] D. O. Abdeslam, P. Wira, J. Merckle, D. Flieller, and Y.-A. Chapuis,
and measurement, and digital signal processing.
“A unified artificial neural network architecture for active power filters,”
IEEE Trans. Ind. Electron., vol. 54, no. 1, pp. 61–76, Feb. 2007.