compliance analysis of pmu algorithms and devices for wide-area stabilizing control of large power...

8/10/2019 Compliance Analysis of PMU Algorithms and Devices for Wide-Area Stabilizing Control of Large Power Systems

1/13

1766 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013

Compliance Analysis of PMU Algorithms andDevices for Wide-Area Stabilizing Control

of Large Power SystemsInnocent Kamwa, Fellow, IEEE, S. R. Samantaray, Senior Member, IEEE, and Geza Joos, Fellow, IEEE

AbstractFor the first time, IEEE Std. C37.118.1-2011 nowprovides metrics for PMU dynamic performance in terms ofclasses P and M filter designs. This paper attempts to determinewhether fulfilling these requirements makes the PMU inherentlywell suited for stability control applications such as wide-areapower system stabilizers (PSSs). In this aim, we considered twodifferent frequency-adaptive approaches for class-P and -Mcompliance to ensure operation over a wide frequency range.The first is based on a finite-impulse response (FIR) with noovershoot in either the phase or the amplitude step responses,

while the second is Kalman filter-based (EKF), which allows fora more refined out-of-band interference rejection at the cost ofa phase step response with overshoot. These two approaches arebenchmarked against Hydro-Qubecs existing PSS requirementsand the conclusion is that the total vector error-based responsetime is not indicative of the phase lag within the frequency bandof interest, nor of the 3-dB bandwidth under sinusoidal ampli-tude/frequency modulation phenomena, which are key criteriawhen specifying PSS PMUs. Using simulated and field-recordednetwork fault responses, we also show that a class-M PMU isunsatisfactory for wide-area stabilizing control, unless its perfor-mance is improved during the fault period, which is not coveredby Std. C37.118.1-2011.

Index TermsAdaptive complex bandpass filtering, changing

harmonics, IEEE Std. C371181-2011, Kalman fi

ltering, phasormeasurement unit (PMU), power system oscillations, syn-chrophasor, wide-area measurement systems (WAMS), wide-areaprotection and control (WAPC).

I. INTRODUCTION

P HASOR measurement unit (PMU) performance has beenthe subject of very intense activity recently. IEEE stan-dard C37.118.1 [1] provides metrics for comparing dynamic

performances of various PMU brands in terms of class-P andclass-M filter designs while Std. C37.118.2 defines the commu-nication protocols more precisely. An IEEE guide for testingand calibrating PMUs more comprehensively and with a greater

Manuscript received May 13, 2012; revised May 27, 2012, July 18, 2012, andAugust 26, 2012; accepted September 24, 2012. Date of publication November12, 2012; date of current version April 18, 2013. Paper no. TPWRS-00494-2012.

I. Kamwa is with the Hydro-Qubec/IREQ, Power System and Mathematics,Varennes QC J3X 1S1, Canada (e-mail: [email protected]).

S. R. Samantaray is with the School of Electrical Sciences, IndianInstitute of Technology, Bhubaneswar, Orissa-751 013, India (e-mail:[email protected]).

G. Joos is with the Department of Electrical and Computer Engi-neering, McGill University, Montreal, QC H3A 2A7, Canada (e-mail:[email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TPWRS.2012.2221168

level of uniformity is currently under ballot [2]. Over ten yearsago [3], authors were arguing that PMU-based wide-area stabi-lizing control of PSS could result in a number of stability ben-efits ranging from markedly improved damping to less severe

post-fault voltage dips and angle shifts[4], [5]. However, despitesubstantial work at the design stage to improve the controllertuning and coordination or at the laboratory and pilot stages tosimulate the whole concept in real time or prove it in open loop[6], [7], no significant progress towards actual implementation

has been achieved so far. The common view is that system oper-ators and reliability regulators are reluctant to move ahead cre-atingthe extra risk posed by a centralized continuous control

bugged by communications uncertainties and delays. However,the gaps in PMU technology and its lack of maturity are sim-

pler explanations for the reluctance of planning and project en-gineers to implement new approaches to improve system sta-

bility. The objective of this paper is to determine whether therecent developments in PMU standards and related commercial

products have significantly improved the prospects of wide-areaPSS implementation.

A literature survey on this topic revealed that some au-thors [8][11] have developed extensive test procedures to

benchmark commercial PMUs against Western ElectricityCoordinating Council (WECC) and North American Syn-chroPhasor Initiative (NASPI) PMU filtering requirements.

Novel synchrophasor algorithms claiming to meet or exceedsimilar requirements are proposed in [12][14]. Specific met-rics which a PMU must meet to make it suitable for wide-areadamping control are summarized in [15] while sample filtersare proposed in [16] showing that the requirements can infact be met. These metrics, which are defined in terms of stepresponse and frequency response (gain and phase) character-istics, under sinusoidal amplitude and phase modulation, areeasy to understand and are the motivation behind the proposed

work. In fact, while Std. C37.118.1 proposes specific criteriaformeasurement bandwidth and step response characterization,these are built around total vector error (TVE), which is not anatural concept for control systems designers. In addition, outof the five commercial PMUs tested in [17], only two met theC37.118.1 class-M step response specification while none ofthem fulfilled the frequency ramp response specification. Asobserved, some work remains to be done to link the metrics in[1] with more conventional figures of [16] and, as a result, morework is now required by manufacturers to fully meet C37.118.1for class-M performance [18].

This paper will impact on both fronts. On one hand, adap-tive algorithms able to help manufacturers meet C37.118.1s

dynamic requirements will be identified. On the other hand,

0885-8950/$31.00 2012 IEEE


2/13

KAMWAet al.: COMPLIANCE ANALYSIS OF PMU ALGORITHMS AND DEVICES 1767

detailed performance analysis of these algorithms will helpbridging the gaps in C37.118.1 in order to enable wide-areastabilizing control with full confidence. Some of these gapsare related to the deficiency of steady-state TVE metrics whenaccurate magnitude is as important as angle accuracy [19][21]and to the non-applicability of TVE as a performance metricduring fault periods when the voltage vanishes [12], [22][24].

The methodology to be followed considers a standard PMUalgorithm complying with C37.118.1 static and dynamic per-formance requirements and an extensive assessment is carriedout to determine whether it complies with the utilities expecta-tions of PMUs in existing PSS and special protection schemes(SPS) applications. Although the standard provides a PMU sim-ulationmodel which can be used for this purpose, it is not imple-mented in a Simulink environment and therefore has difficultyadapting to other purposes. Many other PMU algorithms haverecently been reported but with no comprehensive evidence ofcomplying with the latest C37.118.1 directives and thus are notreadily usable in our project without a prior audit.

In this context, we found it more realistic to consider PMUalgorithms arising from the authors own work [12], [25]. Tworadically different approaches are considered for the sake ofgenerality: a finite-impulse response and a Kalman filter-baseddesign. A PMU of the first kind has a predefined group delayand no overshoot after step input, while the second type ofPMU shows an overshoot in its phase step response. Both arecenter-frequency adaptive and therefore result in an extremelywide frequency range of stable operation, typically from 45 to75 Hz. However, the Kalman filter-based PMU offers the ad-ditional advantage of an unusually high out-of-bandrejection,thanks to an embedded adaptive notch filter. The paper is illus-trated with comparisons of the performance of three commercial

PMUs.

II. PMU FILTERINGREQUIREMENTS ATHYDRO-QUBEC

PMUs have never been used for control at Hydro-Qubecand, as a result, no official guideline exists for selecting themfor this purpose. However, based on the requirements of ex-isting special protection systems (SPS) and power system sta-

bilizer (PSS) devices and systems, it is possible to baseline themetrics to be met by PMUs for future wide-area SPS and PSSapplications.

To provide some context for this discussion, let us first re-

call some characteristics of the Hydro-Qubec grid [26]. It isislanded (i.e., large frequency excursions) and characterized bylong lines with a high level of series and shunt compensation.The system also contains a significant number offlexible ACtransmission systems (FACTS) devices such as high voltageDC (HVDC) and static var compensator (SVC). The geograph-ical location makes the grid sensitive to geomagnetic stormswith a strong likelihood of transformer saturation. The designof measurement systems for WACS/WAMS has been histori-cally driven by this specific context [27]. As an example, hereare typical requirements from the Under Frequency Load Shed-ding project (19982001) [28]:

Fundamental frequency range Hz, but the de-vice should work in the range 40-70 Hz (associated tohydro-generator requirements).

Fig. 1. Summary of Hydro-Qubecs requirements for PMU and SPS algo-rithms design in series-compensated network. (a) Simplified spectrum of elec-tromagnetic phenomena of concern (fundamental frequency at 60 Hz). (b) Re-quired filters with prescribed responsetimesin cycles of fundamentalfrequency.

Rate of change of frequency (ROCOF) range Hz/s, butthe device should work up to Hz/s. The algorithmsshould remain stable up to 25 Hz/s.

Out-of-band filtering of series damped resonance in the

range 2532 Hz (2.5% on voltage, more than 10% oncurrents).

Out-of-bandfiltering of parallel damped resonance in therange 70100 Hz (up to 10% on voltage).

Out-of-band filtering of sub-synchronous parallel reso-nances in the range Hz (5 to 25%).

Intermodulation rejection (due to coupling of transformersaturation and sub-synchronous parallel resonance) in therange and (5% on voltage).

Harmonics rejection (5% for 2nd to 5th and 2% from6th to 10th) at fundamental but also at offset frequencies(5466 Hz).

Instrument transformer transients rejection.The electromagnetic spectrum associated to these require-ments is illustrated in Fig. 1(a). Hydro-Qubec does not specifythe expected accuracy under the above conditions, leaving roomfor innovation. However, in SPS applications, three filters arerequired for voltage and frequency with various expected time-responses. According to Fig. 1(b), Voltage magnitude resultsfrom the filters called V1cy, V3Cy and V12cy, with one, threeand twelve fundamental cycles response-time respectively. Fre-quency is obtained similarly from F1cy, F3cy and F9cy filters,with one, three and nine cycles response time, respectively. Rateof change of voltage (ROCOV) and ROCOF variables are man-dated correspondingly.

In fast control applications (PSS type), the bandwidth ofdc-demodulated signals such as angle, frequency, electric


3/13


Fig. 2. Frequency response of the MB-PSS frequency PMU [29].

TABLE ITYPICAL REQUIREMENTS FORCONTROL WHICH APPLY

TOBOTHAMPLITUDE ANDPHASEMEASUREMENTS

power, etc., is restricted to 10 Hz and the phase lag should notbe larger than that of a 40-ms time constant, under sinusoidalmodulation between 0 Hz and 4 Hz, with a group delay less than1.5 cycles of the fundamental. The possibility of adding a tun-able notch for within-band interference rejection is suggested.

Another source of information for control system filtering

requirements is the MB-PSS, widely used at Hydro-Qubec,which is also known as IEEE PSS4B [29]. Fig. 2 shows thatthe 3-dB bandwidth of the frequency PMU embedded in theMBPSS is 12 Hz, with an attenuation of only dB at 35 Hz.The corresponding phase lag is only 30 at 5 Hz, making it avery fast PMU.

The PMU filtering specifications retained in this work forthe Hydro-Qubec system are given in Table I. They reflect acompromise between the stringent attenuation of interferenceneeded by the defense plan SPS [26][28] and the fast responseneeded by the MBPSS (which takes advantage of the filteringeffect of generator inertia andfield winding to relax interferenceattenuation requirements). In support of the Table I metrics, it isimportant to mention some findings from simulation studies ofwide-area damping control on the large-scale system:

A 50-ms time-constant PMU usually achieves a gooddamping performance of the inter-area modes insimulation.

Typically, a 50-ms pure time delay reduces the inter-areamode damping performance only marginally (needs100 ms to have a detrimental effect).

In addition to Hydro-Qubecs requirements, Table I includesBPA filtering requirements for comparison. Lack of uniformityin the selected metrics makes it difficult to compare the two setsof criteria side-by-side. But it is also clear that they are not ex-

pressed in terms of the TVE and require interpretation to recon-cile them with the C37.118.1 dynamic-performance metrics. Asa further point of comparison, [32] dealt with small-signal os-cillation monitoring in the frequency range Hz] and cameout with more relaxed metrics (20 dB attenuation at Nyquistfrequency, maximum 10 phase lag and 3 dB gain ripple in the

passband).

III. CANDIDATEALGORITHMS FORWIDE-AREACONTROL

For generality, we assume that the PMU input is a singlephase sinusoidal signal (inpu) [37], withsuperimposed DC,interharmonic and fundamental harmonic components:

(1)

where

harmonic order;

highest rank of the harmonic present in the signal;

dc value;magnitude of the th harmonic;

phase of the th harmonic;

fundamental frequency in Hz;

magnitude of the interharmonic component;

frequency of the interharmonic component;

phase of the interharmonic component;

magnitude of the fundamental component;

phase of the fundamental component;

fi

xed sampling period;sampling time index.

For a three-phase system, the symmetrical components canbe derived from separated analysis of the single phase a, b andc of voltage or current signals. This additive combination of thesingle phase phasors result in a positive sequence componentwhich is less noisy than the individual phasors. To extract thefundamental phasor, the frequency response of the filters musthave nulls at the harmonic frequencies that are expected to be

present in the signal and a unity gain at the fundamental fre-quency. If the frequency is not constant, then filter parametershave to be adapted online during frequency estimation. To pro-vide satisfactory measurement over a wide-frequency range, it

becomes necessary to track the system frequency and apply cer-tain corrections on the measuring algorithms and input filters.


4/13


Fig. 3. Frequency response of the anti-aliasfilters.

This principle will guide the design of the two PMU algorithmspresented in the sequel.

We assume that an anti-alias filter with a 3-dB cut-off fre-quency is applied to the signal prior to any processing insidethe algorithm. Two possible choices for this filter are shown inFig. 3. Setting the sampling rate at points/cycle of thenominal fundamental frequency results in a Nyquist frequencyof 12 points/cycle, or the 12th harmonic (i.e., 720 Hz for 60 Hzsystem). At this frequency, the attenuation of the input signal is22 dB and 26 dB for a 4th-order Butterworth and Tchebycheffdesign respectively. Another important point is the phase shift

introduced by these filters at the fundamental frequency (and , respectively) which has to be compensated in the al-gorithm to achieve the TVE specified in [1].

A. FIR Approach

The idea depicted in Fig. 4 is to generate a complex ana-lytic signal corresponding to x(t), by passing the latterthrough a K-taps FIR bandpass filter with complex coefficientswhich are expressed analytically with respect to the filter centerfrequency

with reference to Fig. 4, if the incoming frequency is knownonline, we can re-tune the center frequency of thebandpassfilteraccordingly [12]. Note that the filtering is adaptive but not thesampling.

The time-varying phasor is made stationary by frequencyshifting based on a local reference phasor. Interestingly, thegroup delay of the filter in number of samples is given by

The adaption frequency is derived through a separate fre-quency estimator using the demodulation method [33], [34].

Fig. 4. Adaptive PMU algorithm based on complex bandpass FIRfiltering.

Fig. 5. FIR bandpassfilters for center frequency adaptive PMU algorithms (cf.[11] and [37] for more information about the selected windows).

Simply stated, the frequency is obtained as the low-pass fil-tered derivative of the phasor angle provided by a robust recur-sive DFT. The FIRfilter window is typically of a three-cyclelength (50 ms at 60 Hz). For this purpose, we prefer Kay orTaylor windows which are theoretically justified as more effi-cient than other windows for frequency estimation [34]. To de-sign the bandpass filter, let us consider the following filter bankdefinition which comes from the exponentially modulated (EM)filter bank theory [12], [31]:

(2)

where and representsthe impulse response coefficients of a linear-phase low-pass FIR

prototype filter. Furthermore, the number offilter cells is se-lected so that is the length of the prototype filter. If properlychosen, the scaling factor and the center frequency of the thfilter can be located at the fundamental frequency. Assuming asampling rate of samples per cycle of the fundamental

(3)

To illustrate this band-pass filter design, a few typical proto-type low-pass filters are shown in Fig. 5. For baselining

purposes, the latter include two reference designs C37P andC37M, obtained using the P and M class filters suggested in


5/13


Fig. 6. Kalmanfi lter-based adaptive PMU algorithm. The frequency and am-plitude of the interharmonic (or subsynchronous) component are denoted byand , respectively.

C37.118.1, in cascade with a standard DFT phasor whose am-plitude is frequency-compensated as in [12]. The Kay windowis defined as follows:

(4)

while the Taylor window is given by

(5)

with

According to Fig. 5, the Taylor window-based 4-cycle filter

has a 10-Hz bandwidth and 40 dB of attenuation at 30 Hz with a60-dB stopband above 40 Hz. It can therefore report the phasorat a 60-Hz reporting rate without violating the C37.118.1 re-quirement for Nyquist frequency attenuation. Additional met-rics of FIRfilters are given in Table III.

B. Kalman Filter Approach

Fig. 6 illustrates the second approach for adaptive PMU algo-rithm design. It relies on the fact that a Kalman filter can be con-figured to perform like a discrete Fourier transform [30]. Moregenerally, it can be reformulated as a filter bank with center fre-quencies positioned at arbitrary locations, which makes it suit-able for tracking a fundamental frequency with its harmonics,together with specified inter-harmonic frequencies. The aim ofthe Kalman filter, therefore, is to generate analytical signals

at some given frequency components embedded in the inputsignal:

(6)

where and are the in-phase and in-quadraturecomponents, respectively. For a system with

sinusoidal components, the corresponding state vector at timeinstant is

(7)

with

Assuming that at time instant , we have available measure-

ments of the incoming signal , we can setup the fol-lowing state-variable representation to recursively estimate thestate variable in (7):

(8a)

. . .

(8b)

(8c)

(8d)

Once the state is known, the analytic signal associated

to the fundamental component of can be derived using itsdefinition in (6)(7).Note that in the above equations, the number of sinusoidal

components plus the DC component is equal to .The scalar variable is a zero-mean Gaussian white-noise

process with variance and the initial state is an -meanGaussian randomvariable withcovariance . isthe vari-ance of the additive error of the measuring chain and typically,takes the following form:

The causal estimation of the state , using measurementsso that is minimum, where

represents the estimate of derived from


6/13


Fig. 7. Kalmanfilter gain and frequency response gain (a) without the inter-harmonic component and (b) with a 10-Hz interharmonic interference.

, is an optimal filtering problem whose solu-tion is known to be the Kalman filter [30], [35]. Commonly, theKalman estimation process includes two steps, namely the pre-diction and correction phases.

Let us assume that the state estimate is known with anerror variance . The measured value is then used toupdate the state at instant . The additive correction of the a

priori estimated state at is proportional to the differencebetween the a priori output at instant defined as andthe measurement :

(9)

where is the Kalman gain which guarantees the minimalvariance of the error . Also, at each stepthe variance of the prediction error is calculated:

(10)

This variance matrix is then used to calculate the Kalman gainin the next step of the recursive calculation (correction phase):

(11)

For practical purposes, the steady-state value of the Kalmangain a vector, can be derivedentirely offline using simulations of recursive (9)(11) as ex-

plained in [30]. This way, the filter becomes a fixed-coefficientstate observer with predetermined stability characteristics. Also,

by assigning appropriate values to the frequency components in(8), interharmonics and harmonics can all be tackled togethertransparently, which renders the Kalman filter a more flexibletool than the discrete Fourier transform (DFT).

For illustration purposes, the gain of the two Kalman filterconfigurations used in this paper are shown in Fig. 7, along withthe resulting frequency response gain computed from the statespace (9). Solution (a) assumes a signal model with a funda-mental term, harmonics and a DC component. Solu-tion (b) adds one interharmonic at 10 Hz-frequency to the spec-tral content of (a). It appears that when applied with a constant

gain, the Kalman filter can be considered as an efficient designtool for obtaining the frequency response template in Fig. 7. Theimpact of wrong signal modeling assumptions on the filtered

phasor estimate is easily predicted by looking at thegain response. For instance, the maximum sidelobes is 10 dB in

both configurations while the DC rejection is much better thanthat of a standard [34]. Finally, as illustrated in Fig. 6, knowing

the values of and in real time using separate frequencyestimators as in [12], it is possible to update the state matrix ofthe filter and therefore re-tune the center lobes of the wholefilter

bank on-line, without any need to also update the steady-stateKalman filter gain K. A similar principle was recently justi-fied and applied successfully in [36] using a bank of tunableresonators.

IV. SALIENTFEATURES OFCANDIDATEALGORITHMS

The two families of algorithms discussed above have beenimplemented in Matlab/Simulink. In each family, we have a ten-tative P and M class solution. The class-P FIR solution relies on

the 2.2-cycle filter while the class-M solution is based on the4-cycle Taylorfilter, both illustrated in Fig. 5. To approximateclass-P and -M behavior with the Kalman filter approach, weuse a Kalman filter with and without an adaptive notch add-onat the interference frequency, respectively.

To demonstrate the effectiveness of the algorithms, we sub-jected the four PMU algorithms to basically all tests specified inC37.118.1, following a methodology similar to [8] and [12]. Weused Simulink SimPower-Systems three-phase programmablesources to generate high sampling rate data, including har-monics and interferences as defined in C37.118.1. These testsignals are generated at four times

the algorithm sampling rate. The 4th-order anti-alias Butter-worth filter in Fig. 3 is then applied to the synthesized signals,followed by a decimation by four, in order to set the algorithminput sampling rate at points per cycle of the nominalfrequency (i.e., 1440 Hz.). Therefore, the filtering is adaptive

but not the sampling.The comparative step responses are illustrated in Fig. 8, with

some selected metrics shown in Table II. The group delay of thealgorithms, in the last line of this Table, is an important deter-minant of the PMU latency. As expected, the C37P and C37Mdesigns have predictable delays of 1 and 2 cycles (about halfthe processing window size). It should be noted that the group

delays of Kalmanfi

lter-based PMUs are not integer numbers ofsamples but are quite short compared to the baseline. Note thatthe TVE is computed by lagging the theoretical phasor (sam-

pled at 1440 Hz) by an amount of samples equal to the groupdelay of the studied PMU shown in Table II, before taking thedifference between the theoretical and measured phasor values.

The next three figures report steady-state behavior at funda-mental (Fig. 9), interference (Fig. 11), and harmonic (Fig. 10)frequencies, respectively. Fig. 9 was obtained by simulating thealgorithms behavior when subjected to unit magnitude steadystate voltage of varying fundamental frequency ranging from 0( Hz offset) to 120 Hz ( offset). It can be con-cluded that all proposed algorithms meet the 1% TVE require-ment of C37.118.1 over the range of offset frequency,assuming a 60 Hz fundamental frequency. But the FIR-M filter


7/13


Fig. 8. PMU algorithm responses to 10% magnitude and 10 phase step re-sponse at fundamental frequency.

TABLE IISTEPRESPONSECHARACTERISTICS (TIME INSECONDS)

is quite special since it is entirelyflat from dc to harmonics, witha TVE below 1% throughout the whole range.

Results in Fig. 10 were obtained by subjecting the algorithms

to a steady-state unit-magnitude voltage with a single 10% har-monic added. Varying the fundamental frequency ( to 75Hz) and harmonic rank (1 to 11) in two overlapping loops, theTVE was computed for all combinations of these two parame-ters. Fig. 10 then shows the TVE as level of color (from bluefor low TVE to red for high TVE) with respect to all possible

pairs of offset frequency and harmonic rank. On the top of eachsubplot which corresponds to one specific algorithm, the max-imum TVE over the full range of studied parameters is shown.These numbers confirm that the two proposed M-Class filtersare thoroughly immune against changing harmonics, i.e., har-monics at non-fundamental frequency. Their performance ex-ceeds by a widemargin of 0.1% the C37.118.1 specification,which mandates 1% TVE at fundamental frequency only. Inter-estingly, Fig. 10 shows that the EKF-P PMU (maximum TVE

Fig. 9. PMU algorithm steady-state responses at offset frequency.

Fig. 10. Steady-state TVE under harmonics with varying frequency offset.

of 0.11153%) is better than FIR-M PMU (maximum TVE of0.11159%) for rejecting off-nominal harmonics.

Fig. 11 was obtained by simulating the algorithm TVE andnoise rejection gain when subjected to a steady-state unit mag-nitude voltage at fundamental frequency, with a 10% sinusoidal

additive interference superimposed. The interfering frequencyis varied in 1-Hz step from 1 Hz to 240 Hz. Results shown inFig. 11 allow to verify that the two M-filters do in fact meetthe 40-dB attenuation requirement of C37.118.1 at Nyquist fre-quency. In addition, the 1% TVE threshold is met, even wellabove the Nyquist frequency of 30 Hz.

In fact, for the Kalman filter-based PMU, the narrow-bandmagnitude attenuation is 60 dB for both P and M class designs,which is actually the performance required for a high-sensitivityand high-impact SPS such as that in [27]. Table II includes twoTVE-based response-time criteria set at 1% and 2%, respec-tively. It appears that the performance of the Kalman filter issatisfactory when the criterion is 2%. By contrast, it is three tofour times slower than required by Std. C37.118.1 when the cri-terion is 1%. In general, we can conclude that the FIR-M and


8/13


Fig. 11. PMU algorithm steady-state responses to sinusoidal interference At30 or 90 Hz: maxTVE( ); maxTVE( ).

EKF-M algorithms meet boththe BPA and Hydro-Qubecs stepresponse requirements.

V. FREQUENCY RESPONSES TO SMALL-AMPLITUDESINUSOIDALMODULATIONS

When performing control-tuning studies in power systemstability software, the engineer usually develops an equiva-lent model for the PMU. This model should be compatiblewith a positive-sequence representation of the network in thefrequency range of 05 Hz, which is the range covered bystability software. In addition, the equivalent should capturethe small-signal behavior of the PMU within 05 Hz in terms

of magnitude and phase lag, while having a 3-dB frequencyband, which is representative of the actual device.

The step responses of Fig. 8 can be used to develop thesedynamic models. However, an alternative approach, whichalso allows for checking the frequency domain specificationin Table I, determines the frequency responses to amplitudeand phase modulation signals by simulation of the PMU algo-rithms. These tests are performed separately using sinusoidswith 0.01 pu and 0.01 rad peak magnitude respectively;the results are presented in Figs. 12 and 13. For easy inter-

pretation, the phase and amplitude of the frequency responsesat critical frequency are shown in Table III, together with the

TVE computed with a 10% (and 0.1 rad) sinusoidal modulationas required by C37.118.1 [1]. For comparison purposes, thesame metrics are included for a reference design obtained bycascading a frequency-compensated DFT phasor with the Pand M class filters suggested in C37.118.1. The gain and phaseof the 40-ms time constant are also shown below Table III.

The main observation from Figs. 12 and 13 and Table III isthat, while FIR-based solutions have the same phase (degree)and magnitude response, the EKF-based solutions have a flatamplitude but a resonant phase response. It also appears that the

phase lag of the FIR-M filter closely follows that of a 40-ms timeconstant in the frequency range 04 Hz. By contrast, EKF-Mdoes the same up to 3 Hz only for both amplitude and phasemodulations. But interestingly, at 1 Hz, the EKF-M has a 4advantage over a pure 40-ms time constant. This increases to

Fig. 12. PMU frequency responses to 1% amplitude sinusoidal modulation.Gain and phase responses computed by injecting a single frequency excitationat a time, and taking the ratio of the fast-Fourier transforms of the algorithm

input and output signals.

Fig. 13. PMU frequency responses to 0.01 rad (or 0.57) sinusoidal phase mod-ulation. Gain and Phase computed for each frequency at a time.

10 for the EKF-P filter, which can be very useful in certaincircumstances, such as for a power plant PSS. Finally, the TVEmetrics of phase modulation at 5 Hz shows that all algorithmscan meet the bandwidth requirements set in C37.118.1 for classM accuracy.

VI. FAULTRESPONSETESTS

To test the algorithms under more realistic network condi-tions, we picked a test case used in [27] during the design andcommission of an SPS [38] which is now a major component ofthe defense plan [26]. The results are shown in Fig. 14.

The challenge here is that during a frequency ramp, thevoltage (which needs to be accurately measured in [27]) is sub-

jected to a so-called intermediation phenomenon, manifestingitself through fictitious and sustained voltage oscillations at6 Hz. The EKF based algorithm with interference tracking andnotching is the only solution to remove these perturbations


9/13


TABLE IIIFREQUENCYRESPONSECHARACTERISTICS (PHASE INDEGREES)

Fig.14. Analysis of a parametriccaseusedfor designingthe SPSsin [38].Post-fault response with 4-Hz/s frequency ramp, 15% sub-synchronous resonancewith voltage magnitude intermediation at frequency.

without increasing the 95% time response of the filter up to say200 ms. Naturally, these spurious oscillations also corrupt theROCOF signal, even though the frequency dynamic range doesnot allow visualizing them on the frequency ramp plot. Again,despite its good interference attenuation shown in Fig. 11, theFIR-M still come short of polishing the ROCOF to the levelsuitable for crisp decision making.

To further assess whether a C37.118.1-compliant PMUcould meet utilities requirements for wide-area control, wetested three commercial PMUs on a laboratory setup [8], [9].

Fig. 15. Frequency responses of three commercial PMUs.

The testing was not aimed at checking their compliance withC37.118.1 in terms of TVE-based metrics, since this wouldhave required a well calibrated GPS-based source. Instead,our setup is essentially a relay-testing environment with the

goal of playing back synthesized parametric waveforms orCOMTRADE files of network events recorded in the field oron our Hypersim real-time transient-network simulator.

The frequency responses of these PMUs are given in Fig. 15.To facilitate comparisons, we selected for each PMU brand, thesettings closest to the M class performance level. But only one

brand (C) offered a specific M class selection. According toFig. 15, all three PMUs have a good frequency range of flatmagnitude around the fundamental and a good attenuation ofout-of-band interference. Since the output rate is 60 Hz, the

Nyquist frequency rejection dB) is very close to the re-quirements of boththe utilities and C37.118.1 in this regard.No-

tice however that the frequency response of PMU-A seems left-shifted. After further verification, we confirmed that this is thecorrect behavior of the device. The center of its frequency rangeseems to be around 50 Hz, whatever the nominal fundamentalfrequency selected. Notice that, the corresponding out-of-bandinterference frequency response is not available at the momentto confirm this hypothesis.

Next, a simple startup test consisting in suddenly applyingthe nominal voltage at off-nominal frequency was performedon the three devices and their results compared with the EKF-Mfilter of previous sections. Fig. 16 shows some interesting factsabout this test (obtained by playing back the same waveformwith the EKF-M Simulink model). Firstly, we observed that thetwo PMUs start at zero, suggesting that their output frequencyis 60 Hz when there is no voltage while the EKF-M algorithm


10/13


Fig. 16. (a)Sinusoidalstep: fromzero to nominalvoltageat 63 Hz fundamentalfrequency. Top:initial time frame; Bottom: full time frame . Frequency responsetime at 95%: A (N/A), B (75 ms), C (90 ms). (b) Sinusoidal step: from zeroto nominal voltage at 63 Hz fundamental frequency. Top: initial time frame;Bottom: full time frame . Voltage response time at 95%: A (60 ms), B (55 ms),C (70 ms).

and the brand C PMU produce 0 Hz output under the same con-ditions. Secondly, the PMU-C has a large frequency overshoot

Fig. 17. PMU responses to network event: fault at Nemiscau (James Bay) withthe PMU near Montreal.

with no amplitude overshoot whereas the PMU-B has no over-shoot but the output is delayed by 0.2 s.

The behavior of PMU-A is even more delayed, as the 63-Hztarget is still way out of reach 1 s after startup (actually, a zoomof the figure shows that the PMU-A posts a 60.5-Hz frequencyat 1.5 s but increases slowly towards the target). In other words,except for the PMU-A, the amplitude and frequency responsetimes for a sinusoidal step at off-nominal frequency are closeto utilities specifications. However, since the TVE is not avail-able, we cannot draw any conclusion about the C37.118.1 spec-ified response times for the M-filter.

The full time-frame plots confirm that the PMU A has a sig-nificant error throughout the recorded period but the error isslowly decreasing. Using other tests at off-nominal frequency,

we found that in fact, PMU A is just very slow in respondingwhen the initial frequency far from the nominal frequency. Infact, the frequency tracking performance is very poor and well

below the 1 Hz/s required in Std. C37.118.1. This is a majorpitfall which has been communicated to the vendor who agreedwith the pattern shown in Fig. 16. The issue will be corrected ina future release of this device.

Fig. 17 illustrates the PMUresponses to a network faultevent.The only noticeable fact on the voltage response is the discrep-ancy in time response between the various solutions. PMU-C ischaracterized by an initial undershoot of a non-minimal phaseresponse type while the step response overshoot of PMU-B isevident after fault clearing. Regarding the frequency response,the PMU-B and C exhibit spurious behavior during the fault pe-riod while PMU-A remains on idle for a long time. Interestingly,


11/13


Fig.18. PMU responses to a sub-synchronous responseparametric caseknownas design case No f0014. Steady state voltage error: A-B-C: 5%; EKF-M:0.2%. Steady state Frequency error: A: 50 mHz; B: 12 mHz; C: 50 mHz;EKF-M: 1 mHz.

these observations are also in line with Fig. 16. However, noneof the behavior shown by PMU-A, -B or -C is appropriate fromthe wide-area control point-of-view.

In each case, the control could overreact and produce moreharm than good if it wrongly assumes that the PMU frequencysignal is correct during the fault. These laboratory tests con-firm that, under dynamic conditions, most commercial PMUstend to perform as specified in C37.118.1 regarding the am-

plitude response, although excess overshoot could still be anissue for some PMUs (B). Regarding frequency tracking, [17]showed that none of the five PMUs tested was able to meet allC37.118.1 dynamic-response requirements. The present test re-sults demonstrate that one important pitfall of commercial PMU

in the context of wide-area control is related to phase shift andfrequency tracking, during and shortly after network events. Infact, when the voltage vanishes for any reason, the phasor isnot well defined and each PMU addresses this ambiguity dif-ferently. In the example shown, PMU-C does nothing, PMU-Ais too cautious, PMU-B offers a strategy between A and C,while EKF-M algorithm is the best trade-off overall. Alternativeschemes to deal with this issue, not addressed in C37.118.1, can

be found in [17] and [22][24].It has been argued that the requirements in [27] are too strin-

gent, and counterintuitive. However, as a matter of fact, severalSPS has been implemented in the Hydro-Qubecs networkusing non-synchrophasor based measurements units which metthese requirements [28], [38]. One design case used for as-sessing these existing SPS has been selected to benchmark thecommercial PMUs A-B-C with respect to their ability to meetthe requirements in Section II without further development. Thecorresponding waveform and positive sequence voltage andfrequency measurements are shown in Fig. 18. The waveform

on the top plot was played back on the Hypersim simulatorto obtain PMU A-B-C results. It was also played back on theEKF-M Simulink model in Matlab to compare the proposedalgorithm with the three state-of-art PMUs from major vendors.

The resultsin Fig. 18 show quite clearly that thethree existingPMUs are not suitable to deal with some of the phenomena inHydro-Qubecs SPS requirements of Section II: they displaya 5% amplitude error and between 12 and 50-mHz frequencyerror in the post-fault period. By contrast, the amplitude and fre-quency errors ascribed to the EKF-M PMU algorithm are only0.2% and 1-mHz, respectively.

VII. CONCLUSIONSThis paper presented two families of PMU algorithms able

to meet the dynamic and static accuracy requirements of Std.C37.118.1. Further, it highlighted possible gaps and difficultiesthat may happen when reconciling the TVE-based criteria ofC37.118.1 with the utilities filtering requirements, which areusually expressed in terms of frequency response to modulationin amplitude, phase and frequency signals or control system-

based step-response information. It appears, for instance, that a1% TVE stabilization criterion may be too stringent comparedto a 2% stabilization response time specified by most utilitiesfor control system PMUs.

In addition, the laboratory testing of three PMU brandsshowed that Class M PMUs from different vendors will likelyprovide inconsistent measurements of dynamic activity [8].The three PMUs A, B and C tested on the same step excitationsresulted in quite different dynamic behavior, even though theirsettings were selected to be as close as possible to Class M spec-ification according to each vendor. It was also found that thePMU performance for frequency tracking during faults varieda lot from brand to brand. Unfortunately, the TVE criteriondoes not characterize the above inconstancies in engineeringterms, and it cannot be used to compensate them during dataalignment. A possible improvement in this regard could be tocharacterize the amplitude and phase errors patterns separately[19], [20] and further mandate PMU vendors to provide thegroup delay of their PMU to facilitate data alignment at Phasor


12/13


Data Concentrator (PDC) level. A last complication with usingTVE concept is that direct TVE measurements requires GPSsynchronization, which might be unavailable to an otherwisesufficient test facility such as a typical power utility protectionlaboratory or even the world class IREQs transient networksimulator. In this context, it appears that TVE complianceassessment will be performed by large specialized instrumenta-tion laboratories, not by the PMU end users.

Overall, it was found that the FIR bandpass filtering PMUwas a perfect match to the current C37.118.1, as it exceeded allof its P and M accuracy class requirements, with the advantageofflat step responses, and linear phase. However, the adaptiveKalman filterwas evidently the best compromise in narrow bandapplications with changing harmonics and torsional or sub-syn-chronous phenomena. It was able to address effectively transientresponse to faults while meeting or exceeding other C37.118.1requirements.

ACKNOWLEDGMENT

The first author would like to thank all his Hydro-Qubecscolleagues who have contributed to this work directly andindirectly through data and knowledge sharing. Specialsthanks are extended to C. Lafond, M. Perron, C. Cyr, andJ. Blandfrom IREQ, who performed the laboratory tests in arather short period of time. The leadership role of D. McNabbfor preparing numerous test cases and initiating early re-search projects aiming at assessing the practical feasibility ofHydro-Qubecs requirements for SPS measurement units isgratefully acknowledged.

REFERENCES

[1] IEEE Standard for Synchrophasor Measurements for Power Systems,IEEE Std. C37.118.1-2011, Dec. 28, 2011.

[2] Project Authorization Request (PAR) for a New IEEE Standard: PARNo. PC37.242, Guide for Synchronization, Calibration, Testing, andInstallation of Phasor Measurement Units (PMU) for Power SystemProtection and Control, Sep. 30, 2010. [Online]. Available: https://de-velopment.standards.ieee.org/get-file/PC37.242.pdf?t=10380700003.

[3] I. Kamwa, G. Trudel, and L. Grin-Lajoie, Multi-loop power systemstabilizers using wide-area synchronous phasor measurements, in

Proc. 199 8 Amer . Con trol Conf. , Jun. 2126, 1998, pp. 29632967.[4] P. Denys, C. Counan, L. Hossenlop, and C. Holwek, Measurement

of voltage phase for the french future defence plan against losses ofsynchronism, IEEE Trans. Power Del., vol. 7, no. 1, pp. 6269, Jan.1992.

[5] I. Kamwa, J. Bland, G. Trudel, R. Grondin, C. Lafond, and D.McNabb, Wide-area monitoring and control at Hydro-Qubec:

past, present and future, in Proc. 2006 IEEE/PES General Meeting,Montreal, QC, Canada, Jun. 1822, 2006.[6] C. W. Taylor, C. Erickson, K. E. Martin, R. E. Wilson, and V.

Venkatasubramanian, WACS-wide-area stability and voltage controlsystem: R&D and on-line demonstration,Proc. IEEE, vol. 93, no. 5,pp. 892906, May 2005.

[7] L. Peng, W. Xiaochen, L. Chao, S. Jinghai, H. Jiong, H. Jingbo, Z.Yong XuAidong, and X. Aidong, Implementation of CSGs wide-areadamping control system: Overview and experience, in Proc. IEEE/

PES, Power Syst. Conf. & Expo., PSCE 09, Mar. 1518, 2009, pp.19.

[8] Z. Huang, T. Faris, K. Martin, J. Hauer, C. Bonebrake, and J. Shaw,Laboratory Performance Evaluation of SEL 421 Phasor MeasurementUnit, PNNL Report-16852, Pacific Northwest National Laboratory,Dec. 2007. [Online]. Available: http://www.pnl.gov/main/publica-tions/external/technical_reports/PNNL-16852.pdf.

[9] F. Steinhauser and Y. Ping, Testing the dynamic response of

phasor measurement units, in Proc. Int. Conf. Elect. Eng., pp. 15.[Online]. Available: http://www.icee-con.org/papers/2009/pdf/2.02_I9CP0546_E.Pdf.

[10] J. F. Hauer, W. A. Mittelstadt, K. E. Martin, J. W. Burns, and H.Lee, Integrated dynamic information for the Western power system:WAMS analysis in 2005, in Power System Stability and Controlvolume of the Electric Power Engineering Handbook, L. L. Grigsby,Ed., 2nd ed. Boca Raton, FL: CRC Press, 2007, ch. 14.

[11] PMU System Testing and Calibration Guide. NASPI Report of the Per-formance & Standards Task Team (PSTT), Dec. 30, 2007. [Online].Available: https://www.naspi.org/File.aspx?fileID=555.

[12] I. Kamwa, A. K. Pradhan, and G. Joos, Adaptive phasor and fre-

quency-tracking schemes for wide-area protection and control,IEEETrans. Power Del., vol. 26, no. 2, pp. 744753, Apr. 2011.

[13] M. Karimi-Ghartemani, Boon-TeckOoi, and A. Bakhshai, Applica-tion of enhanced phase-locked loop system to the computation of syn-chrophasors, IEEE Trans. Power Del., vol. 26, no. 1, pp. 2232, Jan.2011.

[14] R. K. Mai, L. Fu, Z. Y. Dong, B. Kirby, and Z. Bo, An adaptive dy-namic phasor estimator considering DC offset for PMU applications,

IEEE Trans. Power Del., vol. 26, no. 3, pp. 17441754, Jul. 2011.[15] D. Kosterev, Dynamic performance requirements for phasor measure-

ment units, in Proc. NAPSI Meeting, Feb. 2010. [Online]. Available:https://www.naspi.org/File.aspx?fileID=138.

[16] D. Trudnowski, PMU dynamic requirements for small-signal mea-surements, in Proc. NASPI Meeting, Chattanooga, TN, Oct. 2009.[Online]. Available: https://www.naspi.org/File.aspx?fileID=562.

[17] J. Ren, Synchrophasor measurement using substation intelligent elec-tronic devices: algorithms and test methodology Ph.D. dissertation,

Elect. Eng. Dept., Texas A&M Univ., College Station, 2011.[18] A. Goldstein, PMU Model C37.118.1 NASPI, Jun. 2011. [Online].Available: https://www.naspi.org/File.aspx?fileID=419.

[19] G. Y. Yang, K. E. Martin, and J. stergaard, Investigation of PMUperformance under TVE criterion, inProc. 2010 5th Int. Conf. CriticalInfrastructure (CRIS), Sep. 2022, 2010, pp. 17.

[20] M. Lixia, C. Muscas, and S. Sulis, On the accuracy specificationsof phasor measurement units, in Proc. 2010 IEEE Instrum. Meas.Technol. Conf. (I2MTC), May 36, 2010, pp. 16.

[21] P. Castello, C. Muscas, and P. A. Pegoraro, Performance comparisonof algorithms for synchrophasors measurements under dynamic condi-tions, inProc. IEEE Int. Workshop Applied Measurements for PowerSyst. (AMPS), Sep. 2830, 2011, pp. 2530.

[22] J. Warichet, T. Sezi, and J. C. Maun, Considerations about syn-chrophasors measurements in dynamic system conditions, Int. J.

Elect. Power Energy Syst., pp. 113, 2009.[23] A. G. Phadke and B. Kasztenny, Synchronized phasor and frequency

measurement under transient conditions, IEEE Trans. Power Del.,vol. 24, no. 1, pp. 8995, Jan. 2009.[24] J. Ren and M. Kezunovic, Real time power system frequency and

phasor estimation using recursive wavelet transform, IEEE Trans.Power Del., vol. 26, no. 3, pp. 13921402, Jul. 2011.

[25] I. Kamwa, R. Grondin, and D. McNabb, Changing harmonics instressed power transmission systemsApplication to Hydro-Quebecsnetwork, IEEE Trans. Power Del., vol. 11, no. 4, pp. 20202027,Oct. 1996.

[26] G. Trudel, J. P. Gingras, and J. R. Pierre, Designing a reliable powersystem:Hydro-Quebecs integrated approach,Proc. IEEE,vol.93,no.5, pp. 907917, May 2005.

[27] J. Lambert, D. McNabb, and A. G. Phadke, Accurate voltage phasormeasurement in a series-compensated network, IEEE Trans. Power

Del., vol. 9, no. 1, pp. 501509, Jan. 1994.[28] P. Cote, S.-P. Cote, and M. Lacroix, Programmable load shed-

ding-systems-Hydro-Quebecs experience, in Proc. 2001 IEEE PES

Summer Meeting, Vancouver, BC, Canada, Jul. 1519, 2001, vol. 2,pp. 818823.[29] I. Kamwa, R. Grondin, and G. Trudel, IEEE PSS2B versus PSS4B:

The limits of performance of modern power system stabilizers, IEEETrans. Power Syst., vol. 20, no. 2, pp. 903915, May 2005.

[30] R. R. Bitmead, A. C. Tsoi, and P. J. Parker, A Kalman filtering ap-proach to short-time Fourier analysis, IEEE Trans. Acoust., Speech,Signal Process., vol. 34, no. 6, pp. 14931501, Dec. 1986.

[31] R. G. Lyons, Understanding Digital Signal processing, 3rd ed. En-glewood Cliffs, NJ: Prentice Hall, 2011.

[32] R. Lira, D. H. Wilson, K. Hay, T. Cumming, and D. Cole, Testing syn-chrophasor data quality for applicability to transmission security andoptimisation tools, inProc. 17th Power Syst. Comput. Conf., Stock-holm, Sweden, Aug. 2011, pp. 2226. [Online]. Available: http://www.pscc-central.org/uploads/tx_ethpublications/fp414_01.pdf.

[33] M. Paolone, A. Borghetti, and C. A. Nucci, A SynchrophasorEstimation Algorithm for the Monitoring of Active Distribution Net-

works in Steady State and Transient Conditions, ibid, Jul. 15, 2012.[Online]. Available: http://www.pscc-central.org/uploads/tx_ethpubli-cations/fp221.pdf.


13/13


[34] I. Kamwa, M. Leclerc, and D. McNabb, Performance of demodula-tion-based frequency measurement algorithms used in typical pmus,

IEEE Trans. Power Del., vol. 19, no. 2, pp. 505514, Apr. 2004.[35] D. Alves and R. Coelho, An adaptive algorithm for real-time

multi-tone estimation and frequency tracking of non-stationary sig-nals, IEEE Trans. Nucl. Sci., vol. 58, no. 4, pp. 15821587, Aug.2011.

[36] M. D. Kuljevi, Simultaneous frequency and harmonic magnitudeestimation using decoupled modules and multirate sampling, IEEE

Trans. Instrum. Meas., vol. 59, no. 4, pp. 954962, Apr. 2011.[37] S. Shinnaka, A novel fast-tracking D-estimation method forsingle-phase signals,IEEE Trans. Power Electron., vol. 26, no. 4, pp.10811088, Apr. 2011.

[38] S. Bernard, G. Trudel, and G. Scott, A 735 kV shunt reactors au-tomatic switching system for Hydro-Quebec network, IEEE Trans.

Power Syst., vol. 11, no. 4, pp. 20242030, Nov. 1998.

Innocent Kamwa (S83M88SM98F05)received the Ph.D. degree from Laval University,Qubec City, QC, Canada, in 1988.

He then joined Hydro-Qubecs research institute(IREQ), where he is currently project manager forpower grid control and automation. He is also thechief scientist for Hydro-Qubecs smart grid. He is aP. Eng. and Adjunct Professor of Power Systems En-gineering at McGill University and Laval University.

Dr. Kamwa serves on many IEEE/PES technicalcommittees as member and officer, including the

Fellow Evaluation, Electric Machinery, and Power System Stability commit-tees. He is an Editor of the IEEE TRANSACTIONS ON POWERSYSTEMS andCo-Editor-in-Chief of IET Generation, Transmission and Distribution. He wasawarded the IEEE PES Prize Paper Award in 1998, 2003, and 2009.

S. R. Samantaray (M08SM10) received theB.Tech. degree in electrical engineering from UCEBurla, India, in 1999 and the Ph.D. degree in powersystem engineering from the Department of Elec-tronics and Communication Engineering, NationalInstitute of Technology, Rourkela, India, in 2007.

He holds the position of Assistant Professor inthe School of Electrical Sciences, Indian Institute ofTechnology Bhubaneswar, India. He visited the De-partment of Electrical and Computer Engineering,McGill University, Montral, QC, Canada, as a

Post-Doctoral Research Fellow and Visiting Professor. His major researchinterests include intelligent protection for transmission systems (includingFACTs) and microgrid protection with distributed generation and dynamicsecurity assessment in large power networks.

Dr. Samantaray is the recipient of the 2007 Orissa Bigyan Academy YoungScientists Award, the 2008 Indian National Academy of Engineering Best Ph.D.Thesis Award, the 2009 Institute of Engineers (India) Young Engineers Award,the 2010 Samanta Chandra Sekhar Award, and the 2012 IEEE PES TechnicalCommittee Prize Paper Award. He is an Editor ofIET, Generation, Transmission& Distributionand Electric Power Components and Systems.

Geza Joos (M82SM89F06) received the

M.Eng. and Ph.D. degrees from McGill University,Montral, QC, Canada.

He has been a Professor with McGill Universitysince 2001 and holds a Canada Research Chair inpower electronics applied to power systems. He isinvolved in fundamental and applied research relatedto the application of high-power electronics to powerconversion (including distributed generation andwind energy) and to power systems. He was previ-ously with ABB, the cole de technologiesuprieure,

and Concordia University. He is involved on a regular basis in consultingactivities in power electronics and power systems.

Dr. Joos is active in a number of IEEE Industry Applications Society com-mittees as well as IEEE Power Engineering Society working groups and CIGREworking groups. He is a Fellow of the Canadian Academy of Engineering andof the Engineering Institute of Canada.

compliance analysis of pmu algorithms and devices for wide-area stabilizing control of large power...

Documents