Taylor-Fourier_PMU_on_a_Real-Time_Simulator_Design_Implementation_and_Characterization

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Taylor-Fourier PMU on a Real-Time Simulator:

Design, Implementation and Characterization


G. Frigo∗ , A. Derviškadić∗ , Y. Zuo∗ , A. Bach† , M. Paolone∗
∗ École Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland
† École Normale Supérieure de Cachan, Université Paris-Saclay, Paris, France
mail: {guglielmo.frigo, asja.derviskadic}@epfl.ch

Abstract—The development of software models of Phasor and currents. In this context, the integration of PMUs can be
Measurement Units (PMUs) within Real-Time Simulators (RTSs) carried out according to two alternative options.
represents a promising tool for the design and validation of In the first option, the RTS is interfaced with real PMUs
monitoring and control applications in electrical power networks.
In this sense, it is necessary to find an optimal trade-off between by means of analog output channels, capable of reproducing
computational complexity and estimation accuracy. In this paper, power signals. Through a Local Access Network (LAN),
we present the design and implementation of two new PMU PMUs are connected to a Phasor Data Concentrator (PDC) that
models within the Opal-RT eMEGAsim RTS. The synchrophasor collects the measurement data and aligns them based on the
estimation algorithm relies on a Compressive Sensing Taylor- corresponding time-stamp [7]. This option enables us to test
Fourier Model (CS-TFM) approach, and enables us to extract the
dynamic phasor associated to the signal fundamental component. the actual device (not limited to the synchrophasor extraction
The estimation accuracy of the proposed models is characterized algorithm) and to reduce significantly the complexity of the
with respect to the compliance tests of the IEEE Std. C37.118.1. RTS model. However, this approach is characterized by several
drawbacks: (i) the uncertainty characterization of the RTS
Index Terms—Real-Time Simulator (RTS), Synchrophasor, power signals, (ii) the limited number of available analog
Taylor-Fourier Transform, Phasor Measurement Unit (PMU)
output channels, (iii) the cost of real PMUs, (iv) the amount of
cabling. Due to these limitations, this solution cannot support
I. I NTRODUCTION
large-scale network models.
In recent years, Phasor Measurement Units (PMUs) have In the second option, instead, the PMUs are simulated and
become an effective and reliable solution for the monitoring included within the RTS model. In this way, the validation
and control of electrical power networks [1], [2]. In this process is independent from the device and accounts also
context, the IEEE Std C37.118.1 [3] and its recent amendment for the transfer function of the instrument transformers. It
[4] (briefly IEEE Std) define PMUs as high-accuracy devices is possible to reproduce large-scale networks and validate
for synchronized phasor and frequency measurements. In monitoring and control applications relying on numerous (e.g.
particular, PMUs are required to characterize the fundamental hundreds) PMUs. The only limitation is represented by the
tone in terms of phasor amplitude and phase, frequency, and computational resources of the RTS. As a consequence, the
Rate-of-Change-of-Frequency (ROCOF). Furthermore, PMU complexity of the PMU model should be possibly reduced, in
measurements are phase-aligned to a common time-reference order to find an optimal trade-off between estimation accuracy
synchronized with Universal Time Coordinates (UTC). and computational requirements [8].
Before the actual deployment into the field, PMUs have Inspired by a similar analysis in [8], in this paper, we present
to be thoroughly validated and characterized in terms of the design, implementation and characterization of two new
estimation accuracy and reporting latency with respect to PMU models within the Opal-RT eMEGAsim RTS [9]. In
both IEEE Std tests and expected operating conditions. Also, particular, we adopt a synchrophasor estimation algorithm,
before their deployment into real power networks, PMU- that has been derived by the recent Compressive Sensing
based control and protection schemes have to be validated in Taylor-Fourier Model (CS-TFM) [10]. To the best of Authors’
terms of reliability and effectiveness. Since an experimental knowledge, this represents the first attempt to implement a
validation might be partially uncontrollable or even unfeasible, PMU model relying on a dynamic phasor formulation [11] and
the employment of a Real Time Simulator (RTS) enables us to thus capable of tracking the signal time-variations and directly
reproduce a wide range of realistic operating conditions and computing ROCOF as one of the system state variables.
therefore evaluate the reliability of PMU-based applications The RTS results rely on the assumption that PMU models
even during critical scenarios [5], [6]. In this context, the are characterized by the same accuracy and responsiveness of
development of PMU software models might become a the actually deployed devices. For this reason, we evaluate
promising solution as they allow for reproducing large-scale its estimation accuracy in terms of Total Vector Error (TVE),
applications involving multiple PMUs. Frequency Error (FE) and ROCOF Error (RFE), and we assess
Once defined the electrical network under investigation, the its computational complexity in terms of maximum number of
RTS reproduces the system behavior in terms of nodal voltages PMUs per RTS core and computation time. For this analysis,

‹,(((

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SRINAGAR. Downloaded on December 16,2024 at 04:43:42 UTC from IEEE Xplore. Restrictions apply.
we consider the IEEE Std tests, with specific reference to the solutions, though, are prone to errors since they rely on the
Protection and Measurement class requirements, referred to unverified assumption that the fundamental frequency is slowly
as P- and M-class respectively, as well as an off-standard test varying between two consecutive observation windows. In the
inspired by real-world operating conditions [12]. presence of sudden parameter changes, PLL and KF introduce
The paper is organized as follows. Section II introduces a filtering effect (with delayed and smoothed transitions) that
the algorithm theoretical background. Section III describes the might result in a significant accuracy degradation [20].
implementation within the Opal-RT simulation environment. In this paper, as in [10], we recover the spectral support
Section IV presents a thorough performance characterization. S through an Orthogonal Matching Pursuit (OMP) algorithm,
Finally, Section V provides closing remarks. i.e. a greedy selection routine that exploits the assumption that
the signal spectrum is sparse and consists only of a limited
II. T HEORETICAL BACKGROUND number of narrow-band components. In particular, we apply a
Let us consider a generic time-varying power signal, CS-based super-resolution technique to improve the frequency
affected by both narrow- and wide-band disturbances: resolution associated to short observation intervals by almost
one order of magnitude (further details in [21]).
x(t) = A · (1 + εA (t)) · cos(2πf t + ϕ + εϕ (t)) + η + ρ (1) Given the recovered support S, we design the corresponding
where A, f and ϕ are the amplitude, frequency and TFM model M and we compute the dynamic synchrophasor
initial phase of the fundamental component, respectively. The p that consists of three Taylor-Fourier series coefficients:
time-varying terms εA and εϕ account for amplitude and
p = {p0 , p1 , p2 } = pinv(M ) · x (2)
phase modulations, η represents any spurious component,
and ρ models the measurement uncertainty as an additive where the superscript denotes the derivative order. Based on
uncorrelated Gaussian random variable. this, we are able to estimate the fundamental parameters as:
Typically, a PMU considers an observation window of finite
length Nw and computes its Discrete Fourier Transform (DFT) Â = |p0 |, Â1 = 2(p1 · e−j ϕ̂ )
X(f ) in order to extract the synchrophasor associated to (p1 · e−j ϕ̂ )
the fundamental component [13]. However, the DFT-based ϕ̂ = ∠p0 , ϕ1 = (3)

approach relies on a static signal model and cannot provide an ϕ̂1 (p2 · e−j ϕ̂ ) − Â1 · ϕ̂1
optimal representation of dynamic conditions. For this reason, fˆ = fˆ0 + , R̂f =
2π 2π · Â
the recent literature has discussed the employment of Taylor
series expansions, whose higher-order derivative terms might where R̂f denotes the ROCOF, and f0 is the fundamental
better account for fundamental time-varying parameters [11]. frequency within the recovered spectral support.
In this context, the CS-TFM algorithm adopts a formulation
of the Taylor-Fourier Transform (TFT) [14], that has been A. Modulated Sliding DFT
suitably modified and generalized in order to deal also with
In terms of measurement reporting latency, the DFT
multi-tone power signals. In its original formulation, the TFT
computation requires a significant amount of processing time,
has been presented as a maximally flat differentiator filter,
as its complexity O(Nw2 ) can be reduced up to O(Nw log Nw )
centered around the nominal system frequency1 [15]. Thanks
with the Fast Fourier Transform algorithm [22]. Nevertheless,
to a Taylor series expansion truncated to the second derivative
the synchrophasor analysis does not consider the entire
order, it is possible to include the fundamental frequency
spectrum, but focuses on a narrow bandwidth centered around
and ROCOF within the estimator state variables as the first
the nominal system frequency. Therefore, it is possible to limit
and second time-derivative of the phase angle, respectively.
the number of DFT bins to be computed. In particular, the
However, the TFT performance strongly depends on the
proposed algorithm restricts its analysis up to the out-of-band
spectral support S employed for the filter positioning [16]:
(OOB) frequency range, i.e. from 5 to 100 Hz (see Sec. II.B)2 .
if any significant spectral component is neglected or badly
identified, the TFT results might suffer from uncompensated In this context, an effective solution for the DFT bins’
spectral leakage and thus lead to inaccurate estimates [17]. computation is represented by the Modulated Sliding DFT
For this reason, the identification of the signal spectral (MSDFT) algorithm [24], a recursive routine for computing
support S becomes one of the most crucial stages of the DFT bins on a sample-by-sample basis. In order to resolve
entire synchrophasor extraction algorithm. On the other hand, multiple spectral tones, we apply a Hanning weighing function
the processing resources destined to S identification should through a frequency-domain convolution that results in a linear
comply with the capabilities of the selected implementation combination of adjacent DFT bin values:
platform. Recent literature has proposed to track the Xk = −0.25 · Xk−1 + 0.5 · Xk − 0.25 · Xk+1 (4)
fundamental frequency by means of a Phase Locked Loop
(PLL) algorithm [18] or a Kalman Filter (KF) [19]. Similar 2 Given a system frequency of 50 Hz, we adopt the inter-harmonic group
formulation as defined in [23]. Nevertheless, according to IEEE Std, PMUs
1 Without loss of generality, in this paper we consider a nominal system should not consider the DC component, and the lowest sub-harmonic group
frequency of 50 Hz, but similar results can be obtained in a 60-Hz scenario. is typically centered around 5 Hz.

,(((0LODQ3RZHU7HFK

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SRINAGAR. Downloaded on December 16,2024 at 04:43:42 UTC from IEEE Xplore. Restrictions apply.
where the subscript k denotes the bin index. In this way, we computes 14 bins as representative of the fundamental
are able to significantly reduce the computational effort as well variation range [45, 55] Hz and the out-of-band3 frequency
as to guarantee the DFT stability and accuracy [25], [26]. range [5, 25] ∪ [75, 95] Hz. As shown in Algorithm 2, we first
project the signal spectrum over a finer grid whose resolution
B. Taylor-Fourier Model Algorithm
is of 0.909 Hz (line 3). Then, we identify the fundamental
As known, the IEEE Std introduces two PMU performance component as the maximum spectral bin (line 4). Based on its
classes: the P-class is intended for protection applications amplitude, frequency and phase, we subtract its contribution
and favors fast responsiveness rather than high accuracy, from the original spectrum (line 5) and we apply the super-
conversely M-class is intended for measurement applications resolution routine over the OOB range (line 6).
and requires high accuracy also in distorted conditions [3]. Within the residual spectrum, we select the maximum bin
In order to cope with these contrasting requirements, we (line 7): if its amplitude exceeds the significance threshold
develop two models specifically designed for P- and M-class θi , the corresponding frequency is included in the spectral
compliance, whose main processing steps are reported in support, otherwise S accounts only for the first four harmonic
Algorithm 1 and 2, respectively. terms (lines from 8 to 12). In concordance with the IEEE Std
a) P-class Model: The P-class PMU adopts a window requirements for the Out-of-Band interference test, we set θi
length of 60 ms, i.e. equal to three nominal cycles at 50 Hz equal to five percent of fundamental amplitude. In practice,
and corresponding to a frequency resolution of 16.67 Hz. though, this parameter has to be set according to the desired
In P-class configuration, we are not interested in detecting sensitivity to spurious tones and the expected noise level, in
and compensating possible inter-harmonic components. The order to avoid the inclusion of negligible components into S.
spectral analysis can be limited to the expected variation range Finally, based on the recovered spectral support, we
of the fundamental component, i.e. [45, 55] Hz. To this end, construct the TFM matrix (line 13), we compute the dynamic
the MSDFT provides a DFT representation X consisting of phasor coefficients (line 14), and we extract the fundamental
just 3 bins, centered at 33.33, 50 and 66.67 Hz, respectively. component parameters (line 15).
As shown in Algorithm 1, the first step consists in enhancing
the frequency resolution by projecting X over the vector space Algorithm 2 M-class Model
spanned by matrix Df . In more detail, the matrix columns are 1: M-class → window length 100 ms, 11 DFT bins
designed to account for leakage effects over a super-resolved 2: input: x, X, Df , Di , θi , output: Â, fˆ, ϕ̂, R̂f
grid, whose bin spacing is set to 1.515 Hz (line 3). 3: Y = Df · X

 super-resolved spectrum
We associate the fundamental frequency fˆ0 to the maximum ˆ
4: (f0 , Â0 , ϕ̂0 ) = max(Y )  fund. parameter estimates
bin of the super-resolved spectrum (line 4), and we include into ˆ
5: R = Y − Â0 ej2(π f0 +ϕ̂0 )  residual OOB spectrum
the spectral support S the first four harmonic terms (line 5). †
6: Y = Di · R  super-resolved spectrum
Based on this information, we construct the TFM matrix (line
7: (fˆi , Âi ) = max(Y )  inter-harmonic selection
6) and compute the corresponding dynamic phasor coefficients
8: if Âi ≥ θi
(line 7). Finally, by applying (3), we extract the fundamental
synchrophasor, frequency and ROCOF (line 8). 9: S = {fˆ0 · [1, 2, 3, 4], fˆi }  spectral support def.
10: else
Algorithm 1 P-class Model 11: S = {fˆ0 · [1, 2, 3, 4]}  spectral support def.
1: P-class → window length 60 ms, 3 DFT bins 12: end if
2: input: x, X, Df , output: Â, fˆ, ϕ̂, R̂f 13: M = T F M (S)  TFM matrix computation

3: Y = Df · X  super-resolved spectrum 14: p = (M † M )−1 M † · x  dynamic phasor extraction
15: fˆ0 , p → {Â, fˆ, ϕ̂, R̂f }  final parameter estimates
4: fˆ0 = max(Y )  fund. frequency estimate
5: S = {fˆ0 · [1, 2, 3, 4]}  spectral support definition
6: M = T F M (S)  TFM matrix computation
III. O PAL -RT I MPLEMENTATION
7: p = (M † M )−1 M † · x  dynamic phasor extraction
8: fˆ0 , p → {Â, fˆ, ϕ̂, R̂f }  final parameter estimates a) Real-Time Simulator (RTS): The development of a
PMU model within a RTS requires the implementation of two
b) M-class Model: Differently from P-class scenario, M- main functionalities: the synchrophasor estimation algorithm,
class PMUs are intended for higher accuracy and robustness in and the synchronization to a UTC-traceable time-source.
the presence of harmonic and out-of-band disturbances, even if For this analysis, we employed the Opal-RT eMEGAsim
this means larger response time and reporting latency. To this PowerGrid Real-Time Digital Simulator [9], whose block
end, the M-class model adopts a window length of 100 ms, scheme is presented in Fig. 1. In more detail, the RTS consists
i.e., equal to five nominal cycles at 50 Hz and corresponding of three main operational blocks: an industrial PC with 12
to a frequency resolution of 5 Hz. cores, a Dolphin DXE410 PCI Express Expansion Chassis
In order to mitigate the leakage effects due to spurious 3 The IEEE Std defines the out-of-band frequency range based on the
components, we need to identify their frequency and include nominal system frequency and the adopted reporting rate. For this analysis, we
it into the spectral support S. For this reason, the MSDFT set both these parameters equal to 50 Hz, as typical of M-class applications.

,(((0LODQ3RZHU7HFK

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SRINAGAR. Downloaded on December 16,2024 at 04:43:42 UTC from IEEE Xplore. Restrictions apply.
For this analysis, we set the sampling frequency equal to 5
kHz, and we consider an overall test duration of 2 s (with
the only exception of Frequency ramp test, whose duration
is set equal to 4 and 10 s for P- and M-class, respectively).
The PMU reporting rate is set equal to 50 fps. In order to
model realistic measurement conditions, the test waveforms
are intentionally corrupted by an uncorrelated Gaussian noise
with a Signal-to-Noise Ratio (SNR) of 80 dB.
As performance indexes, we consider five metrics:
Amplitude Error (AE), Phase Error (PE), Total Vector Error
Fig. 1. Block scheme of the eMEGAsim PowerGrid RTS adapted from [8].
(TVE), Frequency Error (FE) and ROCOF Error (RFE). In this
regard, Table II and III report the worst-case performance as
function of test condition in P- and M-class, respectively, and
TABLE I
PMU M ODELS ’ C OMPUTATIONAL C OMPLEXITY compare them with the corresponding IEEE Std limits.
As regards P-class, it is worth noticing how the PMU model
PMU Model PMUs per Core Computation Time [μs] complies with the IEEE Std requirements in all the static and
P-class 5 16 dynamic tests. In nominal steady-state conditions, the PMU
M-class (no OOB) 2 42 model provides a worst-case TVE of 0.003%, whereas FE and
M-class (with OOB) 1 56
RFE do not exceed 0.21 mHz and 0.03 Hz/s, respectively. As
expected, the worst performance is obtained in the Harmonic
distortion test. Indeed, the scarce frequency resolution (just
[27], and a Spartan-3 Field Programmable Gate Array (FPGA) 16.67 Hz) produces significant spectral leakage from harmonic
board [28]. The industrial PC runs the RTS model with an terms, that results in a performance degradation, e.g. worst-
integration time-step of 100 μs, the FPGA board locks the case FE might achieve 3.88 mHz. Nevertheless, all the
internal time-base to a more stable clock, and the DXE410 performance metrics comply with the IEEE Std requirements.
module enables the communication between the other two Similar considerations hold also for class-M compliance
blocks. In this regard, the alignment to UTC is provided by tests. In nominal steady-state conditions, the estimation
the Spectracom Tsync-PCIe Express board [29]. This hardware accuracy is comparable with P-class results, thus confirming
module employs a Global Positioning System (GPS) antenna that the main limiting factor is represented by the measurement
to provide the UTC timestamp for the PMU data-frames, and noise. The Out-of-band interference test proves that it is
the Pulse-per-Second (PPS) signal for the internal time-base. possible to mitigate distortion effects coming from inter-
b) Computational Complexity: The simulation of a large- harmonic components, if properly included into the spectral
scale PMU-based application depends on the adopted model support S. Once more, the Harmonic distortion test provides
complexity. In this sense, Table I reports the computational the worst performance, with a TVE in the order of 0.2%, yet
requirements of the P- and M-class models. For this analysis, still complies with the IEEE Std limits.
we consider two performance indexes: the number of PMUs Then, we study the models’ behavior in the presence of
that we can include in a single RTS processing core, and the transient events, like amplitude or phase step changes, when
computation time given an integration time of 100 μs. the signal energy is conveyed over the entire spectrum and
In the P-class configuration, the estimation process requires the synchrophasor model looses its significance. In this case,
only 16 μs, corresponding to a maximum number of five the IEEE Std requires to determine the measurement response
PMUs in the same core. The M-class configuration, instead, time, i.e. the time interval necessary for TVE and FE to return
involves a higher number of processing steps that results within the limits for steady-state test (TVE≤1%, FE≤5 mHz).
also in higher complexity. In this regard, we discriminate In this context, Fig. 2 and 3 represent the evolution of TVE
the contribution due to the OOB search and compensation. and FE in the step tests for P- and M-class, respectively,
If we consider only the harmonic terms, the computation time whereas Table IV reports the measured response times and
is limited to 42 μs and we can include two PMUs in the compares them with the IEEE Std limits. The compliance is
same core. If we consider also possible OOB components, the guaranteed in all the configurations, with the only exception of
computation time raises up to 56 μs and we can associate a the P-class response to Phase step test. In fact, the combination
single PMU to each core. of Hanning weighing function and spectral energy spreading
IV. P ERFORMANCE C HARACTERIZATION produces a fundamental main-lobe that can be hardly modeled
by the super-resolution dictionary Df [30]. As a consequence,
In this Section, we report the performance characterization
the support recovery fails and the corresponding TFM model is
of P- and M-class models with respect to the corresponding
not properly centered around the fundamental frequency, with
IEEE Std requirements4 and in a real-world operating scenario.
a consequent degradation of the final estimation accuracy. In
4 A thorough performance assessment of CS-TFM algorithm is available in this sense, a plausible solution might be the employment of a
[10] where also more realistic test conditions are taken into account. narrower weighing window (e.g. Kaiser window) or a shorter

,(((0LODQ3RZHU7HFK

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SRINAGAR. Downloaded on December 16,2024 at 04:43:42 UTC from IEEE Xplore. Restrictions apply.
TABLE II
W ORST-C ASE E STIMATION U NCERTAINTY FOR P-C LASS IEEE S TD C OMPLIANCE - T HREE -C YCLE W INDOW

Test AE [pu] PE [rad] TVE [%] Std [%] FE [mHz] Std [mHz] RFE [Hz/s] Std [Hz/s]
nominal 2.36·10−5 2.64·10−5 0.003 1 0.208 5 0.027 0.4
signal freq. (f0 = ±2 Hz) 5.13·10−5 5.62·10−5 0.007 1 3.378 5 0.089 0.4
harm. dist. (THD = 1%) 3.56·10−4 4.19·10−4 0.044 1 3.883 5 0.189 0.4
ampl. mod. (fm = 2 Hz) 4.99·10−5 4.32·10−5 0.006 3 0.183 60 0.073 2.3
phase mod. (fm = 2 Hz) 3.69·10−5 5.37·10−5 0.005 3 2.754 60 0.066 2.3
freq. ramp (Δf = ±2 Hz) 2.16·10−5 3.27·10−5 0.004 1 0.104 10 0.049 0.4

TABLE III
W ORST-C ASE E STIMATION U NCERTAINTY FOR M-C LASS IEEE S TD C OMPLIANCE - F IVE -C YCLE W INDOW

Test AE [pu] PE [rad] TVE [%] Std [%] FE [mHz] Std [mHz] RFE [Hz/s] Std [Hz/s]
nominal 2.67·10−5 5.85·10−6 0.003 1 0.177 5 0.025 0.1
signal (f0 = ±5 Hz) 3.84·10−5 3.83·10−5 0.004 1 0.261 5 0.037 0.1
harm. dist. (THD = 10%) 1.59·10−3 1.84·10−3 0.192 1 17.82 25 1.210 -
out-of-band (TIHD = 10%) 1.73·10−4 1.98·10−5 0.016 1.3 0.307 10 0.041 -
ampl. mod. (fm = 5 Hz) 4.12·10−4 7.66·10−5 0.050 3 2.693 300 0.789 14
phase mod. (fm = 5 Hz) 4.18·10−4 4.28·10−4 0.055 3 38.15 300 1.107 14
freq. ramp (Δf = ±5 Hz) 2.69·10−5 3.11·10−5 0.004 1 1.009 10 0.057 0.2

(a) (a)

0
100
TVE [%]
TVE [%]

Ampl. Step Ampl. Step


Phase Step Phase Step
-2 Std Limit
10-2 Std Limit

-4
10-4 0.9 0.92 0.94 0.96 0.98 1 1.02 1.04 1.06 1.08 1.1
0.9 0.95 1 1.05 1.1
time [s] time [s]
(b)
(b)
2 Ampl. Step
2 Ampl. Step
FE [mHz]

Phase Step
Phase Step
FE [mHz]

0 Std Limit
0 Std Limit
-2

-2
-4
0.9 0.92 0.94 0.96 0.98 1 1.02 1.04 1.06 1.08 1.1
0.9 0.95 1 1.05 1.1 time [s]
time [s]

Fig. 3. Evaluation of measurement response times for M-class configuration


Fig. 2. Evaluation of measurement response times for P-class configuration in the Amplitude step and Phase step tests in blue and red line, respectively.
in the Amplitude step and Phase step tests in blue and red line, respectively. The dashed black line represents the corresponding IEEE Std limit.
The dashed black line represents the corresponding IEEE Std limit.

an inter-area oscillation of the Continental Europe electricity


window length (e.g. 40 ms) [31], though it might result in a system recorded on December 1, 2016 (further details in [12]).
performance degradation in the other test conditions. As shown in Fig. 4(a) the fundamental frequency exhibits a
Finally, we evaluate the PMU models’ performance by using time-varying trend characterized by superposed modulations
a synthetic dataset inspired by a real-world network event. In and linear ramps. Even in such a challenging operating
particular, the test waveform derives from the recording of condition, both the PMU models prove to provide accurate
estimates, keeping FE within ±200 and ±75 μrad for P- and
M-class configuration, respectively.
TABLE IV
R ESPONSE T IME IN IEEE S TD S TEP C HANGES T ESTS V. C ONCLUSIONS
In this paper, we presented the design, implementation and
Model Step TVE Std FE Std characterization of two TF-based PMU models within a RTS.
Amplitude (±0.1 pu) 39.2 40 56.4 90 First, we discussed the theoretical background of the
P-class
Phase (±π/10 rad) 55 40 56.2 90
adopted synchrophasor extraction algorithm, i.e., CS-TFM. In
Amplitude (±0.1 pu) 69 140 92 280 this sense, we focused on the processing routines employed
M-class
Phase (±π/10 rad) 94 140 94.4 280
to compute the DFT bins and the TF expansion coefficients.

,(((0LODQ3RZHU7HFK

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SRINAGAR. Downloaded on December 16,2024 at 04:43:42 UTC from IEEE Xplore. Restrictions apply.
(a) [9] ”Opal-RT eMEGAsim PowerGrid Real-Time Digital Hardware in
frequency [Hz]

50.06 reference the Loop Simulator,” [online] http://www.opal-rt.com/product/emegasim-


powergrid-real-time-digital-hardware-in-the-loop-simulator
50.04
[10] M. Bertocco, G. Frigo, C. Narduzzi, C. Muscas and P. A. Pegoraro,
50.02 ”Compressive Sensing of a Taylor-Fourier Multifrequency Model for
Synchrophasor Estimation,” in IEEE Trans. on Instrum. and Meas., vol.
50
50 100 150 200 64, no. 12, pp. 3274-3283, Dec. 2015.
time [s] [11] D. Petri, D. Fontanelli and D. Macii, ”A Frequency-Domain Algorithm
(b) for Dynamic Synchrophasor and Frequency Estimation,” in IEEE Trans.
200 on Instrum. and Meas., vol. 63, no. 10, pp. 2330-2340, Oct. 2014.
P-class
100 [12] G. Frigo, A. Derviškadić, Y. Zuo and M. Paolone, ”PMU-Based
FE [ Hz]

M-class
0
ROCOF Measurements: Uncertainty Limits and Metrological Significance
in Power System Applications” in IEEE Trans. on Instrum. and Meas.,
-100 accepted for publication.
-200 [13] A.G. Phadke, J.S. Thorp, ”Synchronized Phasor Measurements and
50 100 150 200 Their Applications,” Springer, 2017.
time [s]
[14] J. A. de la O Serna, ”Dynamic Phasor Estimates for Power System
Oscillations,” in IEEE Trans. on Instrum. and Meas., vol. 56, no. 5, pp.
Fig. 4. In (a), the reference value for the fundamental frequency evolution 1648-1657, Oct. 2007.
during an inter-area oscillation event. In (b), the FEs obtained by P- and [15] M. A. Platas-Garza, J. A. de la O Serna, ”Dynamic Phasor and
M-class PMU model in red and blue line, respectively. Frequency Estimates Through Maximally Flat Differentiators,” in IEEE
Trans. on Instrum. and Meas., vol. 59, no. 7, pp. 1803-1811, July 2010.
[16] P. Castello, M. Lixia, C. Muscas and P. A. Pegoraro, ”Impact of the
Model on the Accuracy of Synchrophasor Measurement,” in IEEE Trans.
Then, we described the RTS platform and quantified the on Instrum. and Meas., vol. 61, no. 8, pp. 2179-2188, Aug. 2012.
models’ computational complexity. Finally, we characterized [17] C. Narduzzi, M. Bertocco, G. Frigo and G. Giorgi, ”Fast-
their estimation accuracy in all the IEEE Std tests, as well as TFM—Multifrequency Phasor Measurement for Distribution Networks,”
in IEEE Trans. on Instrum. and Meas., vol. 67, no. 8, pp. 1825-1835,
in an experimental dataset representative of real-world power Aug. 2018.
system dynamics. In this context, the PMU models prove to be [18] P. Castello, J. Liu, C. Muscas, P. A. Pegoraro, F. Ponci and A. Monti,
compliant with the requirements of both performance classes ”A Fast and Accurate PMU Algorithm for P+M Class Measurement of
Synchrophasor and Frequency,” in IEEE Trans. on Instrum. and Meas.,
(with the only exception of the response time for the P-class vol. 63, no. 12, pp. 2837-2845, Dec. 2014.
Phase step test). Based on these results, the developed models [19] M. Bertocco, G. Frigo, G. Giorgi and C. Narduzzi, ”Frequency Tracking
might represent a useful tool for investigating the performance for Efficient Phasor Measurements Based on a CSTFM Model,” 2015
IEEE AMPS, Aachen, 2015, pp. 1-6.
of dynamic phasor estimators also in real-world scenarios. [20] J. A. de la O Serna and J. Rodriguez-Maldonado, ”Instantaneous
Oscillating Phasor Estimates With TaylorK -Kalman Filters,” in IEEE
ACKNOWLEDGMENTS Trans. on Pow. Sys., vol. 26, no. 4, pp. 2336-2344, Nov. 2011.
[21] M. Bertocco, G. Frigo, C. Narduzzi and F. Tramarin, ”Resolution
This research project is carried out within the frame of Enhancement by Compressive Sensing in Power Quality and Phasor
the Swiss Centre for Competence in Energy Research on the Measurement,” in IEEE Trans. on Instrum. and Meas., vol. 63, no. 10,
pp. 2358-2367, Oct. 2014.
Future Swiss Electrical Infrastructure (SCCER-FURIES) with [22] E. Jacobsen and R. Lyons, ”The sliding DFT,” in IEEE Signal Processing
the financial support of the Swiss Innovation Agency. Magazine, vol. 20, no. 2, pp. 74-80, March 2003.
[23] IEEE Interharmonic Task Force, ”Interharmonics in Power Systems,”
R EFERENCES Cigre 36.05/CIRED 2 CC02 Voltage Quality Working Group [online]
http://grouper.ieee.org/groups/harmonic/iharm/ihfinal.pdf
[1] I. Kamwa, S. R. Samantaray and G. Joos, ”Compliance Analysis of PMU [24] K. Duda, ”Accurate, Guaranteed Stable, Sliding Discrete Fourier
Algorithms and Devices for Wide-Area Stabilizing Control of Large Power Transform [DSP Tips & Tricks],” Signal Processing Magazine, IEEE ,
Systems,” in IEEE Trans. on Pow. Sys., vol. 28, no. 2, pp. 1766-1778, vol.27, no.6, pp.124-127, Nov. 2010.
May 2013. [25] P. Romano and M. Paolone, ”An Enhanced Interpolated-Modulated
[2] C. Muscas, et al., ”Multiarea Distribution System State Estimation,” in Sliding DFT for High Reporting Rate PMUs,” 2014 IEEE International
IEEE Trans. on Instrum. and Meas., vol. 64, no. 5, pp. 1140-1148, May Workshop on Applied Measurements for Power Systems Proceedings
2015. (AMPS), Aachen, 2014, pp. 1-6.
[3] ”IEEE Standard for Synchrophasor Measurements for Power Systems,” [26] P. Romano and M. Paolone, ”Enhanced Interpolated-DFT for
in IEEE Std C37.118.1-2011, Dec. 2011. Synchrophasor Estimation in FPGAs: Theory, Implementation, and
[4] ”IEEE Standard for Synchrophasor Measurements for Power Systems – Validation of a PMU Prototype,” in IEEE Trans. on Instrum. and Meas.,
Amendment 1: Modification of Selected Performance Requirements,” in vol. 63, no. 12, pp. 2824-2836, Dec. 2014.
IEEE Std C37.118.1a-2014, April 2014. [27] ”DXE410 PCI Express Expansion Chassis,” [online] http://www.
[5] D. Ouellette, M. Desjardine, R. Kuffel, Yi Zhang, E. Xu, ”Using a Real dolphinics.com/download/DX/OPEN DOC/DXE410 Product Brief.pdf
Time Digital Simulator with Phasor Measurement Unit technology,” Int. [28] ”Spartan-3 FPGA Family Data Sheet,” [online] https://www.xilinx.com/
Conf. on APAP, 2011, vol.3, no., pp.2472,2476, 16-20 Oct. 2011. support/documentation/data sheets/ds099.pdf
[6] D. R. Gurushinghe, A. D. Rajapakse, D. Muthumuni, ”Modeling of a [29] ”Spectracom PCI Express Slot Cards,” [online] http://www.
Synchrophasor Measurement Units into an Electromagnetic Transient spectracomcorp.com/ProductsServices/TimingSynchronization/Bus-level
Simulation Program,” International Conference on Power Systems Timing/PCIexpressslotcards/tabid/1296/Default.aspx
Transients (ISPT), Vancouver, Canada July 18-20, 2013. [30] G. Frigo, D. Colangelo, A. Derviškadić, M. Pignati, C. Narduzzi and
[7] A. T. Al-Hammouri et al., ”Virtualization of Synchronized Phasor M. Paolone, ”Definition of Accurate Reference Synchrophasors for Static
Measurement Units within Real-Time Simulators for Smart Grid and Dynamic Characterization of PMUs,” in IEEE Trans. on Instrum.
Applications,” IEEE Power and Energy Society General Meeting (PES), and Meas., vol. 66, no. 9, pp. 2233-2246, Sept. 2017.
2012, pp.1,7, 22-26 July 2012. [31] P. Castello, C. Muscas, P. Attilio Pegoraro and S. Sulis, ”Critical
[8] P. Romano, M. Pignati, M. Paolone, ”Integration of an IEEE Std. C37.118 Analysis of PMU Testing Procedures for Step Response Evaluation,” 2015
Compliant PMU into a Real-Time Simulator,” 2015 IEEE Eindhoven IEEE AMPS, Aachen, 2015, pp. 37-42.
PowerTech, Eindhoven, 2015, pp. 1-6.

,(((0LODQ3RZHU7HFK

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SRINAGAR. Downloaded on December 16,2024 at 04:43:42 UTC from IEEE Xplore. Restrictions apply.

You might also like