Riccobono 2017 Zgrid Measurement

This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2017.2783042, IEEE Journal
of Emerging and Selected Topics in Power Electronics
Noninvasive Online Parametric Identification of

Three-Phase AC Power Impedances to Assess
the Stability of Grid-Tied Power Electronic
Inverters in LV Networks
Antonino Riccobono, Member, IEEE, Markus Mirz, Student Member, IEEE and Antonello Monti,
Senior Member, IEEE
Institute for Automation of Complex Power Systems
E.ON Energy Research Center, RWTH Aachen University
Aachen 52074, Germany
[ariccobono, mmirz, amonti]@ eonerc.rwth-aachen.de
Abstract—This paper presents a noninvasive online integration of Renewable Energy Sources (RESs) and loads
parametric identification of three-phase AC power through grid-connected feedback-controlled power electronic
impedances to assess small-signal stability of grid-tied converters [1]-[3]. This transformation is schematically
inverter systems by using well-known impedance-ratio- depicted in Fig. 1, representing an exemplary low-voltage (LV)
based stability criteria. The identification technique is radial feeder. In Fig. 1 (a), the traditional AC LV feeder is
integrated into the control of an existing grid-tied inverter mainly a passive network because the majority of loads are
for the estimation of wide bandwidth AC grid impedances, either AC machines, resistive, or diode rectifiers, fed by the
on top of its original power conversion function. This is MV/LV transformer. In Fig. 1 (b), the counterpart Power
accomplished in practice by injecting a short-time small- Electronics AC LV feeder is an active network where inverters
signal Pseudo Random Binary Sequence (PRBS), a digital interface the distributed RESs and active rectifiers interface all
approximation of white noise which is wide bandwidth in the loads. In the situation of Fig. 1 (b), it is also envisioned that
nature, on the inverter control loop so that all frequencies the MV/LV conversion will be taken by solid-state transformer
of interest at the impedance measurement point can be technology [4].
excited at once. Then, digital processing is performed in the As power electronics penetrates the AC grid, a new challenge
integrated control platform where the parametric AC grid is raising for both power electronic engineers and power
impedance is extracted from the measurement of voltage systems engineers. This proliferation of grid-connected power
and current over the length of PRBS injection. Moreover, a electronics can have a destabilizing effect on the AC voltage
procedure on how to identify the output impedance of the due to interactions among grid-connected feedback-controlled
inverter is deployed so that the parametric source and load power electronic converters and equivalent power grid
impedances can be used to verify the system stability by impedances seen at the various Point of Common Couplings
means of the generalized Nyquist stability criterion. The (PCCs) [5]-[11]. In the literature, this stability problem has been
technique is validated via Hardware In the Loop (HIL) real- explained in two ways. The first explanation considers the
time simulation. The present work focuses on the destabilizing equivalent negative incremental resistance
identification of balanced three-phase AC impedances in dq behaviors of output and input impedance of grid-connected
reference frame and a dq diagonal-dominant stability feedback-controlled inverters and active rectifiers, respectively
analysis which is typical of LV Distribution Grids. [8]-[10]. The second explanation considers the interaction of
the grid-connected converters and their control loops. When
Index Terms—Impedance Measurement, Inverters, Power plotting the impedances of each subsystems connected at the
System Stability, Stability Analysis, Stability Criteria various PCCs, the interaction manifests itself as an impedance
magnitude overlap over a certain frequency range. This
I. INTRODUCTION impedance overlap may bring the system into instability [5]-
[10].
The traditional AC Power System is transforming into a Power-
Initially proposed for DC power distribution systems, such as
Electronics-based AC Power System due to the deeper
the Middlebrook criterion [12] and its extensions [13],
This work extends previously presented work at COMPEL 2016.
2168-6777 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JESTPE.2017.2783042, IEEE Journal
impedance-ratio-based stability criteria have been proposed and measurement point. Then, advanced digital signal processing
widely used to study the small-signal stability of Power- techniques can be applied on the measured voltage and current
Electronics-based AC Power Distribution Systems [5]-[11]. All to identify the impedance of the system under test. The online
these stability criteria are based on the minor loop gain concept, wideband impedance identification methods can be categorized
i.e. an impedance ratio at the source/load interface at the PCC. in steady-state methods and transient methods. In the steady-
Considering that the minor loop gain is a type of loop gain of state methods, a wideband small-signal perturbation signal is
an equivalent negative feedback loop system, the stability study superimposed onto the steady-state operating point for a certain
can be performed by applying the Nyquist stability criterion to amount of time and the impedance is extracted from the
such a minor loop gain. AC impedances of balanced three- response during the perturbation [16]-[22],[24]. In the transient
phase systems are usually represented in dq reference frame and methods, usually a current impulse is injected at impedance
are 2 × 2 matrixes1. Therefore, the impedance ratio, i.e. the measuring point and the impedance is extracted from its
minor loop gain, is also a 2 × 2 matrix. To study the stability of transient response [25],[29]-[30]. Despites the relatively longer
Power-Electronics-based AC Power Distribution Systems in dq perturbation, steady-state methods are recently of greater
domain, the generalized Nyquist stability criterion is used [8]- interest because of the too aggressive nature of the transient
[11]. methods. The online wideband impedance identification
0
Z0 methods can be further categorized in invasive methods [23]-
Z01 1 Z12 2 Z23 3
[31], which require a dedicated power hardware to perform the
impedance identification task, and noninvasive methods [14]-
MV/LV Zg1 Zg2 Zg3 [22], which use the existing power hardware in the system.
ZL1 ZL2 ZL3 Among the noninvasive solutions, methods for wideband
identification of power impedances in conjunction with existing
(a) power electronics converters are quite attractive, especially in
0 Z0 Power-Electronics-based AC Power Distribution Systems [16]-
AC Z01 Z12 [22]. In fact, since many of the converters connected to the AC
1 2
grid have digital control and sensing, they may also be used to
AC Zg1 Zg2 monitor AC power grid impedances on the top of their power
MV/LV
ZS1 ZL1 ZS2 ZL2 conversion function. Digital network analyzer techniques [14]-
AC AC AC AC [15] can be integrated into the converter controllers allowing
DC DC DC DC them to be used as online monitors without any extra power
VS1 ZLL1 VSS2 ZLL2
hardware.
(b) This paper presents an online wideband impedance
identification method, called the Wideband System
Fig. 1. Exemplary traditional passive LV radial feeder (a), and counterpart Identification (WSI) technique, which allows the integration of
power electronics based feeder (b).
the control platform, where the WSI technique is implemented,
into the controller of an existing grid-connected power
In order to apply the impedance-ratio-based stability criteria,
electronic converter. More in the detail, the identification
it is necessary to know the interface impedances. Offline
technique consists of injecting a short-time small-signal
analytical methods to study the stability of Power-Electronics-
approximation of white noise (wide bandwidth in nature), i.e.
based AC Power Distribution Systems are not enough because
Pseudo Random Binary Sequence (PRBS), on the control loop
practically the system impedances continuously vary over time
of the existing grid-connected power electronic converter while
and in relation to too many parameters. Therefore, online
it is performing its power conversion function. Then, digital
methods to measure impedances are required in order to
processing is performed in the integrated control platform to
monitor system stability in real time and take corrective actions
identify the grid impedance, i.e. the load subsystem impedance,
if needed. The ideal impedance measurement technique would
in dq reference frame at the PCC from the voltage and current
complete the measurement in a short time, allowing fast
measurements over the length of the injection. The online WSI
response to system variations. Even though narrowband
technique belongs to the category of the steady-state
identification methods based on injection of sine sweeps give
identification methods because PRBS is injected during the
the best impedance measurement accuracy, they are not suitable
steady-state operation of the system. It also belongs to the
for online applications due to the very long measurement times.
category of noninvasive methods because it does not require
In contrast, wideband identification methods based on injection
any extra power hardware to inject the perturbation and
of a high-frequency content signal able to excite all frequencies
measure voltages and currents needed to identify the impedance
of interest simultaneously meet the requirement of fast
of the system under test.
measurement and have been proposed in the literature to
With respect to past publications in the field of wideband
measure impedances [14]-[31]. These methods were first
measurement of AC grid impedances via existing grid-
proposed for DC systems [14]-[15], and then extended to AC
connected power electronic inverters [16]-[18], the new
systems [16]-[31]. These methods consist in generating a
contribution of the present work resides in the implementation
wideband perturbation controlled in both magnitude and
of the online parametrization of the identified non-parametric
duration for both voltage and current at the impedance
1
AC impedances can be modeled with other methods, such as harmonic dynamic phasors, and reduced-order modeling [46],[47]. These methods are not
linearization or modeling in the phase sequence domain, modeling with object of this paper, but considerations for their usage are made in Section VII.
impedances in the same control platform. This allows the quantities in the abc phase domain, the AC system becomes a
identification of parametric grid impedances in real time. The multiple-input and multiple-output (MIMO) DC system in the
real-time parametrization of identified impedances is the key dq domain. To derive the small-signal impedances in the dq
enabler for several system-level applications, such as online domain, linearization techniques are used [37]. In dq domain,
stability monitoring, harmonic propagation detection, active the obtained source and load AC impedances are the 2 × 2
filter tuning, and adaptive control. Moreover, with respect to matrixes (1) and (2), which clearly also have cross terms
pure nonparametric identification, the authors envision that between the d-component and q-component.
real-time parametrization of grid impedances will enable fast 𝒁𝑺_𝒅𝒒 (𝑠) = [
𝑍𝑆_𝑑𝑑 (𝑠) 𝑍𝑆_𝑑𝑞 (𝑠)
] (1)
and reliable communication links for the system-level 𝑍𝑆_𝑞𝑑 (𝑠) 𝑍𝑆_𝑞𝑞 (𝑠)
𝑍𝐿_𝑑𝑑 (𝑠) 𝑍𝐿_𝑑𝑞 (𝑠)
applications listed above at all automation levels of future 𝒁𝑳_𝒅𝒒 (𝑠) = [ ] (2)
𝑍𝐿_𝑞𝑑 (𝑠) 𝑍𝐿_𝑞𝑞 (𝑠)
Power-Electronics-based AC Power System such as that
The source-load system and its equivalent control block
depicted in Fig. 1 (b). With respect to previously presented
diagram are shown in Fig. 2.
work by the authors [19], where the identification focused on
the phase domain, this paper presents the online parametrization +
ZS_dq
of the identified grid impedance in dq domain. Real-time + ZS_dq
parametric identification was recently presented by the authors VS_dq -
ZL_dq
VL_dq
[20], but with the different goal of improving stability and
accuracy of a Power Hardware in the Loop setup and only - Z-1L_dq
considering passive DC impedances. This paper extends the (a) (b)
identification to three-phase AC impedances. References [21]-
[22] also used PRBS as perturbation signal and the impedance Fig. 2. Small-signal representation of a three-phase system in the dq domain
(a), and equivalent MIMO feedback block diagram (b).
of the system under test was also identified in dq domain.
However, they did not embody any parametric impedance
The source and load voltage vectors are defined as follows.
identification in real time. Moreover, again with respect to [19], 𝑣𝑆_𝑑
the procedure described in [21] to identify the parametric output 𝑉𝑆_𝑑𝑞 (𝑠) = [𝑣 ]
𝑆_𝑞
(3)
impedance of the inverter is used, but parametric identification 𝑣𝐿_𝑑
is embodied. Both the parametric inverter output impedance 𝑉𝐿_𝑑𝑞 (𝑠) = [𝑣 ]
𝐿_𝑞
(4)
and grid impedance serve for small-signal stability analysis of By inspection of Fig. 2, the load voltage can be calculated as
grid-tied inverter systems by applying the generalized Nyquist follows.
stability criterion to the minor loop gain defined as the 𝑉𝐿_𝑞𝑑 (𝑠) = (𝑰 + 𝒁𝑺_𝒅𝒒 (𝑠) ∙ 𝒁−𝟏
−𝟏
(5)
𝑳_𝒅𝒒 (𝑠)) ∙ 𝑉𝑆_𝑑𝑞 (𝑠)
impedance ratio at the inverter/grid interface. The technique is
verified in a Hardware In the Loop (HIL) real-time simulation As a consequence, the so-called minor loop gain is given by
𝑻𝑴𝑳𝑮_𝒅𝒒 (𝑠) = 𝒁𝑺_𝒅𝒒 (𝑠) ∙ 𝒁−𝟏
𝑳_𝒅𝒒 (𝑠) =
setup. The HIL setup features an exemplary plant consisting of (6)
𝒁𝑺_𝒅𝒒 (𝑠) ∙ 𝒀𝑳_𝒅𝒒 (𝑠)
a grid-tied digitally-controlled inverter simulated in a real-time −𝟏
simulator and the digital platform where the identification where 𝒀𝑳_𝒅𝒒 (𝑠) = is the load admittance matrix.
𝒁𝑳_𝒅𝒒 (𝑠)
algorithms are implemented. Using (1) and (2) in (6), the expression of the minor loop gain
The paper is organized as follows. Section II gives the can be expanded as follows.
theoretical background for stability analysis of grid-tied 𝑻𝑴𝑳𝑮_𝒅𝒒 (𝑠) = 𝒁𝑺_𝒅𝒒 (𝑠) ∙ 𝒀𝑳_𝒅𝒒 (𝑠) =
𝑍𝑆_𝑑𝑑 (𝑠) 𝑍𝑆_𝑑𝑞 (𝑠) 𝑌𝐿_𝑑𝑑 (𝑠) 𝑌𝐿_𝑑𝑞 (𝑠)
inverter systems. Section III derives analytic expressions for [ ]∙[ ]=
𝑍𝑆_𝑞𝑑 (𝑠) 𝑍𝑆_𝑞𝑞 (𝑠) 𝑌𝐿_𝑞𝑑 (𝑠) 𝑌𝐿_𝑞𝑞 (𝑠) (7)
both the inverter output impedance and grid impedance in dq 𝑍𝑆𝑑𝑑 (𝑠)𝑌𝐿𝑑𝑑 (𝑠) + 𝑍𝑆_𝑑𝑞 (𝑠)𝑌𝐿_𝑞𝑑 (𝑠) 𝑍𝑆𝑑𝑑 (𝑠)𝑌𝐿𝑑𝑞 (𝑠) + 𝑍𝑆_𝑑𝑞 (𝑠)𝑌𝐿_𝑞𝑞 (𝑠)
domain that will be the references which the results of the [
𝑍𝑆𝑞𝑑 (𝑠)𝑌𝐿𝑑𝑑 (𝑠) + 𝑍𝑆_𝑞𝑞 (𝑠)𝑌𝐿_𝑞𝑑 (𝑠) 𝑍𝑆𝑞𝑑 (𝑠)𝑌𝐿𝑑𝑞 (𝑠) + 𝑍𝑆_𝑞𝑞 (𝑠)𝑌𝐿_𝑞𝑞 (𝑠)
]
online WSI are compared to. Section IV presents the WSI Assuming stand-alone stability of the source and load
technique and explains the methods for identification of both subsystems, the interconnected three-phase system is stable if
the grid impedance and the inverter output impedance in dq and only if the minor loop gain (7) satisfies the generalized
domain. Section V presents the HIL setup providing detailed Nyquist stability criterion. The reader can refer to [8]-[11] for
description of all the routines and algorithms needed to online its detailed formulation. In practice, the interconnected three-
identify AC power impedances. Section VI gives the results of phase system is stable if and only if the Nyquist contours of the
the online parametric impedance identification and stability minor loop gain (7) do not encircle the (−1, 𝑗0) point. Based on
analysis. Conclusions and future work are finally given in the concept of avoiding encircling the (−1, 𝑗0) point in the
Section VII complex plane, many impedance-ratio-based stability criteria
were proposed for interconnected DC Power Systems and they
II. SMALL-SIGNAL STABILITY ANALYSIS OF GRID-TIED are reviewed in [13]. They provide only sufficient, but not
INVERTER SYSTEMS necessary stability conditions by defining various forbidden
As an extension of the case for DC systems [12]-[13], regions for the polar plot of the minor loop gain, defined at the
impedance-ratio-based stability analysis for balanced three- power interface between source and load subsystems. The
phase systems is performed in this paper in the dq rotating concept of forbidden regions perfectly applies to three-phase
frame to circumvent the modeling problem given by the time- AC systems to derive design formulations for a stable
variant nature of AC systems [8]-[11]. By applying a interconnected system as demonstrated in [10].
synchronous reference frame transformation to the measured To perform small-signal stability analysis of grid-connected
inverter systems, the classical approach consists in studying the In practice, it is convenient to apply the generalized Nyquist
linearized system under steady state conditions by assuming stability criterion to 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟏 (𝑠) or the more conservative
this convention: the grid-tied inverter is the source subsystem criteria to avoid encirclement of the (−1, 𝑗0) point [10].
(subscript “S”) and the grid is the load subsystem (subscript Differently from DC systems which are single-input and
“G”), both connected at the PCC interface. Fig. 3 shows the single-output (SISO) systems, studying the stability of three-
equivalent system broken down into the two subsystems phase AC systems by using the generalized Nyquist stability
assumed to be individually stable. The inverter normally criterion applied to the minor loop gains 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟏 (𝑠) or
operates in current mode (Fig. 3 (a)); however, it can also 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠) can be very difficult and unpractical due to its
operate in voltage mode (Fig. 3 (b)) to provide voltage support MIMO nature. However, in LV Distribution Grids,
in case of an islanded microgrid formation or the main grid simplifications can be obtained to at least obtain a diagonal-
becomes weak [36]. dominant matrix. In fact, if the LV power-electronics-based AC
IPCC IPCC
power system of Fig. 1 (b) is considered, then the cross terms
ZG_dq ZS_dq ZG_dq of 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟏 (𝑠) and 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠), defined at the various PCCs
+ +
Is_dq VG_dq VS_dq VG_dq of the same system, can be neglected for the following two
ZS_dq
VPCC
VPCC reasons:
- -
 LV Distribution Grids are mainly resistive, i.e.
(a) (b) 𝑋/𝑅 < 1 (negligible capacitive and inductive
Fig. 3. Small-signal representation of the grid-tied inverter system in the dq effects).
domain. Inverter acting as a current source (a) and as voltage source (b).  Grid-connected inverters embody decoupling
controls which significantly reduce the magnitude
Using the approach in [11], for the inverter operating in of the cross terms of their output impedance
current mode (Fig. 3 (a)), the voltage at the PCC is compared with the terms of the main diagonal [8].
 Grid-connected active rectifiers operate as high
−1
𝑉𝑃𝐶𝐶_𝑑𝑞 (𝑠) = (𝒁−𝟏 −𝟏
𝑺_𝒅𝒒 (𝑠) + 𝒁𝑮_𝒅𝒒 (𝑠)) 𝐼𝑆_𝑑𝑞 (𝑠)
+ 𝒁𝑺_𝒅𝒒 (𝑠) power factor loads and, as results, the magnitude of
∙ (𝒁𝑺_𝒅𝒒 (𝑠) + 𝒁𝑮_𝒅𝒒 (𝑠))
−1
𝑉𝐺_𝑑𝑞 (𝑠) =
the cross terms of their input admittance is
−1 (8) significantly reduced with respect to that of the
(𝑰 + 𝒁𝑺_𝒅𝒒 (𝑠)𝒁−𝟏
𝑮_𝒅𝒒 (𝑠)) ∙ 𝒁𝑺_𝒅𝒒 (𝑠) ∙ 𝐼𝑆_𝑑𝑞 (𝑠) terms of the main diagonal [10],[41].
−1
+ (𝑰 + 𝒁𝑺_𝒅𝒒 (𝑠)𝒁−𝟏
𝑮_𝒅𝒒 (𝑠)) ∙ 𝑉𝐺_𝑑𝑞 (𝑠) Under this assumptions, the minor loop gains 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠)
or 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟏 (𝑠) used for stability analysis of grid-connected
where two minor loop gains are identifiable as follows current-controlled and voltage-controlled inverter systems,
𝑻𝑴𝑳𝑮_𝒅𝒒_𝟏 (𝑠) = 𝒁𝑺_𝒅𝒒 (𝑠)𝒁−𝟏
𝑮_𝒅𝒒 (𝑠) (9) respectively, can be simplified as (12) and (13).
𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠) = 𝒁𝑮_𝒅𝒒 (𝒔) ∙ 𝒁−𝟏 𝑺_𝒅𝒒 (𝑠) =
𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠) = 𝒁𝑮_𝒅𝒒 (𝑠)𝒁−𝟏
𝑺_𝒅𝒒 (𝑠) (10) 𝒁𝑮_𝒅𝒒 (𝒔) ∙ 𝒀𝑺_𝒅𝒒 (𝑠) ≈
𝑍𝐺_𝑑𝑑 (𝑠) 0 𝑌𝑆_𝑑𝑑 (𝑠) 0
Being 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟏 (𝑠) the inverse matrix of 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠), for the [
0 𝑍𝐺_𝑞𝑞 (𝑠)
]∙[
0 𝑌𝑆_𝑞𝑞 (𝑠)
]= (12)
case of inverter operating in current mode it is demonstrated in 𝑍𝐺𝑑𝑑 (𝑠)𝑌𝑆𝑑𝑑 (𝑠) 0
[ ]
[11] that applying the generalized Nyquist stability criterion to 0 𝑍𝐺_𝑞𝑞 (𝑠)𝑌𝑆_𝑞𝑞 (𝑠)
𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠) is equivalent to apply the generalized inverse
Nyquist stability criterion to 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠) and to apply the 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟏 (𝑠) = 𝒁𝑺_𝒅𝒒 (𝑠) ∙ 𝒁−𝟏𝑮_𝒅𝒒 (𝑠) =
𝒁𝑺_𝒅𝒒 (𝑠) ∙ 𝒀𝑮_𝒅𝒒 (𝑠) ≈
generalized Nyquist criterion to 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟏 (𝑠). In practice, it is 𝑍𝑆_𝑑𝑑 (𝑠) 0 𝑌𝐺_𝑑𝑑 (𝑠) 0
sufficient to apply the generalized Nyquist stability criterion to [
0 𝑍𝑆_𝑞𝑞 (𝑠)
]∙[
0 𝑌𝐺_𝑞𝑞 (𝑠)
]= (13)
𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠) or the more conservative criteria to avoid 𝑍𝑆 (𝑠)𝑌𝐺𝑑𝑑 (𝑠) 0
[ 𝑑𝑑 ]
encirclement of the (−1, 𝑗0) point [8]. 0 𝑍𝑆_𝑞𝑞 (𝑠)𝑌𝐺_𝑞𝑞 (𝑠)
Again using the approach in [11], for the inverter operating Therefore, according to the simplifications in (12) and (13),
in voltage mode (Fig. 3 (b)), the voltage at the PCC is small-signal stability analysis of the grid-tied inverter system
𝑉𝑃𝐶𝐶_𝑑𝑞 (𝑠) = 𝒁𝑮_𝒅𝒒 (𝑠) ∙ (𝒁𝑺_𝒅𝒒 (𝑠) + 𝒁𝑮_𝒅𝒒 (𝑠))
−1
𝑉𝑆_𝑑𝑞 (𝑠) requires the parametric identification of only four interface
+ 𝒁𝑺_𝒅𝒒 (𝑠) impedances, i.e. 𝑍𝑆_𝑑𝑑 (𝑠), 𝑍𝑆_𝑞𝑞 (𝑠), 𝑍𝐺_𝑑𝑑 (𝑠), and 𝑍𝐺_𝑞𝑞 (𝑠).
−1 These impedances are online identified through the WSI tool
∙ (𝒁𝑺_𝒅𝒒 (𝑠) + 𝒁𝑮_𝒅𝒒 (𝑠)) 𝑉𝐺_𝑑𝑞 (𝑠) =
described in the Section V.
𝒁𝑺_𝒅𝒒 (𝑠)𝒁−𝟏
−1 (11)
(𝑰 + 𝑮_𝒅𝒒 (𝑠)) ∙ 𝑉𝑆_𝑑𝑞 (𝑠)
−1
+ (𝑰 + 𝒁𝑮_𝒅𝒒 (𝑠)𝒁−𝟏
𝑺_𝒅𝒒 (𝑠))
III. IMPEDANCE MODELING OF GRID-TIED INVERTERS AND
∙ 𝑉𝐺_𝑑𝑞 (𝑠) GRID IMPEDANCES
where the same two minor loop gains as defined in (9) and (10) The system under consideration is shown in Fig. 4 and
are identifiable. For the case of inverter operating in voltage consists of a feedback-controlled inverter connected to a
mode, the same relationship between generalized Nyquist lumped grid model. In this section the analytic expressions for
stability criterion and generalized inverse Nyquist stability both grid-tied inverter output impedance ZS_dq and grid
criterion applied to either 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟏 (𝑠) or 𝑻𝑴𝑳𝑮_𝒅𝒒_𝟐 (𝑠) exist. impedance ZG_dq are given in dq reference frame and they will
be compared to the results of the identification in Section VI.
RL ZS_dq
= Sensing Ea
+
Vs +
Eb
=
1500 V
L +
Ec
=
2 mH RG = 2.5
LG = 1 mH EG = 230 Vrms
CG = 3 µF f = 50 Hz
C ZG_dq
fsw = 20 kHz =
to PWM ia ib ic va vb vc 60 µF
θ abc θ abc Ideal θ

abc θ dq dq PLL
dq
Current Control:
dd dq id iq vd vq Kp_I = 0.016
Ki_I = 36.092
Id_ref = 34.109 A
Decoupling
Iq_ref = 6.4294 A
dq dd Id_ref - Voltage Control:

-
+ + Vd_ref Kp_V = 0.03
GC_I GC_V
Ki_V = 68.046
- Vd_ref = V
- +
+V Vq_ref = 0 V
GC_I GC_V q_ref
Current Iq_ref Voltage

Controller Controller
Fig. 4. Circuital diagram of the feedback-controlled inverter connected to a lumped grid model.
The circuital parameters of the three-phase inverter are Park Transformation. The first step of this procedure yields the
reported in the same Fig. 4 along with its control parameters. open-loop large-signal averaged equations:
The inverter, whose switching frequency is 20 kHz, is 𝐿
𝑑𝑖𝑑
= 𝑑𝑑 𝑣𝑆 − 𝑣𝑑 + 𝜔𝐿𝑖𝑞 − 𝑅𝐿 𝑖𝑑 (14)
connected to a simplified balanced three-phase representation 𝑑𝑡
𝑑𝑖𝑞
of the AC system consisting of an ideal three-phase voltage 𝐿 = 𝑑𝑞 𝑣𝑆 − 𝑣𝑞 − 𝜔𝐿𝑖𝑑 − 𝑅𝐿 𝑖𝑑 (15)
𝑑𝑡
source Eabc and an equivalent linear Thévenin impedance ZG 𝑑𝑣𝑑
modeled as Y-connected configuration as depicted in Fig. 4, 𝐶 = 𝑖𝑑 + 𝜔𝐶𝑣𝑞 − 𝑖𝑙𝑜𝑎𝑑_𝑑 (16)
𝑑𝑡
whose circuital parameter values are also given. The controller 𝑑𝑣𝑞
𝐶 = 𝑖𝑞 − 𝜔𝐶𝑣𝑑 − 𝑖𝑙𝑜𝑎𝑑_𝑞 (17)
consists of an inner current loop and an optional outer voltage 𝑑𝑡
loop and it is designed on a synchronous dq reference frame.
The inverter normally operates in current mode to inject a where vd, vq, id, iq are the output voltage and inductor current
certain amount of active (and reactive) power into the grid; state variables, respectively, and dd, dq are the control input
however, it can also operate in voltage mode to provide voltage variables. The currents iload_d and iload_q are the load currents.
support in case of a microgrid formation [36]. The values of the The averaged model given by (14)-(17) contains cross-
abc duty cycles vary sinusoidally in synchronism with the AC coupling terms between the d and q channels. In this paper, the
grid voltage thanks to the use of an ideal Phase Locked Loop decoupling technique of [40] is used to minimize the effect of
(PLL) (see again Fig. 4). the cross-coupling terms in the model. The small-signal open-
loop model is derived by perturbation-and-linearization
A. Inverter Output Impedance technique [37]. Then, the current and voltage controls are
designed based on this small-signal model with the
In this paper, “Ideal PLL” means that its dynamics are performance given in Table I.
neglected. In other words, an oscillator perfectly synchronized
to the AC grid voltage is used. Inspired by [8], the authors are TABLE I. INVERTER CONTROL PERFORMANCE
currently working on the derivation of a modeling framework Current Control Voltage Control
in dq domain of grid-connected power electronic converters Crossover Frequency (fc) 2 kHz 100 Hz
with real PLLs. By neglecting the dynamics of the PLL, the Phase Margin (PM) 80° 80°
small-signal modeling is derived from the equivalent averaged
model in dq rotating coordinates that is obtained from the The inverter output impedance ZS_dq in dq reference frame is
averaged abc-model of the grid-connected inverter by using the derived under the influence of the current control and under the
influence of the voltage control. In this paper, a simulation Moreover, the dq and qd components have a significant smaller
modeling exercise is presented to derive such small-signal magnitude compared to that of the dd and qq components. This
closed-loop models. This exercise not only will produce the shows that ZS_dq can be considered to be diagonal dominant.
analytic transfer functions which the results of the WSI
technique will be compared to in Section VI, but also it will Inverter
Averaged Model
serve as validation tool for the ongoing modeling framework in Input
Eqs. (14)-(17)
Output
Pertubation Measument
dq domain of grid-connected power electronic converters with iload_d vd
real PLLs. First, the averaged model of the inverter (14)-(15) iload_q vq
with the feedback control was implemented in Simulink as
dd id
depicted in Fig. 5. By performing a linear analysis in the dq iq
“Control Design” toolbox in Simulink, it is possible to linearize
the feedback-controlled averaged model of the inverter around
id iq vd vq
an operating steady-state point by specifying linearization
inputs and outputs and eventually extract transfer functions Decoupling
[43]. To extract the inverter output impedance, the linearization
dq dd Id_ref
inputs are iload_d and iload_q, while the linearization outputs are vd - + - +
GC_I GC_V Vd_ref
and vq. The extracted analytic expressions of the inverter output
impedances are reported in Appendix. These transfer functions - + - +V
GC_I GC_V q_ref
are plotted in Fig. 6. Notice that when the dynamics of the PLL
are neglected, the output characteristics of the inverter resemble Current Iq_ref Voltage
like those of an uninterruptible power supply [42]. In fact, the Controller Controller
components Zdd and Zqq are the same; the components Zdq and Fig. 5. Averaged model of the feedback-controlled inverter with signals
Zqd have also the same magnitude, but a phase shift of 180°. highlighted in red to extract the inverter output impedance.
(c)
(a)
(b) (d)
Fig. 6. Analytic output impedance of the inverter under current control (a)-(b), and voltage control (c)-(d).
to the grid impedance labeled as Case 1, the system results in a

B. Grid Impedance
more robustly stable fashion with respect to those of Case 2, as
In this paper, the grid impedance is modeled as a balanced it will be shown in Section VI. These transfer functions are
three-phase RLC impedance as depicted at the right hand side plotted in Fig. 7. Likewise, for the inverter output impedance,
of Fig. 4. An RL branch is in parallel to a capacitive branch. the components Zdd and Zqq are the same; the components Zdq
The equivalent grid impedance model in dq reference frame can and Zqd have also the same magnitude, but a phase shift of 180°.
be calculated as follows. The RL branch impedance in dq Moreover, the grid impedance can be surely considered as
coordinates is given by diagonal dominant for Case 1 because the magnitude of the dd
𝑠𝐿 + 𝑅𝐺 −𝜔𝐺 𝐿𝐺
𝑍𝑅𝐿_𝑑𝑞 = [ 𝐺 ] (18) (and qq) component is sufficiently larger than the magnitude of
𝜔𝐺 𝐿𝐺 𝑠𝐿𝐺 + 𝑅𝐺
the dq (and qd) component for all frequencies. The grid
where 𝜔𝐺 is the AC system fundamental frequency, i.e. 2 ∙
impedance for Case 2 can be still considered diagonal dominant
𝜋 ∙ 50 𝑟𝑎𝑑/𝑠. The capacitive branch admittance in dq
because the diagonal elements are sufficiently larger than the
coordinates is given by
𝑠𝐶𝐺 −𝜔𝐺 𝐶𝐺
cross terms for all the frequencies except around and below the
−1
𝑌𝐶_𝑑𝑞 = 𝑍𝐶_𝑑𝑞 =[
𝜔𝐺 𝐶𝐺 𝑠𝐶𝐺
] (19) fundamental frequency.
Finally, the RLC grid impedance in dq coordinates is
TABLE II. GRID IMPEDANCE PARAMETERS FOR CASE 1 AND CASE
calculated as follows. 2
−1
𝑍𝐺_𝑑𝑞 = (𝑍𝑅𝐿 + 𝑍𝐶−1 )−1 (20) Case 1 Case 2
RG 2.5 Ω 1.2 Ω
Analytic expressions of the grid impedance can be calculated LG 1 mH 5 mH
from (18) to (20) and can be found in Appendix. Two cases are CG 3 µF 5 µF
under consideration, namely Case 1 and Case 2 whose circuital
parameters are given in Table II. When the inverter is connected
(a) (c)
(b) (d)
Fig. 7. Analytic grid impedance for Case 1 (a)-(b), and Case 2 (c)-(d).
control to inductor current transfer function, Zout is the output

IV. THE WIDEBAND SYSTEM IDENTIFICATION TECHNIQUE impedance, and TI is the current control loop gain defined in
In this section, the principle of operation of the WSI Fig. 9.
Three-Phase Inverter Three-Phase System
technique is presented. First, the technique on how to identify PCC under test
the grid impedance is given. Then, that one of the inverter

LC Filter
output impedance is explained. Both the grid impedance and VS
inverter output impedance are identified in dq domain. The

main feature of the WSI technique is its natural characteristic to ZG_dq
be integrated into the control of the inverter so that it can be PWM Vabc, Iabc
measurements
used as an online monitor for grid stability applications as it will
be shown in Section VI. The choice of the identification in dq Controller
Vdq, Idq
domain comes as requirement from the inverter control which WSI tool
measurements
is implemented in such a domain as shown in Section III. This

PRBS
choice reduces the propagation of errors and limits the delays generation Processing
ZG_dq(s)
given by extra transformations if other modeling domains are
selected. (a)
A. Grid Impedance Identification in dq domain to PWM ia ib ic va vb vc
This subsection presents the online WSI technique exploiting θ

θ abc θ abc PLL
an existing grid-tied feedback-controlled inverter to identify the abc θ dq dq
dq
grid impedance in the dq domain. The WSI technique is based dd dq id iq vd vq
on the perturb-and-observe principle. The peculiarity of this
technique is in how the perturbation is realized. Instead of Decoupling
performing a small-signal frequency sweep which requires long
d’q d’d Id_ref
injection times, a short-time small-signal PRBS is used as test +
-
+
-
+
GC_I GC_V Vd_ref
signal. PRBS is a digital approximation of white noise, + + +
therefore wideband in nature, and this allows creating a short- +
- + -
+V
GC_I GC_V q_ref
time wideband perturbation at the impedance measurement + + +
Iq_ref
point. Current Voltage
Controller Controller
Fig. 8 (a) shows the block diagram for the wide bandwidth a b c
identification of three-phase AC power grid impedance in dq
domain. The inverter, in addition to performing its original
from PRBS to Processing
power conversion function in either current or voltage mode, generator
serves as power amplifier for the injected PRBS and as sensor (b)
of voltages and currents in dq domain. To identify the
Fig. 8. Block diagram of the WSI technique to identify the grid impedance
impedance of the three-phase system under test, i.e. the grid (a), and detailed inverter control implementation and its interface to the WSI
impedance in dq coordinates, the voltages and currents tool (b).
measurements in dq domain are processed in the integrated
control platform to identify in real time the parametric Now, the operation in voltage mode is analyzed. By
impedance ZG_dq(s). Notice that both the PRBS generator and inspection of Fig. 9 with the switch connecting the outer voltage
processing are implemented in such an integrated control loop, the output voltage is given as follows.
platform. 1
1 + 𝑇𝐼
In practice (see Fig. 8 (b)), small-signal PRBS is added to the 𝑣 = 𝐺𝑣𝑑 𝑃𝑅𝐵𝑆_1
𝑍 𝐺𝑣𝑑 𝑇𝐼
duty cycle and to both the current and voltage reference signals 1 + 𝑜𝑢𝑡 + 𝑇𝑉
𝑍𝑙𝑜𝑎𝑑 𝐺𝑖𝐿𝑑 𝐺𝑣𝑐_𝐼 1 + 𝑇𝐼
of the inverter controller. The reason for such a choice is linked 𝑇𝐼
to the frequency response of the closed-loop inverter. Fig. 9 𝐺𝑣𝑑 1 + 𝑇𝐼
+
𝐺𝑖𝐿𝑑 1 + 𝑍𝑜𝑢𝑡 + 𝑇 𝐺𝑣𝑑 𝑇𝐼
𝑃𝑅𝐵𝑆_2 (22)
shows a simplified small-signal block diagram of a converter 𝑍𝑙𝑜𝑎𝑑 𝑉𝐺 𝐺 1 + 𝑇
𝑖𝐿 𝑑 𝑣𝑐𝐼 𝐼
employing an inner current loop and an optional outer voltage 𝑇𝐼
loop. The transfer function Zload represents the impedance of the +𝑇𝑉
𝐺𝑣𝑑 1 + 𝑇𝐼
𝑃𝑅𝐵𝑆_3
load subsystem, and PRBS_1, PRBS_2, and PRBS_3 represent 𝐺𝑖𝐿𝑑 𝐺𝑣𝑐_𝐼 1 + 𝑍𝑜𝑢𝑡 + 𝑇
𝑉𝐺
𝐺𝑣𝑑 𝑇𝐼
𝑍𝑙𝑜𝑎𝑑 𝐺 1 + 𝑇
the disturbance injection at the duty cycle, current reference, 𝑖𝐿 𝑑 𝑣𝑐_𝐼 𝐼
where TV is the voltage control loop gain defined in Fig. 9 and
and voltage reference, respectively.
The operation in current mode is analyzed first. By Gvc_I is the control current to output transfer function [37].
inspection of Fig. 9 with the switch connected to Id_ref, the The controllers under consideration in this paper are PI
output voltage is given as follows. controllers. If correctly designed, the current loop is very large
1 𝑇𝐼 in magnitude within the control bandwidth, while it is very
1 + 𝑇𝐼 𝐺𝑣𝑑 1 + 𝑇𝐼 small beyond the control bandwidth [37]. Therefore, the
𝑣 = 𝐺𝑣𝑑 𝑃𝑅𝐵𝑆_1 + 𝑃𝑅𝐵𝑆_2 (21)
𝑍
1 + 𝑜𝑢𝑡 𝐺𝑖𝐿𝑑 1 + 𝑍𝑜𝑢𝑡 following quantities can be approximated as
𝑍𝑙𝑜𝑎𝑑 𝑍𝑙𝑜𝑎𝑑 𝑇𝐼 1 𝑓𝑜𝑟 ‖𝑇𝐼 ‖ ≫ 1
where Gvd is the control to output transfer function, GiLd is the 1 + 𝑇𝐼
≈{
𝑇𝐼 𝑓𝑜𝑟 ‖𝑇𝐼 ‖ ≪ 1
(23)
1
1
𝑓𝑜𝑟 ‖𝑇𝐼 ‖ ≫ 1 balanced three-phase grid seen at the PCC, which is modeled as
1 + 𝑇𝐼
≈ {𝑇𝐼 (24) Y-connected configuration as depicted in Fig. 8 (a). In order to
1 𝑓𝑜𝑟 ‖𝑇𝐼 ‖ ≪ 1 identify all the dq components of the grid impedance, PRBS is
By using the results of Eqs. (23) and (24), the following first injected on the d-component only and then on the q-
statements are true: component only according to the injection scheme shown in
 If only PRBS_1 is applied, v has smaller amplitude Fig. 8 (b). Assuming that the system frequency is constant, the
than PRBS_1 within the frequency range where TI four impedances are sequencially constructed as follows:
is very large, i.e. within the control bandwidth,
while it has the same amplitude at larger
𝑍𝐺_𝑑𝑑 = 𝑣𝑑 ⁄𝑖𝑑 𝑎𝑛𝑑 𝑍𝐺_𝑑𝑞 = 𝑣𝑑 ⁄𝑖𝑞 (25)
frequencies. 𝑍𝐺_𝑞𝑞 = 𝑣𝑞 ⁄𝑖𝑞 𝑎𝑛𝑑 𝑍𝐺_𝑞𝑑 = 𝑣𝑞 ⁄𝑖𝑑 (26)
 If PRBS_2 (or PRBS_3) is applied, v has the same
amplitude of the injected disturbance within the Notice that orthogonal PRBSs can be generated which allow
frequency range where TI is very large, i.e. within simultaneously identifying d and q components of the
the control bandwidth, while is reduced in impedance of the three-phase system under test [21].
amplitude at larger frequencies. B. Inverter Output Impedance Identification in dq domain
In other words, if PRBS is injected to the duty cycle only,
This subsection shows how the same WSI tool can be used
PRBS, seen as a disturbance, is rejected within the bandwidth
to online identify the output impedance of the inverter with few
of the selected control loop (either current or voltage), while it
modifications and without the addition of any extra power
is not rejected beyond the bandwidth of the selected control
hardware. Fig. 10 shows how the WSI tool is connected to the
loop. On the other hand, if PRBS is injected to the current or
grid-tied inverter system. In this case, PRBS is injected into the
voltage reference signals, it is not rejected within the bandwidth
three-phase grid voltage sources of the load subsystem. Since
of the selected current or voltage control loop, respectively.
PRBS is generated in the dq domain, it has to be transformed in
Therefore, injecting PRBS to the duty cycle and to either the
the abc reference frame to be injected into these grid voltage
current or voltage reference signals ensures that PRBS is not
sources. This is accomplished by using a dedicate Phase Locked
rejected by the selected control action over a wide frequency
Loop (PLL). This representation is surely fictitious but clearly
range.
mimics, in a lumped manner, a more complex system, e.g. that
Notice from Fig. 8 (b) that the injected PRBS requires proper
one of Fig. 1 (b), where other grid-tied inverters are also able to
scaling to obtain the desired controlled perturbation amplitude
perform the injection task (but also the identification task) on
at the impedance measurement point. This can be accomplished
the top of their power conversion function. To identify the
by properly choosing the scaling factors a, b, and c before being
impedance of the three-phase system under test, i.e. the inverter
added to the duty cycle, current reference, and voltage
output impedance, the voltages and currents measurements in
reference, respectively. To maintain a small-signal perturbation
dq domain are processed in the integrated control platform to
at the impedance measurement point, e.g. between 5% and 10%
identify in real time the parametric impedance ZS_dq(s).
of AC voltage and current steady state values, and considering Three-Phase System under test
that 𝑃𝑅𝐵𝑆_1 = 𝑃𝑅𝐵𝑆_𝑐 ∙ 𝑎 (likewise 𝑃𝑅𝐵𝑆_2 = 𝑃𝑅𝐵𝑆_𝑐 ∙ 𝑏 Three-Phase Inverter
PCC
and 𝑃𝑅𝐵𝑆_3 = 𝑃𝑅𝐵𝑆_𝑐 ∙ 𝑐), where PRBS_c is that one
coming from the integrated control platform, the scaling factors VS LC Filter
a and b (or c) can be calculated from (21) and (22) in the
frequency range where PRBS is not rejected. Notice that Eabc
ZS_dq
because of (23) and (24), special care is required in the PWM
PLL
calculations of a and b (or c) in order to obtain a flat
θ abc
perturbation over the spectrum of interest. Controller Vabc, Iabc dq
1/Zload measurements
Vdq, Idq
measurements WSI tool
Converter
iload v ZS_dq (s)
Zout_ OL(s) - Processing
PRBS
generation
+
d Gvd _ OL (s) Gvc_I GC_V = TV
iL Fig. 10. Block diagram of the WSI technique to identify the inverter output
GiLd _ OL(s)
impedance.
TI = GC_I GiLd
Id_ref According to the implementation described above, to identify
- + - +
+ GC_I GC_V Vref all the dq components of the inverter output impedance, PRBS
+ + ic + is first generated on the d-component only and then on the q-
component only. Then, thanks to the use of the dedicated PLL,
PRBS_1 PRBS_2 PRBS_3
PRBS is synchronously added to the three-phase voltage
Fig. 9. Simplified small-signal block diagram representation of a converter sources. Assuming that the system frequency is constant, the
with inner current loop and outer voltage loop as well as PRBS injection. four impedances are constructed as follows:
The impedances to be identified are the dq domain

𝑍𝑆_𝑑𝑑 = −𝑣𝑑 ⁄𝑖𝑑 𝑎𝑛𝑑 𝑍𝑆_𝑑𝑞 = −𝑣𝑑 ⁄𝑖𝑞 (27)
impedances of the Thévenin equivalent impedances of the
𝑍𝐺_𝑞𝑞 = −𝑣𝑞 ⁄𝑖𝑞 𝑎𝑛𝑑 𝑍𝑆_𝑞𝑑 = −𝑣𝑞 ⁄𝑖𝑑 (28) the PRBS injection and identification mechanisms explained in
Sections IV.A and IV.B. The grid-connected system consists of
Notice that the positive direction of the current is from the a feedback-controlled three-leg bridge inverter and grid
source to the load subsystem. impedance, whose full description was given in Section III. To
V. THE HARDWARE-IN-THE-LOOP REAL-TIME SETUP implement the bridge switches with no dead-time, a special RT
block from OPAL-RT libraries is used as it will be explained in
In this section, the HIL real-time setup for online parametric Section V.D. The PRBS signal is sequentially injected upon the
three-phase impedance identification is presented. The d- and q- components of the control loop of the inverter as
description of the full implementation of the WSI technique in explained in Section III. The dq components of currents and
the integrated control platform is provided. voltages at the PCC are used to construct the dq wideband grid
A. The HIL Setup impedances according to the procedure explained in Section III.
The simulation model implemented in OPAL-RT runs in the
The HIL setup, shown in Fig. 11, consists of a real-time
CPU, while the D/A and A/D conversions are implemented in
simulator, i.e. eMEGAsim from OPAL-RT Technologies [34],
the FPGA. The simulation time step is 25 µs.
where a digitally-controlled switching-type inverter connected
to a lumped grid model is implemented. OPAL-RT is connected C. The Implementation of the WSI Technique
to a RT host PC with the NI LabVIEW RT OS and a NI PCIe- The implementation of the WSI algorithm is divided in
7841R RIO [35], where the PRBS generator, data acquisition multiple stages as depicted at the bottom part of Fig. 11. The
(DAQ) and calculation of grid impedance are implemented. first stage consists of the PRBS generation and simultaneous
The eMEGAsim simulator is equipped with an OP5600 data acquisition (DAQ). The following stages realize the data
hardware platform. The hardware configuration of the processing and identification of the parametric impedance.
simulator includes two six-core Intel CPUs and a Xilinx Virtex- 1) PRBS Generation
6 Field Programmable Gate Array (FPGA). Moreover, the To generate the PRBS, a 15-bit linear feedback shift-register
simulator is equipped with 16 16-bit AOs and 16 16-bit AIs, (LFSR) is implemented in FPGA. The XOR-ed value of bit 14
and 32 DOs and 32 DIs. The eMEGAsim is based on RT-LAB, and bit 15 are fed back to the beginning of the register, as
an open real-time simulation environment, which is fully depicted in Fig. 12. The last value of the register is shifted to
integrated with Simulink/Simscape by Mathworks. The RT- achieve a white-noise approximation with zero mean [14].
LAB for OPAL-RT systems link Simulink models to both the Then, PRBS is routed to OPAL-RT using two channels (one for
CPUs and the FPGA. RT-LAB generates and compiles real- d-component and one for q-component) of the analog output
time code that runs on the OP5600 hardware. The RT-LAB port at a rate that can be chosen (25 µs for this HIL setup). The
augments the standard Simulink library with a custom blockset PRBS signal is then properly scaled and added to the duty
that gives the user access to I/O ports, PWM, power electronic, cycles and the control reference signals of the inverter, as
and various other event handling and timing systems on the shown in Fig. 8 (b). The amplitude of PRBS can be selected; it
OP5600 hardware. is recommended to produce a perturbation between 5% and
As previously stated, the PRBS generator, data acquisition 10% of the steady-state operating point as explained in Section
and calculation of grid impedance are implemented in an RT IV.
host PC through LabVIEW. This choice is justified by the fact 2) Data Acquisition (DAQ)
that the implementation of the impedance identification While PRBS is being generated and injected, taking
technique requires a large amount of memory and high- advantage of parallel computation of the FPGA, the current and
performance ADCs and DACs. For this purpose, the chosen voltage signals in dq coordinates coming from OPAL-RT are
platform is equipped with a Direct Memory Access (DMA) sampled using four AIs with a sampling rate that can be chosen
FIFO channel to buffer the data, and 8 16-bit AIs and 8 16-bit (25 µs for this HIL setup). The acquired time window during
AOs. Moreover, the NI RIO device contains a Virtex-5 FPGA which PRBS is injected and voltages and currents are measured
programmable with the LabVIEW FPGA Module that can be can also be chosen (200 ms for the results presented in this
used for this implementation due to its inherent multi-threaded paper). The voltage and current samples are directly written in
architecture, and its ability to interface to ADC/DAC and the memory of the host PC through a DMA channel. This “First-
memory. In First-Out” (FIFO) architecture allows buffering the sampled
The chosen HIL setup emphasizes the fact that the values and processing them as a batch after the DAQ is
identification tool can be simply integrated into an existing completed. The PRBS signal is only injected during the DAQ
inverter controller as long as the tool has access to the inverter phase and turned off immediately after the complete time
measurements in dq domain and has the ability to add the PRBS window is acquired.
injection over the inverter control loop as explained in Section 1) Fast Fourier Transform (FFT)
IV. Notice that implementation in an embedded controller is The proper voltage and current in the dq domain are selected
feasible if the chosen platform that hosts the WSI tool is to build the identified impedance according to (25)-(26) for the
provided with enough dedicated memory. grid impedance or to (27)-(28) for the inverter output
B. The Simulated System impedance. Voltage and current spectra are calculated by
performing an FFT algorithm. A Hanning windowing is
The upper part of Fig. 11 shows the block diagram of the
applied.
system, running in the real-time digital simulator, to identify the
grid impedance and the inverter output impedance according to
Three-Phase Inverter so, the lower and upper frequency boundaries for the data
PCC thinning routine have to be specified. All data points outside
Three-
LC Filter Phase
this window are disregarded for further processing. The least
VS
System square fitting routine matches the thinned data to a polynomial
function and it is an algorithm based on Levy [44]. The order of
PWM
numerator n and denominator m must be selected. If they are
not known a priori, the order can be tentatively estimated by
Vabc, Iabc
measurements
increasing n and m until the fitting result matches the non-
Controller
parametric impedance. The result of the fitting routine is the
OPAL-RT
parametric impedance, given by the coefficients of a
polynomial function, whose generic representation is also given
in Fig. 11.
PRBS a
and Correction 1 t
DAQ and -a
FFT t
Calcultation
NI PCIe-7841R RIO FIFO
 x 0  j  y 0  Non-parametric a
x  jy  Impedance 1 2 3 14 15 PRBS
 1 1  1
PRBS Time Window Windowing
type


 

  
Z e j k
settings size (ms)  xk  j  yk 
Fitting Routine XOR

Least Square Fig. 12. Block diagram of LFSR for PRBS generation.
Thinning
Fitting
a0  a1s   an1sn1  sn
b0  b1s   bm1sm1  sm D. Performance of the WSI Technique and Overcoming
Minimum, maximum Order n and m
Parametric
Impedance Z s Practical Challenges
frequency, and number of
data points The real time performance of the WSI technique is given by
LabVIEW Real-Time Host PC
the execution time of all the routines described in the previous
Fig.11. Block diagram of the HIL setup showing both the real-time subsection starting from when the PRBS injection is activated
simulation system and the WSI technique implementation.Correction and and ending when the parametric impedance result is available.
Calculation The execution time of the implemented WSI technique consists
of the chosen time window and the impedance calculation time.
A correction routine is performed to cancel the existing The impedance calculation time comprises the execution time
harmonics besides the AC system fundamental frequency of the FFT algorithm, which is of complexity 𝑂(𝑁 ⋅ 𝑙𝑜𝑔(𝑁)),
because this latter is already canceled out by the synchronous and the fitting routine, which depends on exit conditions of the
abc-to-qd transformation. The cancellation of harmonics is least-square fitting algorithm [44]. Both the algorithms are
performed by subtracting the voltage and current spectra estimated to have an execution time of 20 ms. Therefore, to
obtained under PRBS perturbation from those obtained without measure a single component of the impedance of the system
any injection [25]-[28]. Then, dividing voltage and current under test, the implemented WSI technique takes about
spectra yields the non-parametric grid impedance as follows 𝑡𝑤𝑖𝑛𝑑𝑜𝑤 + 20 𝑚𝑠.
𝐹𝐹𝑇{𝑣[𝑛]} A major challenge in impedance identification consists of
𝑍[𝑒𝑗𝜔𝑛 ] = (29) maintaining the small-signal condition and, at the same time,
𝐹𝐹𝑇{𝑖[𝑛]}
where 𝑣[𝑛] and 𝑖[𝑛] are the voltage and current samples. The guaranteeing a measurable perturbation within the frequency
non-parametric impedance consisting of N complex data points: range of interest. In theory, the maximum identifiable
𝑁 = 𝑡𝑤𝑖𝑛𝑑𝑜𝑤 ⋅ 𝑓𝑠𝑎𝑚𝑝𝑙𝑖𝑛𝑔 (30) frequency is defined by the Nyquist frequency, i.e. half the
switching frequency of the inverter [14]-[20]. The lower
where 𝑡𝑤𝑖𝑛𝑑𝑜𝑤 is the observation window during which PRBS frequency boundary, instead, is equal to the inverse of the
is injected and voltage and current are measured and 𝑓𝑠𝑎𝑚𝑝𝑙𝑖𝑛𝑔 measured time window 𝑡𝑤𝑖𝑛𝑑𝑜𝑤 during which the PRBS is
is the sampling frequency of 𝑣[𝑛] and 𝑖[𝑛]. injected. In practice, the upper boundary depends on the
2) Fitting Routine amplitude of the injected PRBS signal. In fact, the LC filter of
The fitting routing consists of a real time algorithm able to the inverter attenuates the high frequency perturbation,
return the parametric impedance from the non-parametric data eventually leading to a signal level that is comparable with the
set. The complete fitting routine is implemented using a noise floor (< −60 𝑑𝐵) and therefore too small to be identified.
“Mathscript Node”, which allows including MATLAB code to As a consequence, the attenuation at high frequencies
be executed on the LabVIEW Real-Time Host PC. This yields introduced by the LC filter dictates the minimum amplitude of
access to a large number of powerful functions for data the injected PRBS signal. Therefore, to obtain a good
manipulation and processing. The fitting consists of two steps: identification as close as possible up to the Nyquist frequency,
data thinning and least square fitting. The thinning enforces despite the attenuation of the LC filter, and maintaining at the
equal weighting over the full frequency window of the non- same time the small-signal condition, a good choice is to
parametric data. The thinning technique is used to obtain a perturb voltage and current with magnitude between 5% and
logarithmically spaced subset of the data points. In order to do 10% of their steady-state value. As an alternative solution, a
different perturbation signal may be used; an alternative to time step gives the number of duty cycle values that are possible
white noise is blue noise which increases the high-frequency in the simulation environment, which determines the effective
content of the PRBS without affecting the low-frequency PWM resolution in the simulation. For the case presented in this
content. The implementation of a blue noise filter for paper, the switching frequency is 20 kHz (Ts = 50 µs) and the
identification purposes in described in [16]-[17]. As another fixed simulation time step is 25 µs. As a result, in the simulation
efficient alternative, discrete-interval binary sequence can be only 2 possible values of duty cycle would be possible. To
used [48]. This latter technique allows specifying the harmonics overcome this problem of poor duty cycle resolution, the PWM
so that the injection spectral energy can be maximized without and the Time-Stamped Bridge modules, available in the OPAL-
the need of increasing the pertubation signal amplitude. RT’s RTeDriveTM and RT-EventsTM libraries, replace the
The choice of the time window is another practical challenge counterpart PWM and Universal Bridge blocks available in
worthy of discussion. As the time window is directly linked to SimPower Systems library. These special Simulink models can
the number N of complex data points of the non-parametric run on the CPU real-time target [39] and are supported by the
impedance according to (30), for a fixed sampling rate, a long solver eDRIVEsimTM, which is able to decouple the switching
enough time window should be selected in order to have enough period from the simulation time step. This allows improving the
data points to capture eventual sharp features of the identified time resolution of a switch-mode model for a given switching
impedance, such as lightly damped resonances. On the other frequency and a fixed simulation time step.
hand, the time window should be chosen short enough to avoid
increasing impedance calculation times. Moreover, how short VI. RESULTS OF THE ONLINE IMPEDANCE IDENTIFICATION AND
the time window should be chosen is in relation to the STABILITY ANALYSIS
characteristics of the DUT. For the Power-Electronics-based In this section, both time and frequency domain results of the
AC Power System of Fig. 1 (b), it is important to set the time HIL setup are presented and discussed. These include the
window short enough in order to catch with fast changing results of the online parametric identification of all the dq
impedances for example due to load steps or system components of the interface impedances of the grid-tied
reconfiguration. current-controlled inverter system compared to the analytic
Cancellation of existing harmonics is another challenge that impedances derived in Section III. The identified parametric
requires special attention. The grid-connected power electronic impedances are used in the system stability analysis according
system is time-variant and the correction procedure as to the generalized Nyquist stability criterion applied to the
described earlier is affected by this characteristic. The FFT minor loop (13), i.e. that one when the source is current driven.
assumes infinite periodicity of the signal to be transformed. Fig. 13 shows a screenshot of the GUI used to set the settings
However, such an assumption is not practical because all the of the WSI tool implemented in LabVIEW and for online result
signal acquisitions are limited in time. During the acquisition, visualization. The GUI displays the following plots: at the top
the FFT returns additional spurious frequency components from left to right, acquired voltage and current measurements
around the existing harmonic content because of the during the selected time window, voltage and current spectra
discontinuities at the edges of the time window when the after the FFT to the acquired voltage and current, and Bode plot
number of the acquired periods is not an integer. This well- of the identified non-parametric impedance as result of (20); at
known problem is called spectral leakage [38] and it is the main the bottom from center to left, the Bode plot of the thinned non-
reason for the non-perfect harmonic cancellation. In order to parametric impedance vs. the analytic impedance, and the Bode
minimize the effect of spectral leakage, two recommendations plot of the fitted parametric impedance vs. the analytic
are made: 1) creation of Hanning window of the same length as impedance. PRBS injection and DAQ settings are on the left
the acquired signal, and 2) ensuring periodicity of the injection hand side of the GUI. The user can choose the perturbation
in synchronism with the acquisition over an integer number of amplitude and rate as well as the channel of injection (three
fundamental frequency cycles (integer multiples of 20 ms for channels are available, but channel 1 is mapped to the d-
50 Hz or 16.67 ms for 60 Hz). component and channel 2 to the q-component). The sampling
Another practical aspect is the resolution of the D/A and A/D rate of voltage and current measurements can also be chosen as
converters of both the real-time digital simulator and the NI well as the time window during which PRBS in injected and
RIO device. Low resolutions increase the minimum viable signals are sampled. To construct the impedance of interest,
amplitude of the perturbation. Fortunately, both OPAL-RT and proper voltage and current measurements have to be selected;
NI PCIe-7841R RIO are equipped with 16-bit D/A and A/D for the screenshot of Fig. 13, the ratio V1/I1 returns the
converters which give a resolution of 9.49 mV on a grid voltage impedance Zdd. The fitting settings are also available; the user
measurement of 220 Vrms and a resolution of 0.31 mV on the can choose the number of thinned data points as well as the
injected PRBS of ±10 V. This setup enables accurate minimum and maximum frequencies that will be processed by
measurement even in the presence of small injected the least square fitting routine. This latter requires the user to
perturbations. specify the orders of the numerator and denominator. The result
As far as the implemented real-time simulation in OPAL-RT, of the fitting is the parametric fitted transfer function in the
a proper choice of the simulation time step is critical. Such a following form:
choice allows obtaining not only reliable simulation results but 𝑎0 + 𝑎1 𝑠 + ⋯ + 𝑎𝑛−1 𝑠 𝑛−1 + 𝑎𝑛 𝑠 𝑛
𝑍𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑟𝑖𝑐 (s) = (31)
also an accurate identification, especially at high frequencies. 𝑏0 + 𝑏1 𝑠 + ⋯ + 𝑏𝑚−1 𝑠 𝑚−1 + 𝑏𝑚 𝑠 𝑚
The ratio of simulated inverter switching period and simulation
Fig. 13. Screenshot of the GUI of the WSI tool implemented in LabVIEW.
Alternatively, the circuital parameters can be extracted from validation of the accurateness of the identification technique
the fitted parametric impedance if the topology of the system with the present HIL setup, the online identified non-parametric
under test is known. Both the fitted transfer function in the form and parametric grid impedances are compared in real time to
of (31) and the identified circuital parameters are shown at the the counterpart analytic model and plotted in the GUI as shown
bottom left corner of screenshot of the GUI and are updated in at the right bottom of the screenshot of Fig. 13. For a better
real time (every 𝑡𝑤𝑖𝑛𝑑𝑜𝑤 + 20 𝑚𝑠 as explained earlier). Some visualization, the captures of the Bode plots from the GUI are
action buttons are displayed in the middle of the GUI. It is shown in Figs. 15-16 for the grid impedance labeled as “Case
possible to perform the online parametric identification in 1” and “Case 2” (refer to Section III-B). Fig. 15 shows the Bode
single shot mode or looping mode (i.e. continuously injecting plot of the thinned non-parametric impedance vs. the analytic
PRBS and identifying the impedance), to activate the impedance, while Fig. 16 plots the parametric impedance vs.
cancellation mechanism of the existing frequencies as well as the analytic impedance. Only dd and qq components are
to activate the windowing to the FFT algorithm. Moreover, it is parametrized (fitted transfer functions are in Appendix) because
possible to send, via User Datagram Protocol (UDP), the they are the components needed for the stability analysis
coefficients of the fitted parametric impedance, i.e. the vectors according to (13). In all cases, very good matching is evident.
[𝑎0 𝑎1 ⋯ 𝑎𝑛−1 𝑎𝑛 ] and [𝑏0 𝑏1 ⋯ 𝑏𝑚−1 𝑏𝑚 ], to a receiver PC to
perform remote online stability analysis. The coefficients are
transmitted as single-precision floating-point values consisting
of 4 Bytes each.
A. Online Parametric Identification of the Grid Impedance
Initially the system is in steady-state without PRBS injection. PRBS injection starts
Then, the identification procedure is initiated by injection of
10% 15-bit white noise PRBS over the d-channel first for a time
window of 200 ms and then over the q-channel for the same
amount of time. Fig. 14 shows a time-domain capture of the
inverter’s voltage and current at the PCC during PRBS injection
and without it. It can be seen that injected perturbation does not Fig. 14. Three-phase voltage current at the PCC during PRBS injection
introduce an excessive amount of noise to the system. compared to their steady-state trend.
Non-parametric grid impedances are identified by using (29)
to (25) and (26). Then, the thinning technique is used to obtain B. Online Parametric Identification of the Inverter Output
a logarithmically spaced subset of 800 data points in the Impedance
frequency range of 10Hz-10kHz, and the least square fitting The identification of the inverter output impedance is more
routine returns the parametric impedance. For the sake of challenging than that one of the grid impedance due, on the one
hand, to the switching nature of the inverter, and, on the other suffers from poor accuracy at low frequencies when the current
hand, to the rejection effect of the current control. Looking at spectrum hits the noise floor. To partially cope with such a
Fig. 10, PRBS is seen by the inverter as a load current difficulty, the lower frequency limit for the fitting routine was
disturbance. Similarly as done in Section IV-A, it can be set to 30 Hz. The same time window of 200 µs and 800 number
demonstrated that PRBS is rejected within the current control of thinned points are used.
bandwidth, while it is not at higher frequencies. As a result, the
non-parametric identification of the inverter output impedance
Case 1 Case 2
Bode Plot of Non- Bode Plot of Non- Bode Plot of Non- Bode Plot of Non-
parametric ZG_dd
Load Impedance Magnitude (non-parametric & thinned)
parametric ZG_dq
Load Impedance Magnitude (non-parametric & thinned)
parametric ZG_dd parametric ZG_dq
Load Impedance Magnitude (non-parametric & thinned) Load Impedance Magnitude (non-parametric & thinned)
45 20
60 40
40
Thinned Data Points 30
Magnitude (db)
Thinned Data Points
Magnitude (db)
10 50 Thinned Data Points Thinned Data Points

35
Analytic Function Analytic Function Analytic Function 20 Analytic Function
30 40
0 10
25
30 0
20
-10
15 -10
20
10 -20 -20
10
5 -30
0 -30 0 -40
10 100 1000 10000 10 100 1000 10000 10 100 1000 10000 10 100 1000 10000
Load Impedance Phase (non-parametric & thinned) Load Impedance Phase (non-parametric
Frequency (Hz) Frequency (Hz) & thinned) Load Impedance Phase (non-parametric
Frequency (Hz) & thinned) Load Impedance Phase (non-parametric
Frequency (Hz) & thinned)
100 200 100 200
75
Phase (degree)
150
Phase (degree)
75 150
50 100 50 100
25 50 25 50
0 0 0 0
-25 -50 -25 -50
-50 -100 -50 -100
-75 -150 -75 -150
-100 -200 -200
-100
10 100 1000 10000 10 100 1000 10000 10 100 1000 10000
10 100 1000 10000
Frequency (Hz) Frequency (Hz) Frequency (Hz)
Frequency (Hz)
Frequency (Hz) Frequency (Hz) Frequency (Hz) Frequency (Hz)
parametric ZG_qd parametric ZG_qq parametric ZG_qd parametric ZG_qq
Load Impedance Magnitude (non-parametric & thinned) Load Impedance Magnitude (non-parametric & thinned) Load Impedance Magnitude (non-parametric & thinned) Load Impedance Magnitude (non-parametric & thinned)
30 45 40 60
40
Magnitude (db)
20 30
Magnitude (db)
Thinned Data Points Thinned Data Points Thinned Data Points 50 Thinned Data Points
35 20
10 Analytic Function Analytic Function Analytic Function 40 Analytic Function
30 10
0 30
25 0
-10 20
20 -10
-20 10
15 -20
-30 10 -30 0
-40 5 -40 -10

10 100 1000 10000 10 100 1000 10000 10 100 1000 10000 10 100 1000 10000
Load Impedance Phase (non-parametric & thinned) Load Impedance Phase (non-parametric & thinned) Load Impedance Phase (non-parametric & thinned)
Load Impedance Phase (non-parametric & thinned) Frequency (Hz) Frequency (Hz)
Frequency (Hz) Frequency (Hz)
200 200 100
100
Phase (degree)
150 75
Phase (degree)
150 75
100 50
100 50
50 25
50 25
0 0
0 0
-50 -25
-50 -25
-100 -50
-100 -50
-150 -75
-150 -75
-200 -100
-200 -100
10 100 1000 10000 10 100 1000 10000
10 100 1000 10000 10 100 1000 10000
Fig. 15. Bode plot of the identified thinned non-parametric impedance vs. the analytic impedance for the grid impedance – Case 1 and Case 2.
Case 1 Case 2
Bode Plot of Bode Plot of Bode Plot of Bode Plot of
Parametric ZG_dd
Load Impedance Magnitude (parametric)
Parametric ZG_qq
Parametric ZG_dd
Load Impedance Magnitude (parametric) Parametric ZG_qq
60
60
45 45
50 Fitted Function
40 40 50 Fitted Function
Magnitude (db)
Magnitude (db)
Fitted Function Fitted Function

Analytic Function
35 35 40 Analytic Function
Analytic Function Analytic Function 40
30 30
30
30
25 25
20 20
20 20
15 15 10
10
10 10
0
0
5 5 10 100 1000 10000
10 100 1000 10000
10 100 1000 10000 10 100 1000 10000 Load Impedance Frequency
Phase (parametric) Load Impedance Phase (parametric)
Load Impedance Phase (parametric) Load Impedance Phase (parametric) (Hz)
Frequency (Hz)
Frequency (Hz) Frequency (Hz) 100 100
100 100
75 75 75
75
Phase (degree)
Phase (degree)
50 50 50
50
25 25 25
25
0 0 0
0
-25 -25 -25
-25
-50 -50 -50
-50
-75 -75 -75
-75
-100 -100 -100
-100
10 100 1000 10000 10 100 1000 10000 10 100 1000 10000
10 100 1000 10000
Frequency (Hz) Frequency (Hz) Frequency (Hz)
Frequency (Hz)
Fig. 16. Bode plot of the identified parametric impedance vs. the analytic impedance (only dd and qq components) for the grid impedance – Case 1 and Case 2.
parametric ZS_dd parametric ZS_dq parametric ZS_qd parametric ZS_qq
Load Impedance Magnitude (non-parametric & thinned) Load Impedance Magnitude (non-parametric & thinned) Load Impedance Magnitude (non-parametric & thinned) Load Impedance Magnitude (non-parametric & thinned)
40 20 40 20
Magnitude (db)
Magnitude (db)
30 30
0 0
20 20
-20 -20
10 10
-40 -40
0 0
Thinned Data Points Thinned Data Points Thinned Data Points Thinned Data Points
-60 -60
-10 Analytic Function Analytic Function -10 Analytic Function Analytic Function
-20 -80 -20 -80

10 100 1000 10000 10 100 1000 10000 10 100 1000 10000 10 100 1000 10000
Load Impedance Phase (non-parametric & thinned) Load Impedance Phase (non-parametric & thinned) Load Impedance Phase (non-parametric & thinned) Load Impedance Phase (non-parametric & thinned)
-10 200 0 200
-20 150 -20 150

Phase (degree)
Phase (degree)
-30 -40 100
100
-40 -60
50 50
-50 -80
0 0
-60 -100
-50 -50
-70 -120
-100 -140 -100
-80
-90 -150 -160 -150
-100 -200 -180 -200

10 100 1000 10000 10 100 1000 10000 10 100 1000 10000 10 100 1000 10000

Fig. 17. Bode plot of the thinned non-parametric impedance vs. the analytic impedance for the inverter output impedance.
Bode Plot of Bode Plot of

Parametric ZS_dd
Parametric ZS_qq
40 40
Magnitude (db)
30 30
20 20
10 10
0 0
Fitted Function Fitted Function
-10 Analytic Function -10 Analytic Function
-20 -20
10 100 1000 10000 10 100 Phase (parametric)
1000 10000
Load Impedance Phase (parametric) Load Impedance
-10 -10
-20 -20
Phase (degree)
-40 -40
-60 -60
-80 -80
-100 -100
10 100 1000 10000 10 100 1000 10000
Fig. 18. Bode plot of the parametric impedance vs. the analytic impedance (only dd and qq components) for the inverter output impedance.
Non-parametric grid impedances are identified by using (29) to coefficients, constructs the impedances according to (31), and,
(27) and (28). Figs. 17-18 show the captures of the Bode plots in line with (13), plots the Nyquist diagram of the following
of the inverter output impedance from the GUI of the WSI tool minor loop gains:
in LabVIEW. Fig. 17 shows the Bode plot of the thinned non- 𝑇𝑀𝐿𝐺_𝑑𝑑 =
𝑍𝑝𝑎𝑟𝑎𝑚𝑐_𝐺_𝑑𝑑 (𝑠)
(32)
parametric impedance vs. the analytic impedance, while Fig. 18 𝑍𝑝𝑎𝑟𝑎𝑚_𝑆_𝑑𝑑 (𝑠)
plots the parametric impedance vs. the analytic impedance. 𝑍𝑝𝑎𝑟𝑎𝑚_𝐺_𝑞𝑞 (𝑠)
𝑇𝑀𝐿𝐺_𝑞𝑞 =
𝑍𝑝𝑎𝑟𝑎𝑚_𝑆_𝑞𝑞 (𝑠)
(33)
Again, only dd and qq components are parametrized (fitted
transfer functions are in Appendix) because they are the It was demonstrated in Section III that both the inverter
components needed for the stability analysis according to (13). output impedance and grid impedance are diagonally dominant.
In all cases, again, very good matching is evident. This was also confirmed by the online measurements.
Therefore, it makes sense to perform the stability analysis on
C. Stability Analysis the minor loop gains (32) and (33). Fig. 19 shows the Nyquist
The current-controlled inverter system is tested with two grid diagram of the minor loop gains (32) and (33) that use the
impedances labeled as Case 1 and Case 2, whose circuital transmitted fitted parametric coefficients of the impedances
parameters are given in Table II. To perform the stability identified by the online WSI tool. The Nyquist contour for both
analysis of the inverter system, the fitted coefficients Case 1 and Case 2 do not encircle the (−1, 𝑗0) point, thus
[𝑎0 𝑎1 ⋯ 𝑎𝑛−1 𝑎𝑛 ] and [𝑏0 𝑏1 ⋯ 𝑏𝑚−1 𝑏𝑚 ] of the identified predicting a stable inverter system for both situations. However,
parametric impedances 𝑍𝑝𝑎𝑟𝑎𝑚_𝑆_𝑑𝑑 (𝑠), 𝑍𝑝𝑎𝑟𝑎𝑚_𝑆_𝑞𝑞 (𝑠), the phase margin for the inverter system with the grid
𝑍𝑝𝑎𝑟𝑎𝑚_𝐺_𝑑𝑑 (𝑠), and 𝑍𝑝𝑎𝑟𝑎𝑚_𝐺_𝑞𝑞 (𝑠) are transmitted via a UDP impedance labeled as Case 1 is larger than for Case 2. This
to a receiver PC. A MATLAB script in this PC loads the predicts that the inverter system with grid impedance labeled as
Case 1 is more robustly stable than Case 2. Notice that only controlled inverter system. The accuracy and performance of
𝑇𝑀𝐿𝐺_𝑑𝑑 (𝑠) could have been used for assessing the system the implemented WSI tool is evaluated through a HIL setup,
stability because 𝑍𝑝𝑎𝑟𝑎𝑚_𝑆_𝑑𝑑 (𝑠) = 𝑍𝑝𝑎𝑟𝑎𝑚_𝑆_𝑞𝑞 (𝑠) due to the comprising a grid-connected switch-mode feedback-controlled
fact that in this paper the dynamics of the PLL have been inverter running in real-time in OPAL-RT. The chosen HIL
neglected. If the dynamics of the PLL are included the system setup emphasizes the fact that the WSI identification tool can
becomes unsymmetric and 𝑍𝑝𝑎𝑟𝑎𝑚_𝑆_𝑑𝑑 (𝑠) ≠ 𝑍𝑝𝑎𝑟𝑎𝑚_𝑆_𝑞𝑞 (𝑠) be simply integrated into an existing inverter controller as long
[45]. Therefore, both (32) and (33) needed to be used to assess as the tool has access to the inverter voltage and current
the system stability [8]-[10]. measurements in dq domain, and has the possibility to add the
Fig. 20 depicts the transient current response of the inverter PRBS injection over the inverter control loop. The full
system in correspondence of a current step from Id_ref (defined description of the HIL setup and all the routines of the
in Fig. 4) to 50 A. When the inverter is connected to the grid implemented WSI tool are provided. Technical challenges are
impedance labeled as Case 2, the dq-components of the current also discussed and implemented solutions are presented.
result in a more oscillatory trend with respect to those of Case Frequency domain results reveal high accuracy of the
1, especially the q-component. The time domain results are implemented WSI tool in the dq domain parametric
perfectly in line with the frequency-domain stability prediction. identification of balanced three-phase interface impedances.
The online parametric identified interfaces are then used to
VII. CONCLUSIONS AND FUTURE WORK assess the stability of the grid-tied inverter system by applying
This paper presents the implementation of a noninvasive the generalized Nyquist stability criterion to the minor loop gain
online Wideband Systems Identification (WSI) technique in a of the inverter/grid interface. Results in the frequency domain
RT LabVIEW target to online identify three-phase parametric predict the system stability in accordance with the time domain
interface impedances in dq domain of a grid-tied feedback- results.
PM_Case1 = PM_Case1 =
54.2° @ 495 Hz 54.9° @ 505 Hz
PM_Case2 =
PM_Case2 =
31.8° @ 241 Hz
29.9° @ 237 Hz
(a) (b)
Fig. 19. Nyquist plot of 𝑇𝑀𝐿𝐺_𝑑𝑑 (𝑠) (a) and 𝑇𝑀𝐿𝐺_𝑞𝑞 (𝑠) (b) then the grid impedance is as Case 1 and Case 2.
decoupled positive- and negative-sequence converter

impedances, when phase- or dq-domain current control systems
are implemented” [47]. This feature enables several practical
advantages of the phase sequence domain over the dq domain
[46]: 1) SISO stability analysis for balanced three-phase
systems, 2) both balanced and unbalanced systems can be
handled, and 3) the impedances depend only on the converter
dynamics and not on the system frequency. Therefore, future
work will need to look into the phase sequence domain for the
impedance identification and make a comparative study to the
identification in dq domain.
Moreover, as underling assumptions, the paper has dealt with
Fig. 20. Current transients in corrispondence of a current step on the d ideal PLL and constant system frequency. In a real system
component when the grid impedance is as Case 1 and Case 2. where the PLL has its own dynamics and the system frequency
may vary, injecting a pure d-component and pure q-component
Notice that in the paper the identification of impedances and PRBS is challenging, and this may corrupt the impedance
stability analysis were performed in dq domain, thus resulting identification accuracy. Future work will look into a more
with some extra analysis to assess whether the off-diagonal
realistic situation when a real PLL and a system frequency
elements of the impedance matrices can be effectively
control are deployed. Practical requirements and conditions for
neglected. Such a difficulty is completely avoided if the phase the frequency control and the PLL to assure proper PRBS
sequence domain is used because this modeling method “yields injection and accurate impedance measurements will need to be
found. (1.67e04*s^3 + 1.161e08*s^2 + (1.669e04*s^3 + 7.563e07*s^2 +

8.865e11*s + 1.492e15)/(s^4 + 6.889e11*s + 6.342e11)/(s^4 +
Future work will also need to look into the dynamic effect of 7483*s^3 + 6.644e07*s^2 + 1.387e11*s 5115*s^3 + 5.449e07*s^2 + 4.46e10*s
a real PLL on the identified impedances. In such a case, it will + 5.086e13) + 4.581e13)
Zanalytic_S_dq = -Zanalytic_S_qd = Zanalytic_S_dq = -Zanalytic_S_qd =
be shown that the WSI tool will be able to measure different (47.68*s^3 + 6.261e06*s^2 + 7.7e09*s (5.85e06*s^4 + 8.486e09*s^3 +
parametric impedance dd and qq components (phase of qq + 2.968e14)/(s^4 + 1.259e04*s^3 + 2.582e14*s^2 - 4.978e15*s +
components starts at −180° at low frequency), but the dq and 8.89e07*s^2 + 5.447e11*s + 2.317e14) 4.037e17)/(s^6 + 1.171e04*s^5 +
8.768e07*s^4 + 5.462e11*s^3 +
qd components are the same in magnitude (except 180° phase 5.017e14*s^2 + 4.747e17*s + 4.971e19)
shift). In practice, from the online parametric impedance Analytic expressions for the grid impedance:
identification, the negative incremental resistance behaviors of Grid Impedance – Case 1 Grid Impedance – Case 2
Zanalytic_G_dd = Zanalytic_G_qq = Zanalytic_G_dd = Zanalytic_G_qq =
output and input impedances of grid-connected feedback- (3.386e05*s + 8.211e08)/(s^2 + 2573*s (2.071e05*s + 6.504e07)/(s^2 + 390.9*s
controlled inverters and active rectifiers will be captured. From + 3.335e08) + 3.983e07)
the stability analysis standpoint, therefore, the PLL will be Zanalytic_G_dq = -Zanalytic_G_qd = Zanalytic_G_dq = -Zanalytic_G_qd =
(1.057e08*s^2 + 6.498e11*s - (6.723e07*s^2 + 5.083e10*s -
identified as the main responsible for potential stability 3.473e16)/(s^4 + 1.154e04*s^3 + 2.339e15)/(s^4 + 4442*s^3 +
problems in LV Power-Electronics-based AC Power Systems. 7.134e08*s^2 + 3.901e12*s + 1.188e17) 8.158e07*s^2 + 1.791e11*s + 1.61e15)
This result will be compared to those of [8]-[10], where The identified parametric impedances obtained by the WSI
nonparametric impedances were used for stability analysis. tool in LabVIEW and used for the stability analysis are the
Also, the practical advantages of parametric identification over followings.
Grid Impedance – Case 1 Grid Impedance – Case 2
nonparametric identification will be sistematically presented in Zparam_G_dd = Zparam_G_dd =
a special remote online stability analysis setup. By referring to (-24.3e03 + 777.9e06*s + (59.7e06 + 200.9e03*s)/(40.3e06 +
Fig. 21, a real-time application in a secondary substation will 328.1e03*s^2)/(3.9e06 + 333.6e06*s + 347.2e0*s + 1.0e0*s^2)
2.7e03*s^2 +1.0e0*s^3)
not only be coordinating the injection of PRBS at the converters Zparamc_G_qq = Zparam_G_qq =
connected to the several PCCs, but also acquiring from a (5.6e03 + 783.7e06*s (53.9e06 + 208.9e03*s)/(40.2e06 +
+328.5e03*s^2)/(125.7e03 +333.5e06*s 358.7e0*s + 1.0e0*s^2)
communication link the coefficients of the parametric +2.7e03*s^2 +1.0e0*s^3)
impedance identifications (red arrows) and build the Inverter in Current Mode
corrispondent minor loop gains out of which stability margins Zparam_S_dd =
(946.1e09 + 125.5e06*s +16.5e03*s^2)/(32.7e09 + 76.7e06*s + 7.3e03*s^2
can be calculated in near real-time as well. If stability margins +1.0e0*s^3)
violate certain conditions, then corrective stabilizing actions Zparam_S_qq =
(505.0e09 + 89.2e06*s +17.6e03*s^2)/(17.2e09 + 41.3e06*s +6.2e03*s^2
can be taken (blue arrows). +1.0e0*s^3)
Secondary ACKNOWLEDGMENT
Substation
This work is supported by the European Union’s Horizon
2020 research and innovation programme under grant
agreement No 727481.
Z0
0
AC 1 2
REFERENCES
AC [1] M. Liserre, T. Sauter and J. Y. Hung, "Future Energy Systems: Integrating
MV/LV
ZS1
Zg1
ZL1 ZS2
Zg2
ZL2
Renewable Energy Sources into the Smart Power Grid Through Industrial
Electronics," in IEEE Industrial Electronics Magazine, vol. 4, no. 1, pp.
AC AC AC AC 18-37, March 2010
DC DC DC DC
[2] D. Bogdanov, O. Koskinen, A. Aghahosseini and C. Breyer, "Integrated
VS1 ZLL VSS ZLL renewable energy based power system for Europe, Eurasia and MENA
1 2 2 regions," 2016 International Energy and Sustainability Conference
Fig. 21. Power-Electronics-based LV Feeder equipped with a (IESC), Cologne, 2016, pp. 1-9
communication link to perform stability analysis at substation level. [3] B. K. Bose, "Global Energy Scenario and Impact of Power Electronics in
21st Century," in IEEE Transactions on Industrial Electronics, vol. 60,
Finally, another piece of future work will focus on the no. 7, pp. 2638-2651, July 2013
derivation of a modeling framework in dq domain of grid- [4] T. Guillod, F. Krismer and J. W. Kolar, "Protection of MV Converters in
connected power electronic converters with real PLLs. This the Grid: The Case of MV/LV Solid-State Transformers," in IEEE
modeling framework will not only capture the dynamics of Journal of Emerging and Selected Topics in Power Electronics, vol. 5,
no. 1, pp. 393-408, March 2017
PLLs, but going beyond what already published in [8], it will
[5] J. Sun, "Impedance-Based Stability Criterion for Grid-Connected
include the requirement of taking into account the topology of Inverters," in IEEE Transactions on Power Electronics, vol. 26, no. 11,
the power system of Fig. 21 so that of being easily applicable pp. 3075-3078, Nov. 2011
to well-known impedance-based stability criteria. In practice, [6] Z. Bing, K. J. Karimi and J. Sun, "Input Impedance Modeling and
the modeling framework will end up with a matrix of a kind of Analysis of Line-Commutated Rectifiers," in IEEE Transactions on
mixed nodal admittance matrix which includes the inteface Power Electronics, vol. 24, no. 10, pp. 2338-2346, Oct. 2009
impedances at the various PCC of the system. [7] M. Liserre, R. Teodorescu and F. Blaabjerg, "Stability of photovoltaic and
wind turbine grid-connected inverters for a large set of grid impedance
APPENDIX values," in IEEE Transactions on Power Electronics, vol. 21, no. 1, pp.
263-272, Jan. 2006
Analytic expressions for the inverter output impedances: [8] B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli and Z. Shen, "Analysis
Inverter in Current Mode Inverter in Voltage Mode of D-Q Small-Signal Impedance of Grid-Tied Inverters," in IEEE
Zanalytic_S_dd = Zanalytic_S_qq = Zanalytic_S_dd = Zanalytic_S_qq = Transactions on Power Electronics, vol. 31, no. 1, pp. 675-687, Jan. 2016
[9] B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli and Z. Shen, "Small- [27] L. Jessen and F. W. Fuchs, "Modeling of inverter output impedance for
Signal Stability Analysis of Three-Phase AC Systems in the Presence of stability analysis in combination with measured grid impedances," 2015
Constant Power Loads Based on Measured d-q Frame Impedances," in IEEE 6th International Symposium on Power Electronics for Distributed
IEEE Transactions on Power Electronics, vol. 30, no. 10, pp. 5952-5963, Generation Systems (PEDG), Aachen, 2015, pp. 1-7
Oct. 2015 [28] Jessen, L.; Gunter, S.; Fuchs, F.W.; Gottschalk, M.; Hinrichs, H.-J.,
[10] B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli and Z. Shen, "D-Q "Measurement results and performance analysis of the grid impedance in
impedance specification for balanced three-phase AC distributed power different low voltage grids for a wide frequency band to support grid
system," 2015 IEEE Applied Power Electronics Conference and integration of renewables," in Energy Conversion Congress and
Exposition (APEC), Charlotte, NC, 2015, pp. 2757-2771 Exposition (ECCE), 2015 IEEE , vol., no., pp.1960-1967, 20-24 Sept.
[11] B. Wen, D. Boroyevich, R. Burgos, P. Mattavelli and Z. Shen, "Inverse 2015
Nyquist Stability Criterion for Grid-Tied Inverters," in IEEE Transactions [29] T. T. Do, M. Jordan, H. Langkowski, and D. Schulz, " Novel Grid
on Power Electronics, vol. 32, no. 2, pp. 1548-1556, Feb. 2017 Impedance Measurement Setups in Electrical Power Systems," Applied
[12] Middlebrook, R.D., "Input Filter Considerations in Design and Measurements for Power Systems (AMPS), 2016 IEEE International
Application of Switching Regulators," IEEE IAS Annual Meeting, 1976 Workshop on, 28-30 Sept. 2016
[13] A. Riccobono and E. Santi, "Comprehensive Review of Stability Criteria [30] Sumner, M.; Palethorpe, B.; Thomas, D.W.P.; Zanchetta, P.; Di Piazza,
for DC Power Distribution Systems," in IEEE Transactions on Industry M.C., "A technique for power supply harmonic impedance estimation
Applications, vol. 50, no. 5, pp. 3525-3535, Sept.-Oct. 2014 using a controlled voltage disturbance," Power Electronics, IEEE
Transactions on , vol.17, no.2, pp.207,215, Mar 2002
[14] B. Miao, R. Zane, and D. Maksimovic, “System Identification of Power
Converters With Digital Control Through Cross-Correlation Methods,” [31] G. Francis, R. Burgos, D. Boroyevich, F. Wang and K. Karimi, "An
IEEE Trans. Power Electron, vol. 20, no. 5, pp. 1093–1099, 2005 algorithm and implementation system for measuring impedance in the D-
Q domain," 2011 IEEE Energy Conversion Congress and Exposition,
[15] A. Barkley and E. Santi, "Online Monitoring of Network Impedances
Phoenix, AZ, 2011, pp. 3221-3228
Using Digital Network Analyzer Techniques," Applied Power Electronics
Conference and Exposition, 2009. APEC 2009. Twenty-Fourth Annual [32] L. Ljung, System Identification: Theory for the User, 2nd ed. Englewood
IEEE, Washington, DC, 2009, pp. 440-446 Cliffs, NJ: Prentice-Hall, 1999
[16] D. Martin, E. Santi, and A. Barkley, "Wide bandwidth system [33] J. Duncan Glover, Power System Analysis and Design, 5th ed. Cengage
identification of AC system impedances by applying pertubations to an Learning, 2012
existing converter," Energy Conversion Congress and Exposition [34] http://www.opal-
(ECCE), 2011 IEEE, vol., no., pp.2549,2556, 17-22 Sept. 2011 rt.com/sites/default/files/OP5600%20Simulator%20User%20Manual.pdf
[17] D. Martin, I. Nam, J. Siegers, and E. Santi, "Wide bandwidth three-phase [35] http://sine.ni.com/nips/cds/view/p/lang/en/nid/207368
impedance identification using existing power electronics inverter," [36] J. Wang, N. C. P. Chang, X. Feng and A. Monti, "Design of a Generalized
Applied Power Electronics Conference and Exposition (APEC), 2013 Control Algorithm for Parallel Inverters for Smooth Microgrid Transition
Twenty-Eighth Annual IEEE, vol., no., pp. 334,341, 17-21 March 2013 Operation," in IEEE Transactions on Industrial Electronics, vol. 62, no.
[18] A. Riccobono, S. K. A. Naqvi, A. Monti, T. Caldognetto, J. Siegers, and 8, pp. 4900-4914, Aug. 2015
E. Santi, "Online wideband identification of single-phase AC power grid [37] R. W. Erickson and D. Maksimovic, Fundamentals of power electronics,
impedances using an existing grid-tied power electronic inverter," Power 2nd ed. Norwell, MA: Kluwer Academic, 2001
Electronics for Distributed Generation Systems (PEDG), 2015 IEEE 6th
[38] Understanding FFTs and Windowing, White Paper available at
International Symposium on, vol., no., pp.1-8, 22-25 June 2015
http://www.ni.com/white-paper/4844/en/
[19] A. Riccobono, E. Liegmann, A. Monti, F. Castelli Dezza, J. Siegers and
[39] OPAL-RT: “RT-Events 4.0 – Block Library Reference Guide”.
E. Santi, "Online wideband identification of three-phase AC power grid
Available: http://www.opal-rt.com/download-center/
impedances using an existing grid-tied power electronic inverter," 2016
IEEE 17th Workshop on Control and Modeling for Power Electronics [40] Riccobono and E. Santi, "Positive Feedforward Control of Three-Phase
(COMPEL), Trondheim, 2016, pp. 1-8 Voltage Source Inverter for DC Input Bus Stabilization With
Experimental Validation," in IEEE Transactions on Industry
[20] A. Riccobono, E. Liegmann, M. Pau, F. Ponci and A. Monti, "Online
Applications, vol. 49, no. 1, pp. 168-177, Jan.-Feb. 2013
Parametric Identification of Power Impedances to Improve Stability and
Accuracy of Power Hardware-in-the-Loop Simulations," in IEEE [41] R. Burgos, D. Boroyevich, F. Wang, K. Karimi and G. Francis, "On the
Transactions on Instrumentation and Measurement, vol. 66, no. 9, pp. Ac stability of high power factor three-phase rectifiers," 2010 IEEE
2247-2257, Sept. 2017 Energy Conversion Congress and Exposition, Atlanta, GA, 2010, pp.
2047-2054
[21] T. Roinila, T. Messo and A. Aapro, "Impedance measurement of three
phase systems in DQ-domain: Applying MIMO-identification [42] B. Wen, D. Boroyevich, R. Burgos and P. Mattavelli, "c," 2014 IEEE
techniques," 2016 IEEE Energy Conversion Congress and Exposition Energy Conversion Congress and Exposition (ECCE), Pittsburgh, PA,
(ECCE), Milwaukee, WI, 2016, pp. 1-6 2014, pp. 163-169
[22] T. Roinila, M. Vilkko and J. Sun, "Broadband methods for online grid [43] https://de.mathworks.com/help/mpc/gs/using-simulink-to-develop-lti-
impedance measurement," 2013 IEEE Energy Conversion Congress and models.html
Exposition, Denver, CO, 2013, pp. 3003-3010 [44] E. C. Levy, "Complex-curve fitting," in IRE Transactions on Automatic
[23] J. Huang, K. A. Corzine and M. Belkhayat, "Small-Signal Impedance Control, vol. AC-4, no. 1, pp. 37-43, May 1959
Measurement of Power-Electronics-Based AC Power Systems Using [45] L. Harnefors, "Modeling of Three-Phase Dynamic Systems Using
Line-to-Line Current Injection," in IEEE Transactions on Power Complex Transfer Functions and Transfer Matrices," in IEEE
Electronics, vol. 24, no. 2, pp. 445-455, Feb. 2009 Transactions on Industrial Electronics, vol. 54, no. 4, pp. 2239-2248,
[24] M. Jaksic, Z. Shen, I. Cvetkovic, D. Boroyevich, R. Burgos and P. Aug. 2007.
Mattavelli, "Wide-bandwidth Identification of small-signal dq [46] J. Sun, "Small-Signal Methods for AC Distributed Power Systems–A
impedances of ac power systems via single-phase series voltage Review," in IEEE Transactions on Power Electronics, vol. 24, no. 11, pp.
injection," Power Electronics and Applications (EPE'15 ECCE-Europe), 2545-2554, Nov. 2009
2015 17th European Conference on, Geneva, 2015, pp. 1-10 [47] M. Cespedes and J. Sun, "Impedance Modeling and Analysis of Grid-
[25] Z. Staroszczyk, "A method for real-time, wide-band identification of the Connected Voltage-Source Converters," in IEEE Transactions on Power
source impedance in power systems," in IEEE Trans. on Instrumentation Electronics, vol. 29, no. 3, pp. 1254-1261, March 2014
and Measurement, vol. 54, no. 1, pp. 377-385, Feb. 2005 [48] Roinila T., Vilkko M., Sun J., Online Grid Impedance Measurement
[26] R. Stiegler, J. Meyer, P. Schegner, and D. Chakravorty, "Measurement of Using Discrete-Interval Binary Sequence Injection, IEEE Journ. of
network harmonic impedance in presence of electronic equipment," Emerg. and Select. Topics in Power Electr., Vol 2, Issue 4, pp 985-993,
Applied Measurements for Power Systems (AMPS), 2015 IEEE 2014
International Workshop on, vol., no., pp. 49-54, 23-25 Sept. 2015

Riccobono 2017 Zgrid Measurement

Uploaded by

Copyright:

Available Formats

Riccobono 2017 Zgrid Measurement

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Riccobono 2017 Zgrid Measurement

Uploaded by

Copyright:

Available Formats

This article has been accepted for publication in a future issue of this journal, but has not been

Noninvasive Online Parametric Identification of

This work extends previously presented work at COMPEL 2016.

be compared to the results of the identification in Section VI.

θ abc θ abc Ideal θ

dq dd Id_ref - Voltage Control:

Current Iq_ref Voltage

to the grid impedance labeled as Case 1, the system results in a

control to inductor current transfer function, Zout is the output

the grid impedance is given. Then, that one of the inverter

inverter output impedance are identified in dq domain. The

is implemented in such a domain as shown in Section III. This

A. Grid Impedance Identification in dq domain to PWM ia ib ic va vb vc

This subsection presents the online WSI technique exploiting θ

The impedances to be identified are the dq domain

Fitting Routine XOR

10 50 Thinned Data Points Thinned Data Points

-40 5 -40 -10

Fitted Function Fitted Function

-20 -80 -20 -80

-20 150 -20 150

-100 -200 -180 -200

Frequency (Hz) Frequency (Hz) Frequency (Hz) Frequency (Hz)

Bode Plot of Bode Plot of

-10 Analytic Function -10 Analytic Function

decoupled positive- and negative-sequence converter

found. (1.67e04*s^3 + 1.161e08*s^2 + (1.669e04*s^3 + 7.563e07*s^2 +

You might also like

found. (1.67e04s^3 + 1.161e08s^2 + (1.669e04s^3 + 7.563e07s^2 +