Electronic Power Waves in Networks of Inverters

Slobodan N. Vukosavic Aleksandar M. Stankovic

Dept. of Electrical Engineering Dept. of Electrical and Computer Engineering
The University of Belgrade Tufts University
Belgrade, Serbia Medford, MA, USA
boban@etf.rs astankov@ece.tufts.edu

Abstract—The paper describes the phenomenon of electronic described and stabilized. Under such circumstances, we believe
power waves in large networks of inverters. These waves have that physics-based strategies are most promising in light of
similarities with electromechanical waves that have been possibly complicated behavior of heterogeneous components.
observed in actual electric power systems, and also have Time-horizons for stabilization that are of the order of
significant differences, including dependence on control tuning of
milliseconds will likely result in stabilization policies that
inverters and on operating point. Electronic power waves are
significantly faster than their electromechanical counterpart, but combine local and non-local (area-wide) components.
way below the speed of light. Analytical models are presented and We are convinced that the study of electronic power waves
illustrated by computer simulations. They provide a basis for is useful for understanding system-wide phenomena in large
further study of spatial phenomena in emerging inverter-rich networks of inverter-connected sources and loads. Engineering
power systems, made possible by new generations of sensors (such applications of such studies will likely follow the path blazed
as phasor measurement units) equipped with very accurate time by electromechanical dynamics, and will include stabilization
stamps. and protection improvements in the near-term and event
Index Terms—Power Electronic Converters, Power Grids, forensics and cyber-physical security in the longer-term. In
Inverters, Power system Dynamics, Power System Stability. addition, it is very likely that electromechanical and inverter-
based sources and loads will co-exist in power systems for a
I. INTRODUCTION long time to come, so there is definitely a need to understand
The study of spatial aspects of power system transients has their hybrid behavior.
a rich, fascinating history that intertwines conceptual advances II. DYNAMICS OF INDIVIDUAL GRID-SIDE CONVERTERS
with practical engineering concerns and solutions. The first
paper proposing a detailed analytical model that we are aware In grid-connected converters, a phase-locked loop is used to
of is [1]. Within a decade, authors of [2] arrived to a very similar synchronize the d-q frame to the line voltages. In balanced
model starting from practical stability concerns in the Western three-phase systems, the line voltage is represented by a
US interconnection. An influential sequence of papers started revolving vector with the amplitude and frequency that
by [3] has generalized the models and analytical methods, and correspond to the line-voltage, and with the spatial orientation
serves as a point of departure for most papers on the subject that determined by the phase of the line-voltage (ϕL in Fig. 1). With
followed. A recent two-part series [4-5] summarizes the d-q frame synchronized with the line-voltage vector, and with
theoretical aspects and explains their practical relevance in a the q-axis aligned with the line-voltage vector, the q-axis
sequence of examples. In this paper we present some current of the grid-side converter controls the active power,
preliminary modeling and control ideas for electronic power while the d-axis power controls the reactive power (Fig. 1).
waves that are most aligned with [6]. The electromechanical
waves predicted and later observed in actual power systems [1-
6] can only be quantified in practice with high-bandwidth
sensors. In the case of electronic power waves that we consider
one would need input (kHz-sampled, unprocessed) samples in
phasor measurement units (PMUs) [7,8] which have been
appropriately time-stamped [9].
Conventional models of power system transients use nodal
descriptions of components and interconnections, so spatial
aspects of transients are hard to understand. With the adoption
of grid-side inverters faster transient phenomena need to be
978-1-5386-7138-2/18/$31.00 ©2018 IEEE Fig. 1 Synchronization of the d-q frame to the line voltage [10].
 III. WAVE PROPAGATION ON A STRING OF INVERTERS restore the nonlinear model for power flows between cells. We
Consider the case shown in Fig. 5 when many inverter/load aim to demonstrate that the linearized model shown above
cells are connected in a string, and, focusing on the power already has a wealth of relevant information. Note, however,
balance for the k-th cell that no such simplifications are used in simulations presented
later. For example, following [3], we can estimate the velocity
, − , = , − , (5) of the waves predicted by (14) as
Where , , , are power generated and consumed in the
cell k, and , is the power flow between cells k and k+1.
= (15)
| |
Let b be the susceptance of the line between any two cells
Note that the velocity depends on the control gain , on the
, = sin( , − , ) (6) operating point , and on the line parameters. To get a rough
estimate, consider a (very long) string at 220kV, with line
Which is well approximated for small angular differences with
reactance at 0.4 Ω/km and the base power of 100MVA (so ZBase
, ≈ ( , − , ) (7) = 484 Ω), as in [3]. Then, to get the pre-transient power flow
of the order of 1 p.u. and the overall phase difference of 2 ,
Thus the right-hand side of (5) becomes (assuming uniform we consider the case of 60 sections, each approximately 120
voltage magnitudes along the string) km long, so that in (14) b =10 p.u. and V = 1p.u. For =
(2 , − , − , ) (8) (2 ) , so that the closed-loop bandwidth of the PLL is 1Hz
[10], and =-1 in per unit, we get the speed of ≈ 20 sections
Next, we consider an incremental version of (6), and with per second or ≈ 2400 km/s.
assumed constant power load withing the cell , we get Please note that this estimate is approximately five times
∆ = (2∆ −∆ −∆ ) (9) faster than in the case of electromechanical waves [2,3,6]. On
, , , ,
the other hand, the range of velocities is broader than in the
We note that the term in the parentheses approximates the case of electromechanical waves, as the control settings of
spatial derivative of the line voltage angle and the operating point are likely to vary widely. If the
overhead lines were to be replaced with cables at the same
∆
(10) voltage, the susceptance b would increase 3-4 times. At the
same time, which depends on both line parameters and load
From the previous section we note the (second order) characteristics, would also be affected, most likely upward.
differential equation description of dynamics for in the Note that Q0 = 0, which makes the speed estimate (15)
forward branch; for compactness, we define gains = unbounded, signals a modeling issue more than a true
and = . Then, multiplying that singularity, as it models all inverters as constant power
equation with − (reactive power generation before the sources. Given our assumption about constant power loads,
transient), we obtain that would mean that a power disturbance propagates
instantaneously due to power balance in each cell.
∆
, + , =∆ , = (11) IV. SIMULATION RESULTS
The string of grid inverters (Fig. 5) comprising a total of 64
From the block diagram in the previous section we note that
inverters is modeled and simulated. The gains within PLL units
, = , + , + , (12) of grid inverters are set to achieve the closed loop bandwidth
of 1 Hz. The initial net power increment at each node is set to
From the typical values for the constants , and from the zero, and the line frequency is set to the nominal value (50 Hz).
fact that , = , in steady state, we make the following
It is assumed that the voltages along the string retain their rated
approximation value (1 p.u.), and that the transmission line reactance between
the two inverters is 0.1 p.u. (b = 10 p.u.), as shown in Fig. 5.
∆ ∆ In Fig. 6, a total of 64 inverters are connected in series. The
≈ = (13)
input disturbance is applied to the leftmost node in the form of
Which allows us to summarize our electronic power wave in a a 500 ms wide power pulse having an amplitude of 1 p.u. The
single equation rightmost node could either be left open, or supplied by the
impedance-matching power source [6] that quenches the
+ = (14) reflection of the electronic power waves. Reactive power of
, ,
each inverter is set to Q0 = 1 p.u.
Please note that an analogous derivation can be made for two- Simulation results shown in Fig. 7 present the change of
dimensional grids as in [3]; similarly, one could model the the line-voltage phase at the connection points of
inverter power deviation for large angular deviations, and corresponding inverters in the case when the string end on the
right is left opened. It takes roughly 2.5 seconds for the wave
to reach the end of the string. With f = 50 Hz and xl Δ = 0.1 p.u, (11), where the square of the speed is reciprocal to the reactive
the distance between two inverters is Δ = 100km (so the string power of inverters. In Fig. 7, reflected electronic power waves
length is roughly L = 100×64 = 6400 km), which corresponds clearly contribute to the overall oscillatory nature of the
to the speed of v = 2560 km/s, an estimate in agreement with transient.

Fig. 6. A total of 64 inverters (GCS) are connected in series. The input disturbance is applied to the leftmost node in the form
of a 500 ms wide power pulse. The rightmost node could be left open, or, supplied by the impedance-matching power source
that quenches the reflection of the electronic power waves.

Simulation results given in Fig. 8 are obtained by adding the the speed of electronic power waves in the case of grid-side
wave-quenching compensation. The compensation is applied at inverters is considerably higher (likely 5 times or more).
the right end of the string, and it provides the power pulse that Dynamic phenomena in an ac network with grid-side converters
rests in proportion with the phase-shift deviation. Reflection at is largely affected by control settings, such as the PLL, and by
the right end of the string is suppressed, as well as the reactive loading of the inverters (which depends on actual loads
oscillations of the line-voltage phase-shift and active power. It and on transmission line parameters). The quenching of the
is also of interest to estimate the required power and energy waves requires some means of local energy storage, as we
rating of the wave-quenching device at the end of the line. To discuss below.
that aim, the power supplied from the compensator is plotted in 0.5

Fig. 9. The waveform represents the power drawn from the

wave-quenching compensator at the end of the line, and it is 0

obtained simultaneously with the results plotted in Fig. 8. The -0.5

peak power remains below 0.2 p.u., while the positive pulse
dwells for less than 2 seconds. -1

1.5
-1.5
1

-2
0.5

0 -2.5

-0.5
-3

-1
-3.5
0 5 10 15
-1.5 Time (seconds)

-2
Fig. 8. In addition to settings of Fig. 7, the impedance-
-2.5 quenching compensation is applied at the right end of the
-3
string, providing the power pulse that rests in proportion with
the phase-shift deviation.
-3.5
0 5 10 15
Time (seconds) Finally, we consider example of a power system in which
both electromechanical and electronic power waves can co-
Fig. 7. The power pulse of 1 p.u. and 500 ms is supplied
exist. We connect in a ring structure a string with grid-side
from the left. The right end of the string is left open. The
inverters (top portion of Fig. 10) with a string with conventional
waveforms represent the line-voltage phase shift at connection
electromechanical sources (bottom portion of Fig. 10). Each
points. The waveforms are intentionally scale-shifted in
string has 64 sections, with electromechanical parameters as in
vertical direction.
[6]. The initiating event is the same as in Figs. 7-8, and it occurs
From the analytical considerations and simulation results, at a junction of two strings.
we note wave phenomena can occur in networks of inverters In Fig. 10 we observe that the wave propagates much faster
just as electromechanical waves have been observed on through the top (inverter) portion than through the bottom
networks with conventional synchronous generators. While the (synchronous generator) portion. We also note numerous
speed of electromechanical waves is of the order of 500 km/s,
 REFERENCES
[1.] A. Semlyen, “Analysis of disturbance propagation in
power systems based on a homogenous dynamic
model,” IEEE Trans. Power App. Syst.,vol. PAS-93, no.
2, pp. 676–684, Mar. 1974.
[2.] R. L. Cresap and J. F. Hauer, “Emergence of a new
swing mode in the western power system,” IEEE Trans.
Power App. Syst., vol. PAS-100, no. 4, pp. 2037–2045,
Apr. 1981.
[3.] J. S. Thorp, C. E. Seyler, and A. G. Phadke,
“Electromechanical wave propagation in large electric
power systems,” IEEE Trans. Circuits Syst. I, Fundam.
Theory Appl., vol. 45, no. 6, pp. 614–622, Jun. 1998.
[4.] T. Li, G. Ledwich, Y.Mishra, J. Chow, and A.
Vahidnia, ‘‘Wave Aspect of Power System Transient
Stability-----Part I: Finite Approximation’’, IEEE Trans.
Power Systems, vol. 32, No. 4, pp. , 2493-2500, July
2017.
[5.] T. Li, G. Ledwich, Y.Mishra, J. Chow, and A.
Vahidnia, ‘‘Wave Aspect of Power System Transient
Stability-----Part II: Control Implications’’, IEEE Tran.
Power Systems, vol. 32, No. 4, pp. , 2501-2508, July
2017.
[6.] B. C. Lesieutre, E. Scholtz, and G. C. Verghese, “A
zero-reflection controller for electromechanical
disturbances in power networks,” in Proc. 14th Power
Syst. Comput. Conf., Sevilla, Spain, 2002, pp. 1–7.
[7.] IEEE Standard for Synchrophasor Measurements for
Power Systems, IEEE Power and Energy Society, IEEE
Std C37.118.1TM-2011 and IEEE Std C37.118.1aTM-
2014.
[8.] M.K. Penshanwar, M. Gavande, M.F.A.R. Satarkar,
“Phasor Measurement Unit Technology and its
Applications: A Review,” International Conference on
Energy Systems and Applications (ICESA), Nov. 2015.
[9.] A. Traub-Ens, J. Bordoy, J. Wendeberg, L.M. Reindl,
and C. Schindelhauer, “Data Fusion of Time Stamps
and Transmitted Data for Unsynchronized Beacons”,
IEEE Sensors J., vol. 15, no. 10, pp. 5946-5953, Oct.
2015.
[10.] S.N. Vukosavic, Grid Side Converters -Design and
Control, Springer, 2018.
 The use of the synchronous d-q frame in grid-side proportional and integral action of the controller, the driving
converters is often implemented via a simple and effective signal is brought to the voltage controlled oscillator (VCO).
phase-detector. In Fig. 2, the d-q frame gets synchronized to the Thus, the PLL output is
line voltage by the phase-locked loop. The position ϕPLL of the
k k k k k k 
q-axis is obtained at the output of the PLL (Fig. 3). Assuming φ PLL ( s ) =  VCO PD P + VCO 2PD i  ⋅ Δφ ( s ) (1)
that the line voltage angle ϕL goes ahead of ϕPLL by Δϕ =ϕL -  s s 
ϕPLL, the line voltage vector advances with respect of the q-axis
by the phase error Δϕ. Therefore, the d-axis voltage component
assumes a non-zero value, proportional to -sin(Δϕ) ≈ -Δϕ for
small angles. Thus, the d-axis voltage component is an indicator
of the phase error, and the phase-detection is implemented in a
straightforward way.

Fig. 4 Simplified block diagram of the PLL.

The power delivered by the grid-side inverter depends on
the voltage amplitude, current amplitude and the angle between
the voltage and current vectors,
 
PG = uL ⋅ idq ⋅ cos ( Δϕ +ψ ) = U ⋅ I ⋅ cos ( Δϕ +ψ ) . (2)

Transient changes of the line-voltage angle produce the

phase error Δϕ, which results transient changes of the inverter
active power. For small changes of Δϕ,

ΔPG = −U ⋅ I ⋅ sin (ψ ) ⋅ Δϕ = −Q0 ⋅ Δϕ , (3)

Fig. 2 The d-q frame based phase detector.
where Q0 is the reactive power delivered by the inverter.
From (1), (3) and from Fig. 4,

Q0 ⋅ s 2
ΔPG ( s ) = − ⋅ φ L ( s ) . (4)
s 2 + s ⋅ k P k PD kVCO + ki k PD kVCO
It is of interest to study the propagation of the phase-shift
and power disturbances along a sting of the grid-side inverters,
as shown in Fig. 5, where ϕL is the line-voltage phase at the
connection point of each inverter (e. g. , , for the k-th
Fig. 3 The phase-locked loop (PLL). inverter), PG is the inverter active power, and PLoad is the local
power consumption.
The phase detector of Fig. 3 is represented by the gain block
kPD in the simplified diagram of the PLL (Fig. 4). With

Fig. 5 A string of grid-side inverters.

reflections occurring at the two connection points between power Pnom cannot be maintained over the pulse dwells longer
strings. than 5-10 ms. In order to upgrade a grid-side inverter into a
wave-quenching device, it is necessary to provide the means for
0.2
considerably larger energy storage, capable of a near-unlimited
number of charging cycles. This would eliminate any form of
0.15
batteries available today. A suitable topology enhancement of
0.1
grid-side inverters that is based on super-capacitors and the
switching power interface are shown in Fig. 11.
0.05

-0.05

-0.1

-0.15
0 5 10 15

Fig. 9. The power drawn from the wave-quenching

compensator at the end of the line, obtained simultaneously
with the results plotted in Fig. 9.
The wave-quenching compensator has to supply the
transient power pulses that are illustrated in Fig. 9. With the
pulse width below 2 seconds, the required energy storage is Wes
= 2Pnom, where Pnom is the rated power of the inverter.
Conventional 3-phase inverters with capacitor-supported dc- Fig. 10. A ring structure with grid-side inverters (half-ring,
bus have a rather limited internal energy storage. With the top portion) and conventional electromechanical sources (half
allowed change of the dc-bus voltage of 20%, the transient ring, bottom portion).

Fig. 11. Topology enhancement of grid-side inverters that permits local energy storage for the wave-quenching power
compensator. The storage block comprises the super-capacitor CSC and the switching power interface.

V. CONCLUSIONS networks of inverters is that some decisions will likely remain

In this paper we argue for the existence of electronic power local. The stabilization via feedback control will have to take
waves in cases with networked inverters. Our main tool is an into account spatial and temporal properties of electronic
idealized continuum model in the form of a wave equation. We power waves. In that regard, physics will continue to provide
use simulations with discrete cells to establish credibility the much needed guidance in terms of compensation strategy
predictions of the continuum model. Such models are a and energy storage requirements.
primary means for understanding spatial aspects of disturbance Future studies on electronic power waves will likely use
propagation in power systems. In particular, they bring to the more detailed inverter and load representations and spatially
fore global (or at least non-local) facets of transient behavior. irregular connections. While we do not expect that detailed
These are hard to grasp from traditional descriptions, which inverter models will fundamentally change the main findings
capture nodal dynamics at a fixed location. Electronic power reported here, the role of different network topologies needs to
waves have similarities with electromechanical waves, but also be examined and quantified. Another intriguing direction is the
display some key differences. The fact that electronic waves study of power systems with mixed electromechanical and
are faster is not very surprising, as inverters tend to have faster electronic power waves, as illustrated in our final simulation
dynamics than traditional generators. However, electronic example.
waves also depend strongly on control settings and on the
operating point. One implication for system protection in