IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 8, NO.

2, JUNE 2020 925

Grid-Forming Inverters: A Critical Asset

for the Power Grid
Robert H. Lasseter , Life Fellow, IEEE, Zhe Chen , Student Member, IEEE,
and Dinesh Pattabiraman, Student Member, IEEE

Abstract— Increasing inverter-based sources reduces the sys- I. I NTRODUCTION

tem’s inertia resulting in possible frequency stability issues.
Understanding low-inertia systems and their stability properties
is of crucial importance. This article introduces fundamental
T HE massive penetration of variable renewable
energy (RE) in addition to other distributed energy
resources (DERs) poses major operational challenges to
ways to integrate high levels of renewable energy (RE) and
distributed energy resources (DERs) in the power system while utility system operators. Power system operation traditionally
creating a more flexible power system. Using RE and DER in the assumes that synchronous generators provide frequency
distribution system has many advantages such as reducing the stability via their stored kinetic energy. With increasing
physical and electrical distance between generation and loads, inverter-based sources, the system’s inertia is reduced
bringing sources closer to loads contributes to the enhancement
of the voltage profile, reduction in distribution and transmission and frequency stability becomes a concern. For example,
bottlenecks, improved reliability, lower losses, and enhances the operators in Ireland [1], Texas [2], and South Australia [3]
potential use of waste heat. A basic issue for high penetration are already facing obstacles regarding the high penetration
of DER is the technical complexity of controlling hundreds of of inverter-based sources during certain periods of the
thousands to millions of inverters. This is addressed through day. Understanding low-inertia systems and their stability
autonomous techniques using local measurements eliminating the
need for fast control systems. The key issues addressed in this properties is of crucial importance. We need to find
article include using inverter damping to stabilize frequency in fundamental ways to integrate high levels of RE and DER
systems with low or no inertia, autonomous operation, methods in the power system while creating a more flexible power
for relieving inverter overload, energy reserves, and their imple- system.
mentation in photovoltaics (PV) systems. This article provides Traditionally, inverter-based sources such as photo-
important insight into the interactions between inverter bases
sources and the high-power system. The distinction between grid- voltaics (PV) and wind have been deemed to possess zero
forming (GFM) inverter and grid-following (GFL) inverter is inertia; they are typically operated at their rated power output
profound. GFM inverters provide damping to frequency swings and are not expected to respond dynamically to frequency
in a mixed system, while GFL inverter can aggravate frequency changes [4]. With increasing inverter penetration levels due
problems with increased penetration. Rather than acting as a to growth in installations of variable RE sources, the total
source of inertia, the GFM inverter acts as a source of damping
to the system. On the other hand, the application of inverters in stored mechanical energy is reduced. This can result in larger
the power system has two major issues. One is the complexity frequency swings as a larger fraction of kinetic energy storage
of controlling hundreds of thousands to millions of inverters. is decommissioned. Larger deviations can cause reliability
This is addressed through autonomous techniques using local issues such as frequency-based tripping of loads and legacy
measurements. The other is the potential of high overcurrent in equipment in the system. Smaller islanded power systems,
GFM inverters and techniques for explicitly protecting against
overloading. To exploit the innate damping of GFM inverters, such as Australia, Hawaii, face imminent low inertia-related
energy reserves are critical. issues. This does not need to be the case. Inverters can be
controlled to increase frequency damping with increased pene-
Index Terms— Grid-following (GFL), grid-forming (GFM),
inverter damping, low-inertia power systems, renewables, tration. Fundamentally grid-forming (GFM) inverter frequency
reserves. control could be very advantageous, particularly for islanded
power systems with frequency issues. Compared to large
synchronous machines, inverter-based resources are able to
Manuscript received September 21, 2019; revised November 13, 2019;
accepted December 2, 2019. Date of publication December 13, 2019; date change their output much faster, arresting system’s frequency
of current version May 6, 2020. This work was supported in part by the U.S. changes before any load shedding is triggered.
Department of Energy Office of Energy Efficiency and Renewable Energy, DER and RE bring a level of variability and nuance never
in part by the Office of Electricity through contracts administered by the
Lawrence Berkeley National Laboratory, and in part by Wisconsin Electric before seen by network operators. This tangible convergence
Machines and Power Electronics Consortium (WEMPEC). Recommended of DER interconnections to networks ill-suited to integrate
for publication by Associate Editor Joseph O. Ojo. (Corresponding author: variable demand-side behaviors represents ground zero for
Robert H. Lasseter.)
The authors are with the Electrical and Computer Engineering Depart- the disruption of the global energy landscape caused by
ment, University of Wisconsin–Madison, Madison, WI 53706 USA (e-mail: DER. We need to rethink our distribution system includ-
[email protected]; [email protected]; [email protected]). ing the integration of high levels of DER, to provide a
Color versions of one or more of the figures in this article are available
online at http://ieeexplore.ieee.org. smarter and more flexible distribution system. Our overar-
Digital Object Identifier 10.1109/JESTPE.2019.2959271 ching objective is to transform the installation of a very
2168-6777 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 926 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 8, NO. 2, JUNE 2020

Fig. 1. Block diagram of a typical GFL inverter. Fig. 3. System frequency regulation with GFL and GFM inverters.

Fundamentally, if the GFL inverter loses a voltage/frequency

source it must shut down.
GFM inverters are intrinsically different from GFL inverters.
A GFM inverter is a controllable voltage source behind
a coupling reactance much like grid-tied synchronous gen-
erators. Voltage source inverters with droop characteristic
allows for direct control of voltage and frequency. During
contingencies, the droop-controlled GFM sources will increase
or decrease their output power instantaneously to balance
loads and maintain local voltage and frequency. There is no
significant delay between the change of output power and
the change of output frequency in droop-controlled GFM
Fig. 2. Block diagram of a typical GFM inverter.
inverters. Therefore, GFM sources respond much faster to any
contingencies than the response of the GFL sources. Providing
large number of inverter-based sources from a major potential primary frequency control from inverter-based resources could
liability to a critical asset for both the power grid and utility be very advantageous, particularly for “low-inertia” power sys-
customers. tems. Compared to large synchronous machines, inverter-based
Using RE and DER in the distribution system reduces resources are able to change their output much faster thus
the physical and electrical distance between generation and arresting system’s frequency changes before any load shedding
loads. The benefits include enhanced voltage profiles, reduced is triggered.
distribution and transmission bottlenecks, improved reliabil- An excellent example is the O’ahu power system in
ity, and enhanced potential use of waste heat and lower Hawaii [5]. Peak load summer case with a total load of
losses. The basic issue for high penetration of DER is the about 1080 MW was used in this example. The system has
technical complexity of controlling hundreds of thousands to 16 synchronous generators with a total output of 660 MW
millions of inverters. This complexity is greatly reduced using and transmission-connected renewable sources of 80 MW.
autonomous techniques to eliminate the need for fast control The rest of the generation fleet is comprised of distrib-
systems. The key issues addressed in this article include uted PV with a total of 360 MW. The initiating con-
inverter damping to stabilize frequency in systems with low tingency is the loss of a 200-MW synchronous generator
or no inertia, autonomous techniques for relieving inverter unit in the system. The response with PV GFL inverters
overload, use of energy reserves, and implementation of PV incorporating Frequency–Watt function is indicated in the
systems. red “FW” trace in Fig. 3. The response of GFM inverters
with droop control is indicated in the blue “Consortium for
II. GFL AND GFM I NVERTERS Electrical Reliability Technology Solutions (CERTS)” trace.
The frequency damping provided by the GFM inverters is
There are two basic control technologies for utility-based stunning.
inverters. They are grid-following (GFL) (Fig. 1) and GFM The magnitude of the imbalance between generation and
(Fig. 2). GFL inverters control the output of real and reactive load implies a sufficient level of untapped capacity or head-
power by injecting current at a given phase angle. A phase- room for frequency control. This can take the form of storage,
locked loop (PLL) is used to track the grid phase angle in real spinning reserves, and/or variable RE sources that are oper-
time. The GFL inverter cannot directly provide regulate system ated sufficiently below the maximum available power level.
voltage and frequency. Voltage and frequency reference is pro- Inverter-interfaced batteries can also be deployed. A drawback
vided externally either by a GFM inverter or the power system. is that unused capacity may impose additional operating costs.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 LASSETER et al.: GFM INVERTERS 927

TABLE I
T HEORETICAL S TUDY PARAMETERS

Fig. 4. Aggregate generator model for slow frequency dynamics studies.

III. G ENERATORS /I NVERTER DYNAMICS

A. Generator Model
As demonstrated in the earlier work [6] using small-signal
analysis, the inertia-dependent slow electromechanical mode
is responsible for the swing type response observed in power
systems. A simple mathematical formulation is presented
to satisfactorily capture this behavior in a system with all
generators.
The slow frequency dynamic response is predominantly
affected by the mechanical inertia and turbine-governor
dynamics of generators. In this timescale, the intergenerator
oscillatory modes can be ignored as they are typically faster.
A simple aggregate model can capture the frequency dynamics
by assuming that the transient frequency change is the same
at all buses [7], which is justifiable at a quasi-steady state
in this relatively slower timescale. This power system inertial Fig. 5. GFM inverter model and its inertia equivalent structure.
aggregate model can be developed by summing up generator
rotor speed differential equations. The aggregated inertia  con- dynamics. The oscillation frequency and damping values can
stant and damping values are then written as M g,a = Mg,i
be calculated as
and Dg,a = Dg,i (noteM  H /ωs ). The aggregate
1
mechanical power PM,a = PM,i is written as a function ωn = 
of frequency change PM,a (ω). The governor-turbine system Mg,a T A M p,gen
 
can be reduced to an aggregate equivalent structure. This 1 Mg,a M p,gen TB
approach is proposed in the model reduction package Dynamic ζ = √ + . (2)
2 TA Mg,a T A M p,gen
Reduction (DYNRED) to aggregate turbine-governor transfer
functions of coherent machines by summing up individual For a nominal choice of parameters given in Table I at 1-p.u.
mechanical power output responses by perturbing a speed power output, one can calculate ωn ≈ 0.6 rad/s and ζ ≈ 0.37,
input [8]. Such an approach is also used in [9] to aggregate which indicate a slow and poorly damped response. It is clear
governor turbine dynamics. from the expressions that a lower value of inertia would yield a
An equivalent linearized and the per-unitized aggregate higher oscillation frequency and smaller damping factor value
transfer function is used here for studying the frequency (hence, a faster and poorly damped response).
dynamics. In this article, a simple aggregated first-order trans-
fer function with one pole and zero and a 5% governor speed B. GFM Inverter Model
droop (M p,gen ) is used for the turbine-governor response of the Only the frequency/power dynamics of the GFM inverter
aggregated system. This results in a convenient second-order are modeled, with a first-order power measurement filter as
transfer function model, enabling the derivation of simple shown in Fig. 5 [10]. Voltage control and other dynamics are
closed-form expressions. The transfer function model for the relatively fast and can be ignored. Limit controls are ignored in
aggregate generator is shown in Fig. 4. the theoretical model. GFM inverter equations can be simply
Assuming negligible mechanical damping (Dg,a = 0), the rewritten to obtain structural resemblance to a generator unit,
characteristic equation can then be derived from the system with inertia and damping coefficients [11], where
transfer function as
    Mi  TM /M p,inv Di  1/M p,inv . (3)
1 TB 1
s2 + + s+ . (1) Typical values of these virtual terms at unit power output
TA Mg,a T A M p,gen Mg,a T A M p,gen
(=1 p.u.) are significantly different from their generator coun-
The roots of the second-order characteristic equation give the terparts (Mi = 0.16/ωs Di = 20/ωs pu), with significantly
dominant electromechanical mode that affects the frequency lower inertia and higher damping.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 928 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 8, NO. 2, JUNE 2020

C. GFL Inverter Model

The GFL inverter model is simplified for the relevant
timescales. Voltage-related dynamics are not modeled for the
theoretical study. Delays due to frequency measurement PLL,
current control loop, etc., are not included as they are relatively
fast and do not to play a role in this timescale. PLL is not Fig. 6. GFL inverter model for theoretical study.
as fast as the other faster control blocks such as current
control, and they are still slower than the power controller
modeled in this article. In general, such PLL models are
often ignored in bulk-system transient stability simulation
tools for this reason. However, they are generally used for
studying voltage stability issues which can occur under weak
grid conditions. The terminal voltage angle is used as the
feedback input to determine the frequency and determine a
power command using the FW function (droop). Consequently,
a simple first-order response to approximate the power control
response (Fig. 6). TL = 0.5 s is used as the time constant for
the power response in this article. TL is sometimes set based
on the requirements by applicable standards or can be limited
by inverter power response bandwidth.
The GFL case is structurally different from a generator or
a GFM inverter; the inputs and outputs are switched in the
GFL case. Hence, the inertia equivalent structure cannot be
formulated for the GFL case. However, the same definition Fig. 7. Mixed Source System with GFM Inverter.
for damping as in the GFM case is adopted (Di  1/M p,inv ).

D. Mixed Generator/Inverter Model Furthermore, in the two-source case, it turns out that K 1 ≈ K 4
and K 2 ≈ K 3 for a wide range of interconnecting impedance
In the mixed source case, generators and inverters are each
values. These simplifications lead to the reconstructed block
aggregated into respective equivalent units and modeled as
diagram in Fig. 8, where the inverter is represented by a
a two-source structure. Each aggregated unit is connected
linear function x Di and a first-order response with time
through a line impedance to the aggregate load modeled as
constant T . The value can be derived as T = |x Di /K 4 |,
an impedance. The linearized mixed source system model
which can be considered a delay in the feedback path due
for the GFM case is constructed as shown in [7] using the
to network connections. The value of K 4 is dependent on the
aggregate generator and inverter models. The model is derived
load and network admittances and is in the range of 0.5–1.5 at
as a function of inverter penetration level(x); the generator
varying penetration levels. In general, this network delay value
power rating reduces to (1 −x) p.u. and the inverter rating
|x Di /K 4 | < 0.05 s is insignificant at all penetration levels and
to “x” p.u. to meet a constant 1-p.u. system load. This scales
can be ignored (T = 0).
the generator inertia and damping values by (1 − x) and the
The model for the mixed source case with a GFL inverter
inverter inertia and damping terms by a factor of x. The K x
can be developed using a similar procedure. Surprisingly, sim-
coefficient terms are obtained from the small-signal power–
ple algebraic manipulations lead to the same model structure
angle relationship based on external connections between the
shown in Fig. 8, albeit the value of time constant T . The time
generator and inverter
   constant for the first-order inverter response can be derived as
Pg K1 K2 δg
= . (4) Di
Pi K3 K4 δi T = TL + x ≈ TL . (5)
K4
A similar approach can be used to develop a mixed source sys-
In this case, the time constant is affected by the added
tem model for the GFL case. As mentioned earlier, the inputs
delay due to the inverter power response which can be much
and outputs are reversed with inverter power being the output
larger than the network delay. Again, this network delay value
and the terminal inverter angle being a feedback input [11].
|x Di /K 4 | is considered insignificant at all penetration levels
and can be ignored.
E. Reduced Order Mixed Source System Models Hence, both GFM and GFL cases lead to a similar structure
The aggregate model for the GFM case shown in Fig. 7 except for the values of the T . In the mixed source GFM
can be simplified due to the fast time constants involved in case, T → 0 and the first-order delay is effectively neglected.
the inverter dynamics. Time constant Tm = Mi /Di ≈8ms Hence, the GFM inverter can be represented by a simple
is insignificant compared to the generator inertia and other linear damping feedback for the generator frequency and is an
time constants in Table I and can be ignored. The K matrix indicator for virtual damping provided by GFM inverters. The
values is a rank 1 matrix and is related by K 1 K 4 = K 2 K 3 . damping and oscillation frequency of electromechanical modes

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 LASSETER et al.: GFM INVERTERS 929

For a finite value TL , the damping term on generator frequency

is reduced along with the generator inertia.

F. System Dynamics
Variation of oscillation frequency and damping of electro-
mechanical mode with variation in inverter penetration level
is shown in Fig. 9. As the penetration level of GFL inverters
increases, an increase in damping is initially observed up to
∼20%, beyond which the damping declines. Theoretically,
the oscillation frequency, (ωn ), is small at lower penetration
levels, and damping feedback from (6) is effective. With
Fig. 8. Reduced order model of the mixed source (GFM or GFL).
increasing penetration levels, ωn increases significantly in the
GFL case and attenuates the damping feedback. From this
article, it is quite clear that such delays result in reduced
damping from inverters and contribute negatively to the system
frequency response.
The GFL inverter plot of damping versus penetration level
shown in Fig. 9(b) is sensitive to system inertia (H ) and
the power response time constant (TL ). For larger inertia
values, the peak point at which the decline in damping begins
shifts right, occurring at higher penetration levels and a higher
value of damping. For values of TL (slower power response),
the peak damping value decreases, but the penetration level at
which the decline begins remains roughly the same.
Comparisons of GFM and GFL models derived earlier
are simulated to verify the frequency dynamics. The aggre-
gated two-source system case is simulated, for a 5% load
decrease event and the generator frequency plots are shown
in Fig. 10. Simulations are repeated at inverter penetration
levels of 20%, 50%, and 80% for GFM and GFL cases.
Time-domain responses with the aggregate GFM model visibly
overlap with that of the simplified reduced model as shown
in Fig. 8 (with T = 0). Minor differences in the fast transient
are observed at 80% GFM penetration, which is relatively
insignificant and is inconsequential for the frequency dynamic
response.
It is clear that the aggregate GFM inverter provides damping
feedback on the generator frequency to the system. Rather
than acting as a source of inertia, the GFM inverter acts
as a source of damping to the system. This helps clarify
Fig. 9. Variation of oscillation frequency and damping function of inverter concerns regarding low system inertia or emulating virtual
penetration level. (a) Oscillation frequency versus penetration level. (b) Damp- inertia through inverters. Improvement in frequency dynamic
ing factor versus penetration level. performance can be directly achieved by using inverters for
damping the inertial response, rather than adding inertia to the
system. Artificially introducing delays in the form of inertia
are shown in Fig. 9, which indicates an overdamped response is unnecessary and may not be as helpful.
(ζ > 1) beyond 25% penetration. See the Appendix for the
closed form expressions. For the GFL case, time response is IV. GFM I NVERTER : C OMPLEXITY
dominated by TL and cannot be ignored as it is comparable AND OVERLOAD C ONTROL
to generator time constants. This results in attenuating the
A. Generalized Overload Control
damping feedback on the generator frequency that can be
provided by the inverter. Qualitatively, the attenuation of The application of inverters in the power system has two
damping power feedback due to GFL inverter is clear, when major issues. One is the complexity of controlling hundreds
evaluated at the value of oscillation frequency (ωn ) of the of thousands to millions of inverters. The other is the potential
electromechanical mode of high over current since GFM inverters have no direct
  control of current. The complexity issue requires designs that
1 allow each GFM inverter to operate as an autonomous energy
Deff ( j ωn ) ≈ x Di . (6)
TL2 ωn2 + 1 source. This concept is not new to the bulk-power system.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 930 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 8, NO. 2, JUNE 2020

Fig. 11. Overload mitigation controller.

Fig. 12. Demonstration of the overload issue of microsources in transient

state. (a) Power and (b) frequency for sources 1 and 2.

to compensate for the extra load. However, the sources that

have reached their power limits must restrict their overload.
This is achieved by the overloaded inverter rapidly decreasing
its output frequency changing the loading between sources
due to the change in system phase angles. If there are no
adequate reserves, relief comes from frequency load tripping.
The basic overload function is to force the overloaded source’s
frequency down faster than the droop for power levels greater
than its Pmax .
An overload mitigation controller is shown in Fig. 11.
Fig. 10. Simulation of the aggregated system response to a 5% load decrease.
The top half is the droop controls, where Mp and Mq are
the slopes of P versus f droop, and Q versus V droop,
respectively. The dotted box in Fig. 9 contains the overload
Most large generators track fast load changes autonomously mitigation controller. Pmax is the maximum power of the
using frequency droop, not signals from a centralized con- source, the error between Pmax and P is the input of the
troller. Relative to microgrids, these autonomous concepts have proportional-integral (PI) controller, where K ppmax and K ipmax
been demonstrated in the CERTS microgrid at the American are the proportional and integral gains, respectively. The output
Electrical Power’s test site [12]. of the PI controller will regulate the frequency. When the
Autonomous GFM sources can also use frequency to explic- droop-controlled source becomes overloaded, the PI controller
itly protect themselves from self-overloading [13]. A load will reduce the frequency rapidly to mitigate its overload.
increase or fault can cause a GFM source to exceed its power The basic overload function is demonstrated in a two-source
rating (Pmax ). A sustained overload can stall prime movers, microgrid (see Fig. 12). The two sources initially operate
collapse PV dc bus voltage, or cause the inverter to shut at 60 Hz with unit-1 near its maximum power P1max , and
down on overcurrent. In systems with massive penetration unit-2 well below its rating. Loss of the grid at 0. 2-s results in
of GFM inverters, we must expect that some sources will overloading of unit-1. To prevent this overload, unit-1 rapidly
reach their power limits before other sources. Step loads that decreases the frequency command in order to transfer the
result in all GFM sources instantaneously increase their output extra power to the unit-2. After the transient, the system’s

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 LASSETER et al.: GFM INVERTERS 931

new frequency is f b . Unit-1 is at its maximum value while

unit-2 has increased its output power to accommodate the load
increase. The frequency of unit-1 drops faster than unit-2,
resulting in a needed power-angle change for the loading
transfer [see Fig. 12(b)]. Overload power is transferred to
unit-2 in less than 0.2 s.
An alternative to frequency-based overload control is for the
overloaded GFM sources to switch to GFL mode. This will
force unit-2 to provide extra power. The problem with this
approach occurs when all sources are overloaded resulting
in no voltage or frequency control forcing the system to
shut down. On the contrary, GFM’s frequency-based over-
load control will continue to force-frequency down until
frequency-based shedding of noncritical loads provides relief.
At this point, the microgrid can recover with reduced loads
while maintaining stable voltage and frequency [14]. Inverter
fault protection is ultimately a manufacture’s design issue, Fig. 13. Simulated dynamic responses of the mixed-source system operating.
but there are techniques that can be used to reduce the (a) Power response for both sources. (b) Frequency at the load. (c) DC voltage
likelihood of inverter tripping on overcurrent. They include at the PV/inverter interface.
wye-delta transformers, local protection design, and the droop
controllers. Overcurrent due to power is mitigated by the the generator compensates for the load change by increasing
overload controller discussed in this section. Over-current due its output power by 5%. Since the PV source operates at the
to reactive power is reduced by the reactive power droop MPP, the steady-state frequency of the system is determined
versus voltage by reducing the output voltage with an increase by the generator’s droop characteristic, which changes from
in the reactive power [22]. 60 to 59.7 Hz in this case. However, it is found that the system
including the PV source needs more than 20 s to reach a steady
B. PV Overload Control state. The dynamic damping shown in Fig. 10 is not possible
for PV operating at MPP since there are no reserves.
Using a GFM inverter with a PV panel implies that the Pmax
shown in Fig. 11 is also the maximum power point (MPP) [15].
V. R ESERVES FOR L OW I NERTIA S YSTEMS
Setting the maximum power limit at the MPP prevents the PV
source from trying to deliver more power than is available. A. Damping Reserves
When demand exceeds the MPP the frequency decreases It is clear from Figs. 9 and 10 that aggregated GFM
rapidly. This frequency drop causes a rapid change in the inverters can provide damping feedback to the system. Rather
PV source’s phase angle that causes other sources that have than acting as a source of inertia, the GFM inverter acts
not reached their limits to automatically assume load-tracking as a source of damping to the system. Traditional power
responsibilities. The maximum power limit is designed to system relies on spinning reserve to compensate for power
respond to dynamic changes in the PV power capacity caused shortages or frequency drops within a given period of time.
by varying insolation and temperature. Traditionally, the spinning reserve is a concept for systems
The interaction between synchronous generators and PV with large synchronous generators. If the largest generator in
operating at the MPP can be illustrated using the mixed the power system is tripped, the remaining generators need to
two-source power system introduced in Table I. For 50% increase their output to recover the power shortage. However,
penetration, the expected response to a 5% load increase would leveraging traditional generation assets for creating reserved
be for both sources to equally share the increased load. When capacity creates a number of inefficiencies. For example,
the PV inverter is operating at its MPP, this sharing is not because these generators are operated below their rated values,
possible and all the load increase must be provided by the the utility is not maximizing their power output that could
synchronous generators. Details shown in Fig. 13 are the be used for baseload supply. Also, it requires the use of
power outputs from each source, the load voltage, and the PV additional fuel to ramp these generators up in the event that
inverter’s dc voltage. The power plots show not only how the their reserved generation potential is needed, which increases
transient overload of the PV source has been transferred to the emissions while reducing the net efficiency of the power
generator but also the frequency swings caused by the transient system [16].
PV overloading. With the application of the 5% load increase, In principle, GFM DER can be implemented with “spinning
both sources increase their output power instantaneously to reserve” assets. In autonomous operation, the reserve alloca-
meet this positive load step, resulting in transient overloading tion between DERs and synchronous generators is dependent
of the PV source. The Pmax controller shifts the excess load on the penetration ratio and the percent droop. Let us return
of the PV source to the generator. The PV dc bus voltage to the mixed two-source power system. In a system with
recovers to its initial value to restore the delivery of its 50% penetration and 5% droop, the autonomous response to
maximum power. In the new steady-state operating points, a 5% step load is for each source to increase its output by

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 932 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 8, NO. 2, JUNE 2020

2.5% or one half of the applied load. In a more complex

bulk power system, this distribution is also dependent on load
and source locations and the impedance distribution. In this
mixed two-source system, both the synchronous generator
and GFM inverter need to have reserves greater than 2.5%.
Communication would allow different allocations, but the
complexity could be enormous.
The bulk system could easily have hundreds of thousands
to millions of inverters throughout a large geographic area
making centralized, secondary, and tertiary control impracti-
cal without appropriate aggregate-control schemes and sys-
tem partitioning. An alternative is to operate each source
autonomously. In this case, each autonomous PV source must
have the necessary reserve capacity to ensure that it does
not exceed its maximum available power. Generally, for a
mixed-source system with n source, each inverter or generator
may have a unique penetration level, x j , with a droop of M p, j
so that the power change in each source Pi after the load
change is determined as follows:
n
xi j =1, j  =i M p, j
Pi = PL 
n n
m=1 xm j =1, j  =m M p, j
where n is the source number, the subscripts m, i , and j
represent the sources m, i , and j , respectively. The power
change constraint is
M p,1 M p,n
ω = P1 = . . . = Pn
x1 xn
n
where x m is the instantaneous source penetration, m=1
x m = 1.
For our mixed two-source power system, the distribution
Fig. 14. Simulated dynamic responses of the mixed-source system with
of the load change between the generator and inverter is reserves. (a) PV 5% droop. (b) PV 1% droop.
a function pentation and droop. For equal droops and 50%
penetration the load allocation is equal. For a 1% inverter
droop and 5% generator droop, the allocation of the load the load seen in Fig. 14(b) is far from equal. The inverter
is 5/6 for the inverter and 1/6 for the generator radically provides 5/6 of the need power for the load increase, while
changing the required reserves. For the first example, the the generator provides 1/6. The change in frequency is defined
reserves requirements are equal. For the second example, the by the 1% droop achieving steady state is less than a second.
inverter reserve need is five times that of the generator. For the This event needs to be followed with the local redispatch of the
example shown in Fig. 13, the generator’s reserves requirement generator to return the PV to a state with necessary reserves.
is the full-load increase.
VI. I MPLEMENTATION OF PV P OWER R ESERVE C ONTROL
B. GFM PV Sources With Reserves The concept of PV power reserve is to operate the PV at
The frequency dynamics shown in Fig. 13 can be improved a power level lower than the available maximum power from
by introducing power reserve margins to the PV source [17]. the PV array. The reserve power value is
This reserve is achieved by dispatching the PV power below its
P = Pmpp − Ppv,o (7)
MPP. Using the same initial conditions and load step used in
the above simulation, Fig. 14(a) presents the time response of where P is the amount of reserved power and Ppv,o is the
the system having adequate power reserve capabilities for the value of output power from the PV source. It is implemented
PV source. In this case, droop and penetration are equal. These by adjusting the power setpoint of the PV source to insure
two sources evenly share the 5% load increase. Compared needed reserved power. It also needs to point out that due
to Fig. 13, the dynamic performance is obviously improved. to the nonmonotonic characteristics of the PV array’s P − V
The frequency reduction is less than half, and the system curve, there are two possible operating points associated with
achieves steady-state in half the time. Clearly better system the desired PV power Ppv,o . The operating point with voltage
dynamic performance is achieved using PV power reserves. less than MPP can introduce overmodulation issues and insta-
In Fig. 14(b), the penetration levels remain equal but the bility for GFM PV sources. The control system must ensure
inverter droop is changed from 5% to 1%. The allocation of operation over the entire range of available needed reserved

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 LASSETER et al.: GFM INVERTERS 933

Fig. 15. Single-diode electrical equivalent circuit of the PV panel.

power at a voltage higher than the MPP. In order to implement

the power reserve concept on GFM PV sources, tracking the
time-variant maximum available power is a key requirement.
A model-based MPP estimation is proposed that avoids the
need for irradiance and temperature sensors [18] and requires
much less complexity than control-based methods [19] and
curve-fitting algorithms [20].
The equivalent circuit of the PV panel in Fig. 15 is described
by the well-known nonlinear current–voltage equation [21] as
 Vpanel+Rs Ipanel 
Vpanel + Rs Ipanel
Ipanel = Iph − Is e NBVT −1 − (8)
Rsh
where
−E g
Is = K panel AT 3 e VT (9)

is the diode’s saturation current [A], Iph is the photocurrent

source [A], N represents the number of series-connected cells,
B is the diode ideality factor, and Rs and Rsh are the series
and shunt resistances [], respectively. In (8), K panel is the
thermal coefficient [A/(m2 ·deg-K3)], A is the surface area of
a single cell [m2 ], E g is the energy gap for silicon [eV], and
VT = kT /q is the temperature equivalent voltage [V], where
k is the Boltzmann’s constant [eV/deg-K], q is the electron
charge [coulomb], and T is the temperature [deg-K]. This
equation can also be expressed as voltage in terms of current
Fig. 16. Performance of the proposed estimation method for two different
using the Lambert-W function. Without additional hardware days. (a) June 7, 2019. (b) June 17, 2019.
requirements, model-based MPP estimates can be derived by
manipulating this current–voltage equation.
Equations (9)–(11) and equations in [21] used for translation
Using PV voltage and current measurements, the PV panel
of parameters that are dependent on atmospheric conditions
equivalent circuit can be used to play the role of light irradi-
can be used to roughly estimate likely ranges of the tem-
ation and temperature sensors. First, with voltage and current
perature and irradiance values to reduce the search time.
measurements at the PV panel terminals, the operation region
Furthermore, by substituting (9) into (8) and replacing Vpanel
on the PV I − V curve can be decided. If measurements are
and Ipanel with a few sets of sampled voltage and current read-
made to the right of the MPP, the PV panel’s open-circuit
ings, the temperature and irradiance values can be estimated.
voltage Voc can be estimated to be the intercept of the
This process can be accelerated using the Newton–Raphson
curve-fitted voltage-source region of the I − V curve with the
method.
voltage axis. The short-circuit current Isc can also be estimated
Next, circuit parameters that are relevant to the temperature
using a similar process if measurements are made to the left
and irradiance, such as Iph , Is , Rsh , and VT , are updated and
of the MPP. Equation (8) can be expressed in terms of the PV
full P-V /I −V curve can be generated. Since the MPP current
panel’s open-circuit voltage Voc as
(Impp ) has a nearly proportional relationship with Isc , Impp can
 
Voc Voc be approximately expressed as
Iph = Is e NBVT − 1 + . (10)
Rsh
Impp = K Isc Isc where 0.78 < K Isc < 0.92 (12)
And the short-circuit current Isc can be estimated by solving
the following rearrangement of (8):
 RI  where the value of K I sc is essentially constant, and its value
s sc Rs can be estimated from the specified PV panel’s datasheet. The
Isc = Iph − Is e NBVT − 1 − Isc . (11)
Rsh MPP voltage (Vmpp ) can be approximately expressed by using

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 934 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN POWER ELECTRONICS, VOL. 8, NO. 2, JUNE 2020

the nonlinear Lambert-W function based on [21] GFM case can be derived from the transfer function as
⎧  √ ⎫  
1 TB x Di
⎪
⎪ +
Rs + Rsh Rs +Rs2 ⎪
⎪ s +
2
+ +
⎨ mpp
I 1 Rsh ⎬ TA Mg,a T A Rg 1 − x Mg,a
∼
Vmpp = NBVT W − Rs Impp .  
⎪ ⎪ 1 1 x
⎪
⎩
Is ⎪
⎭ + + Di (14)
Mg,a T A Rg 1−x
(13) Closed-form expressions are obtained for oscillation frequency
and damping, as a function of inverter penetration level (x)
After Impp and Vmpp are estimated using (12) and (13),
respectively, the MPP power (Pmpp ) can be calculated from 1
ωn = 
their product [23]. Mg,a T A Rg (1−x)
The performance of the MPP estimation has been evaluated  
TB √ TA x
for a PV array consisting of series- and parallel-connected ζ ≈  1−x + √ . (15)
2 Mg,a T A Rg TB 1 − x
SunPower X21-345 PV panels that are installed on the rooftop
of the Wisconsin Energy Institute at the University of Wis- With increasing values of x from 0 to 1, (1 − x)1/2 decreases
consin (UW)–Madison. Circuit parameter characterization that and 1/(1 − x)1/2 increases and leads to an increase in ωn ,
accounts for deviations from datasheets due to a variety of indicating a faster response. The increase in x/(1 − x)1/2 term
reasons such as aging, degradation, cable resistances, and so more than compensates for the decrease due to (1 − x)1/2
on, is determined using curve scan operation. term as T A  TB . Hence, the damping factor (ζ ) improves
Fig. 16 plots the estimation performance for this PV array significantly by increasing inverter penetration, and we have
operating in maximum power point tracking (MPPT) mode found that GFM inverters can damp frequency swings for
during two different days, which is compared with measured changes in load or generation.
MPPs. These figures demonstrate that the MPPs can be esti-
mated with acceptable errors by using this proposed method in
R EFERENCES
hardware PV source equipment in the presence of real-world
disturbances such as measurement noises. [1] DS3: System Services Review TSO Recommendations, EirGrid SONI,
Dublin, Ireland, May 2013.
[2] J. Matevosyan and P. Du, “Wind integration in ERCOT,” in Integration
VII. C ONCLUSION of Large-Scale Renewable Energy into Bulk Power Systems. Cham,
Switzerland: Springer, 2017, pp. 1–25.
This article provides important insight into the interactions [3] Black System South Australia, Australian Energy Market Operator,
between inverter bases sources and the bulk-power system. Melbourne, Australia, Mar. 2017.
[4] T. Ackermann, T. Prevost, V. Vittal, A. J. Roscoe, J. Matevosyan,
The distinction between GFM inverter and GFL inverter and N. Miller, “Paving the way: A future without inertia is closer
is profound. GFM inverters provides damping to frequency than you think,” IEEE Power Energy Mag., vol. 15, no. 6, pp. 61–69,
swings in a mixed system while GFL inverter can aggravate Nov./Dec. 2017.
[5] M. E. Elkhatib, W. Du, and R. H. Lasseter, “Evaluation of inverter-
frequency problems with increased penetration. Rather than based grid frequency support using frequency-watt and grid-forming
acting as a source of inertia, the GFM inverter acts as a source PV inverters,” presented at the IEEE Power Energy Soc. Gen. Meeting
of damping to the system. On the other hand, the application of (PESGM), Portland, OR, USA, 2018.
[6] D. Pattabiraman, R. H. Lasseter, and T. M. Jahns, “Comparison of grid
inverters in the power system has two major issues. One is the following and grid forming control for a high inverter penetration power
complexity of controlling hundreds of thousands to millions of system,” in Proc. IEEE Power Energy Soc. Gen. Meeting (PESGM),
inverters. This was addressed through autonomous techniques Aug. 2018, pp. 1–5.
[7] P. W. Sauer, M. A. Pai, and J. H. Chow, Power System Dynamics and
using local measurements. The other is the potential of high Stability: With Synchrophasor Measurement and Power System Toolbox.
over-current in GFM inverters and techniques for explicitly Hoboken, NJ, USA: Wiley, 2017, ch. 10.
protecting against overloading. To exploit the innate damping [8] J. H. Chow, Power System Coherency and Model Reduction. Cham,
Switzerland: Springer, 2013. Accessed: Jun. 3, 2019. [Online]. Available:
of GFM inverters energy reserves was shown to be critical. https://www.springer.com/us/book/9781461418023
In autonomous operation, the reserve allocation between GFM [9] D. Pattabiraman, J. Tan, V. Gevorgian, A. Hoke, C. Antonio, and
sources and synchronous generators is shown to be approxi- D. Arakawa, “Impact of frequency-watt control on the dynamics of a
high DER penetration power system,” in Proc. IEEE Power Energy Soc.
mately equal to the penetration ratio with equal droops. It was Gen. Meeting (PESGM), Aug. 2018, pp. 1–5.
found that unequal droops can have profound effects on the [10] P. J. Hart, R. H. Lasseter, and T. M. Jahns, “Reduced-order harmonic
needed reserves and can greatly reduce frequency swings. The modeling and analysis of droop-controlled distributed generation net-
works,” in Proc. IEEE 7th Int. Symp. Power Electron. Distrib. Gener.
last section derives and tests a reserve allocation algorithm Syst. (PEDG), Jun. 2016, pp. 1–9.
for PV sources. The test demonstrates that the MPPs can [11] S. D’Arco and J. A. Suul, “Equivalence of virtual synchronous machines
be estimated with acceptable errors by using this proposed and frequency-droops for converter-based microgrids,” IEEE Trans.
Smart Grid, vol. 5, no. 1, pp. 394–395, Jan. 2014.
method. [12] R. H. Lasseter et al., “CERTS microgrid laboratory test bed,” IEEE
Trans. Power Del., vol. 26, no. 1, pp. 325–332, Jan. 2011.
A PPENDIX [13] W. Du and R. H. Lasseter, “Overload mitigation control of droop-
controlled grid-forming sources in a microgrid,” in Proc. IEEE Power
Closed-form expressions for the mixed source GFM case Energy Soc. Gen. Meeting, Chicago, IL, USA, Jul. 2017, pp. 1–5.
[14] W. Du, R. H. Lasseter, and A. S. Khalsa, “Survivability of autonomous
can be easily obtained due to the second-order nature of the microgrid during overload events,” IEEE Trans. Smart Grid, vol. 10,
system model. The characteristic equation for the mixed source no. 4, pp. 3515–3524, Jul. 2019.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.
 LASSETER et al.: GFM INVERTERS 935

[15] W. Du, Q. Jiang, M. J. Erickson, and R. H. Lasseter, “Voltage-source Zhe Chen (Student Member, IEEE) received the
control of PV inverter in a CERTS microgrid,” IEEE Trans. Power Del., B.S. degree in electrical engineering and automation
vol. 29, no. 4, pp. 1726–1734, Aug. 2014. from the Nanjing University of Aeronautics and
[16] A. J. Wood and B. F. Wollenberg, Power Generation, Operation, and Astronautics, Nanjing, China, in 2012, and the M.S.
Control, 2nd ed. Hoboken, NJ, USA: Wiley, 1996. degree in electrical engineering from the University
[17] B.-I. Craciun, T. Kerekes, D. Sera, and R. Teodorescu, “Frequency of Wisconsin–Madison, Madison, WI, USA, in 2016,
support functions in large PV power plants with active power reserves,” where she is currently pursuing the Ph.D. degree in
IEEE J. Emerg. Sel. Topics Power Electron., vol. 2, no. 4, pp. 849–858, electrical engineering.
Dec. 2014. Her research interests include modeling, analysis,
[18] G. Vahan, “Highly accurate method for real-time active power reserve and control of renewable energy sources, microgrids,
estimation for utility-scale PV power plants,” Nat. Renew. Energy Lab., power electronics, and electric machines.
Golden, CO, USA, Tech. Rep. NREL/TP-5D00-73207, 2019.
[19] E. Batzelis, G. Kampitsis, and S. Papathanassiou, “Power reserves
control for PV systems with real-time MPP estimation via curve fitting,”
IEEE Trans. Sustain. Energy, vol. 8, no. 3, pp. 1269–1280, Jul. 2017.
[20] S. Kolesnik et al., “Solar irradiation independent expression for pho-
tovoltaic generator maximum power line,” IEEE J. Photovolt., vol. 7,
no. 5, pp. 1416–1420, Sep. 2017.
[21] W. De Soto, S. A. Klein, and W. A. Beckman, “Improvement and
validation of a model for photovoltaic array performance,” Solar Energy,
vol. 80, no. 1, pp. 78–88, 2006.
[22] J. Eto et al., “CERTS microgrid laboratory test bed: Appendix J,”
California Energy Commission, Sacramento, CA, USA, PIER Final
Rep. CEC-500-2009-004, Feb. 2009.
[23] R. Lasseter and T. Jahns, “Active power reserve control for grid-forming
PV sources in microgrids using model-based maximum power point
estimation,” in Proc. ECCE, Baltimore, MD, USA, Sep./Oct. 2019,
pp. 1–8.

Robert H. Lasseter (Life Fellow, IEEE) received

the Ph.D. degree in physics from the University of
Pennsylvania, Philadelphia, PA, USA, in 1971.
He was a Consulting Engineer with General
Electric, Philadelphia, until he joined the Depart- Dinesh Pattabiraman (Student Member, IEEE)
ment of Electrical and Computer Engineer, Uni- received the B.S. degree in electrical engineering
versity of Wisconsin–Madison, Madison, WI, USA, from the National Institute of Technology, Tiruchi-
in 1980. rappalli, India, in 2014, and the M.S. degree from
Dr. Lasseter is internationally recognized as one the University of Wisconsin–Madison, Madison, WI,
of the earliest and most influential pioneers in the USA, in 2016, where he is currently pursuing the
microgrid field. His professional career during the Ph.D. degree in electrical engineering.
past 40 years has been dedicated to applying power electronics to utility His research interests include utility applications
systems. He is the Technical Lead for the Consortium for Electrical Reliability of power electronics, inverter control, power system
Technology Solutions’ (CERTS) Microgrid Project. CERTS’ microgrid archi- dynamic response, microgrid control, and so on.
tecture is widely implemented and recognized for its plug-and-play flexibility.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on April 17,2023 at 04:11:50 UTC from IEEE Xplore. Restrictions apply.