This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON ENERGY CONVERSION 1

A Model Modification Process for Grid-Connected

Inverters Used in Islanded Microgrids
Jacob A. Mueller, Student Member, IEEE, Md. Rasheduzzaman, Student Member, IEEE,
and Jonathan W. Kimball, Senior Member, IEEE

Abstract—A method of modifying existing grid-connected in- crogrids consisting of distributed generation sources and the
verter models for use in droop-controlled microgrids is presented. development of accurate dynamic models of their behavior.
The modification involves combination with a model of a grid- A typical microgrid control strategy for islanded operation is
forming inverter to accurately represent the coupling between
complex power, bus voltage, and frequency. The combination is the P/f , Q/V droop control method [5], [6]. In this scheme,
performed after the individual models are linearized, adding little autonomously operating inverters regulate the grid frequency
in terms of computational complexity. The method is applicable to and bus voltage based on the active and reactive power require-
any three-phase inverter operating in a grid-supporting capacity ments of the system load. These grid-forming generation units
and is scalable for any number of parallel inverters at the same share power and follow the load requirements, provided they
point of connection. To examine the modification process and its
effect on model performance, a generic grid-tied inverter model is are within the bounds of output power limitations. In a system
derived and used as a test case. The newly derived model is modi- consisting of renewable energy resources such as wind or pho-
fied according to the proposed method. The validity of this process tovoltaics (PV), this load-following operation necessitates some
is assessed through comparisons of model predictions—both from amount of energy storage to decouple the system from the inter-
before and after modification—to results of hardware experiments. mittency of the generation sources. Energy storage allows the
A simple design example is given to demonstrate the application
of this process in the design of inverters in distributed-generation- needs of the loads to be satisfied independently from the operat-
based microgrids. ing point requirements of the source. The source is then free to
vary its operating point according to some other directives. For
Index Terms—Distributed power generation, microgrids, power
system dynamics. example, energy storage allows a PV source to operate accord-
ing to the commands of a maximum power point tracker (MPPT)
regardless of the demands of the load, significantly increasing
I. INTRODUCTION the source’s efficiency. However, due to the prohibitive cost of
energy storage, particularly for distributed systems of low power
N RECENT years, a diverse array of control strategies for in-
I terfacing distributed generation sources have been proposed.
Among these solutions is the microgrid concept, which itself has
sources, it is unlikely that all sources within the system will have
this energy buffering capability. It is far more likely, and indeed
necessary, that some generating units within the system will not
a wide variety of manifestations. One of the distinguishing char-
follow the requirements of the load. Continuing the PV source
acteristics of all microgrids, though, is their ability to operate
example, so long as one source is regulating the frequency and
in islanded mode, or disconnected from the main grid [1], [2].
voltage of the system, other PV sources may follow the com-
While this defining characteristic makes the microgrid an attrac-
mands of their own MPPT. More generally, a microgrid may
tive approach for systems of distributed sources, it also presents
consist of one or more droop-controlled grid-forming units and
significant challenges to system control and stability. Perhaps
any number of grid-supporting units, or generating units that op-
the most obstructive of these challenges are those that involve
erate according to some externally generated commands [1], [7].
distributed and coordinated actions, such as black-start opera-
In [8], the term grid supporting is used to refer to inverters
tion or procedures for grid desynchronization and resynchro-
that help regulate the voltage and frequency at their local bus
nization, because they require accurate knowledge of the state
but do not operate in grid-forming operation. The classification
of the system as a whole [3], [4]. Additionally, because many
is further subdivided according to whether the inverter acts as a
distributed sources are interfaced through power-electronic con-
controlled current source or as a controlled voltage source. For
verters, the low inertia and fast dynamics allow little room for
the purposes of this study, grid supporting refers to any inverter
predictive error. These challenges motivate the study of mi-
in a microgrid that requires an externally generated ac voltage
source and cannot function in standalone operation. In the clas-
sification scheme of [8], this includes grid-feeding inverters and
Manuscript received February 25, 2015; revised July 9, 2015 and August
13, 2015; accepted August 17, 2015. This work was supported in part by the grid-supporting inverters that act as controlled current sources.
U.S. Department of Energy SunShot Initiative under Award DE0006341 (Mid- Since the objective of this study is the modification of inverter
America Regional Microgrid Education and Training (MARMET) Consortium). models originally derived with the assumption of fixed voltage
Paper no. TEC-00154-2015.
The authors are with the Department of Electrical and Computer Engineering, and frequency, the subclass of grid-supporting inverters that act
Missouri University of Science and Technology, Rolla, MO 65409 USA (e-mail: as controlled voltage sources are necessarily excluded.
[email protected]; [email protected]; [email protected]). Grid-supporting inverters fill a variety of roles in a microgrid.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. The simplest example is the PV source operating according
Digital Object Identifier 10.1109/TEC.2015.2476600 to the commands of its MPPT. More broadly, grid-supporting
0885-8969 © 2015 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 inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON ENERGY CONVERSION

inverters may include volt-VAR compensators or active power creases the complexity of the linearization step. Because of this
filters, which operate with the objective of improving power increase in computational cost, it is desirable that the individ-
quality [9], [10]. The most general of grid-supporting inverters ual grid-supporting inverters in the system be fully designed
is able to source or sink active or reactive power at an arbitrary beforehand to prevent repeated linearizations. The modeling
power factor according to commands provided by some process in this study provides an efficient means of accomplish-
external controller. This general variant may act as an agent ing this preliminary design. The key advantage of this approach
of some higher level of the controller, often referred to as the is that the grid-supporting inverter models are linearized inde-
microgrid central controller (MGCC) [11], [12]. The MGCC pendently before being included in the model of the microgrid
is responsible for coordinating the distributed functions such system. By treating the grid-supporting inverters separately, pa-
as black-start and synchronization, but acts through commands rameter changes during the design process will not require new
sent to grid-supporting inverters. The efficacy of the MGCC’s system-level equilibrium solutions to be found. This allows the
actions is heavily dependent on the dynamic behavior of the dynamic responses of grid-supporting inverters to be tuned in-
grid-supporting inverters, and it is, therefore, necessary to dividually. The cost of linearizing the models separately is that
establish accurate models of their behavior for successful a consistent steady-state operating point for the full system can-
design and implementation of microgrid systems. not be determined. However, once the grid-supporting inverters
Inverters in stiff grids and grid-supporting units in micro- have been designed appropriately, they may be included in a full
grids are conceptually the same. The differences between the microgrid model such as that given in [21] for the purposes of
two result from the microgrid’s highly variable frequency and system-level analysis.
voltage. Changes in the output power of the inverter will affect The structure of this study is as follows. First, a small-signal
frequency and voltage as a consequence of the droop control model is derived for a grid-tied inverter capable of sourcing ac-
strategy [13], while the same power changes will have no ef- tive and reactive power at an arbitrary power factor. This model
fect on a stiff grid. Accurate models of inverters connected to is intended to be similar in form and function to existing in-
stiff grids and their various control systems are well-known verter models and is derived using established methods. Next,
and established in the literature [14]–[17]. Despite these mod- a model of a droop-controlled inverter is briefly described and
els’ accuracies for inverters in stiff grids, they are not able to the process of combining these two models is introduced. The
represent dynamic behavior of the same inverter operating in results from a set of hardware experiments are then discussed,
a microgrid because of the coupling between complex power, and it is shown that the original grid-tied model is accurate for
bus voltage, and frequency. To represent this coupling effect, an inverter connected to a stiff grid but performs poorly for an
the droop equations themselves must be modeled. Linearized inverter in a microgrid. The proposed model combination pro-
models of inverters operating in droop control have been pro- cess is shown to result in a new model that is able to accurately
posed and validated [18]–[20], all of which accomplish some describe the microgrid system dynamics.
form of this representation. To achieve an accurate representa-
tion of the dynamics of the grid-supporting inverter, a model II. DERIVATION OF GENERIC GRID-CONNECTED MODEL
that encompasses both the grid-supporting inverter itself and
Accurate models of grid-tied inverters are readily available
the grid-forming unit governing the frequency and voltage at its
in the existing literature. In order to apply these (and any sim-
local bus is needed. While this sort of model could be derived
ilarly derived models) to microgrid analysis and design, it is
outright by applying the same techniques as used in the deriva-
necessary to remove the assumption of a grid with fixed voltage
tions of the previously mentioned models, the resulting system
and frequency. As a demonstration of the mechanics of this pro-
of equations would not be generalizable to grid-supporting in-
cess, a generic grid-tied inverter model is briefly introduced in
verters with different functions. Instead, this study proposes a
this section. This model will be used as a test case for the model
method of manipulating existing inverter models in such a way
modification process, which is more broadly applicable to all in-
that the coupling between complex power, bus voltage, and fre-
verter models of similar structure. While inverters in application
quency is accurately represented. This allows the functionality
are developed around a set of objectives specific to their desired
of existing inverter models to be extended to microgrid contexts
function, the design objective for this inverter is generality. As
for the purposes of system design and analysis.
a result, the hardware topology and control system used are not
The model modification process outlined in this study fits
the most optimal for any specific functional purpose, but rather
within the greater task of building scalable models of islanded
are those most commonly found in an inverter literature. Each
microgrids. In [20], a procedure for developing a model of a mi-
constituent part of the inverter system represents an avenue for
crogrid consisting of multiple grid-forming units is discussed.
further research and optimization.
The system models resulting from this process are simplified
through model order reduction in [21], and the reduced order
A. Grid-Connected Inverter Model
models are shown to be an efficient and accurate means of sim-
ulating the system-level responses to changing grid and load The generic inverter and control system is shown in Fig. 1,
conditions. However, the scalability of these models is lim- connected to a fixed grid. The topology is a three-phase voltage
ited by the linearization process, which requires the calculation source inverter typical of those used to interface distributed
of an equilibrium point for the full set of nonlinear equations generation sources. An LCL filter is used to connect the inverter
that describe the microgrid system. Including grid-supporting to the grid. Physical values of filter current, capacitor voltage,
inverters in these nonlinear equations is straightforward, but in- and output current are sensed and used by the control system,
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MUELLER et al.: MODEL MODIFICATION PROCESS FOR GRID-CONNECTED INVERTERS USED IN ISLANDED MICROGRIDS 3

Fig. 1. Generic grid-connected inverter and control system. Note that v o , il , io , and v b are three-phase quantities in stationary (abc) reference frame.

which consists of three loops acting in a synchronously rotating The current control loop also removes the cross-coupling terms
reference frame. A phase-locked loop (PLL) provides phase inherent in the reference frame transformation. Proportional-
and frequency references for the synchronous reference frame Integral (PI) controllers are used to ensure that the currents fol-
transformation. The remaining two loops are the inner loop, low the provided references. The PI controllers introduce two
which controls the inverter filter current, and the outer loop, integrator states, γd and γq . The outputs of this control loop are
which controls the active and reactive power output (P and Q, commanded d-axis and q-axis voltages, from which appropriate
respectively) and provides current references to the inner loop. duty ratios are calculated using space vector modulation (SVM).
This structure is similar to those used in [8], [14]–[16], [18], The current controller equations are
[20], and [22]–[26]. Where these inverters tend to differ is in
the outermost loop, which is typically designed for a specific γ̇d = i∗ld − ild (7)
function. In some cases, this outer loop is omitted entirely, and γ̇q = i∗lq − ilq (8)
current references are instead provided by a set of algebraic
∗
equations. vid = kpid γ̇d + kiid γd − ωn Lf ilq (9)
The LCL filters are commonly used to connect voltage source ∗
viq = kpiq γ̇q + kiiq γq + ωn Lf ild . (10)
inverters to ac grids [14], [17], [18], [20]. The equations used to
model the LCL filter are The gains of these controllers (and all PI controllers discussed
1  
in this study) are labeled kpid and kiiq , where p and i indicate
i̇ld = vid − vod − rf ild + ωg ilq (1)
Lf proportional or integral gain, respectively, and the subscript de-
1   notes the controlled variable. In this study, the averaged switch
i̇lq = viq − voq − rf ilq − ωg ild (2) modeling technique is used to approximate the output of each
Lf
switching phase leg as a continuous voltage source. This ap-
1  
proximation is viable provided that the converter’s switching
i̇od = vod − vbd − rc iod + ωg ioq (3)
Lc frequency is high compared to the dynamics of the control sys-
1   ∗
tem. Furthermore, it is assumed that vidq = vidq , neglecting
i̇oq = voq − vbq − rc ioq − ωg iod (4) semiconductor loss. The feed-forward terms in (9) and (10) re-
Lc
  move the cross-coupling resulting from the transformation from
1  
stationary to synchronous reference frame [27]. Since the objec-
v̇od = Rd i̇ld − i̇od + ild − iod + ωg voq (5)
Cf tive of these controllers is to control ild and ilq , only the cross
  1   coupling due to Lf needs to be considered.
v̇oq = Rd i̇lq − i̇oq + ilq − ioq + ωg vod . (6) The outer control loop consists of instantaneous power cal-
Cf
culations, first-order low-pass filters, and PI controllers. Instan-
These equations include the series resistances of the capac- taneous active and reactive power (p and q, respectively) are
itors and inductors in the filter. To distinguish the inductors, calculated from the capacitor voltage and coupling inductor
the inverter-side inductor is referred to as the filter inductor, current. The low-pass filter, whose cutoff frequency ωc is much
whereas the grid side is referred to as the coupling inductor. The lower than the nominal line frequency ωn , is used to provide
subscripts of their respective labels, Lf and Lc , reflect this des- the average power. The PI controllers ensure that the inverter
ignation. The capacitor’s resistance is lumped together with the follows references P ∗ and Q∗ by providing current commands
larger damping resistance Rd . The resistance of the coupling i∗ld and i∗lq to the inner control loop. Like the current controllers,
inductor may represent that of a discrete component or of an these power controllers introduce integrator states φP and φQ .
isolating transformer, as in [16]. All magnetics are assumed to The power controller equations are
be linear. In (3) and (4), vbd and vbq represent the d- and q-axis
voltages of the bus to which the inverter is connected. 3 
p= vod iod + voq ioq (11)
The current control loop is responsible for ensuring that the 2
sensed filter inductor currents ild and ilq follow current refer- 3 
ences i∗ld and i∗lq , which are provided by the power control loop. q= voq iod − vod ioq (12)
2
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON ENERGY CONVERSION

 
Ṗavg = ωc p − Pavg (13) The full nonlinear grid-tied model consists of 14 equations.
The model is in the form
 
Q̇avg = ωc q − Qavg (14) ẋ = F (x, u) (24)
 T
φ̇P = P ∗ − Pavg (15) u = P ∗ Q∗ vB D vB Q ωg (25)

φ̇Q = Q∗ − Qavg (16) x = δ Pavg Qavg φP φQ γd γq
T
i∗ld = kpQ φQ + kiQ φ̇Q (17) vod voq ild ilq iod ioq φPLL (26)
i∗lq = kpP φP + kiP φ̇P . (18) where x and u are vectors of system states and inputs, respec-
tively, and F is a vector of (1)–(8), (13)–(16), (19), and (22).
The PLL conforms to a standard PI structure typical Inputs P ∗ and Q∗ are provided externally while inputs vB D ,
of synchronous-reference-frame-based inverter control sys- vB Q , and ωg are fixed, set by the grid itself. For the grid-tied
tems [8], [14]–[16], [18], [20], [26]. The d-axis component of system, setting these constant values as inputs is unnecessary,
the capacitor voltage is regulated to 0 to track the grid frequency, but this will later allow the necessary modification for use in
ωg . The phase angle calculated by the PLL, θPLL , is then used microgrid applications. Note that the bus voltage inputs are in
in the reference frame transformation. In this study, the Park global reference frame, and that the model is responsible for
Transformation as given in [27] is used. The PI controller of the transforming these voltages into its own local reference frame.
PLL contains an integrator state, φPLL . The equations pertaining Selecting appropriate inputs and constants, the model can be
to the PLL are linearized by calculating the steady-state values and perturbing
φ̇PLL = −vod (19) the system around this steady-state operating point. This is done
  by taking the Taylor series expansion of the system equations,
ωPLL = kpPLL φ̇PLL + kiPLL φPLL (20) ignoring higher order terms. The resulting small-signal model
is in the traditional state-space form
θPLL = ω̇P L L (21)
x̃˙ = Ax̃ + B ũ. (27)
δ̇ = ωg − ωPLL . (22)
This model is an accurate representation of the inverter system
B. Reference Frame Transformation and System Model connected to a stiff grid and is, in that capacity, a useful design
and analysis tool.
In modeling the inverter, it is necessary to represent the differ-
ence between the actual grid phase angle and the reference angle
III. MODEL MODIFICATION PROCEDURE
calculated by the PLL. This difference in angles, δ, is calculated
according to (22). Angle δ is used in a rotational transformation In order for the previously described grid-connected in-
that translates the d-axis and q-axis quantities from the global verter model to accurately represent a similarly structured grid-
reference frame to the local reference frame, which is set by supporting inverter in a droop-controlled microgrid, the model
θPLL . The rotation transformation used is commonly applied to itself must reflect the coupling between complex power, bus volt-
modeling inverters and synchronous machines [18], [28], [29]. age, and frequency. Changes in the grid-supporting inverter’s
The transformation is output power will elicit variations in frequency and bus voltage
  due to the droop control rules instituted by the grid-forming
lo cal
cos δ − sin δ inverters of the system. These variations are governed by the
Tglobal = (23)
sin δ cos δ algebraic droop control equations and the dynamics of the grid-
forming inverters’ controllers. As a first step in introducing a
where the subscript and superscript denote a transformation representation of the mechanics behind these variations into
from global reference frame to local reference frame. In the the grid-supporting inverter model, a small-signal model of a
case of a grid-tied inverter model, the global reference frame is grid-forming inverter must be explored.
set by the angle of the grid. In microgrids, the global reference
frame may be defined by any of the grid-forming units in the A. Grid-Forming Inverter Model
system. In either situation, the modeling equations are only valid
when all terms are in the same reference frame. Equations (3) An accurate small-signal model of a grid-forming inverter is
and (4) in the preceding section include bus voltage terms vbd given in [20]. This model includes the traditional droop control
and vbq , which are not directly sensed by the inverter. The corre- equations
sponding global reference frame quantities, vB D and vB Q , must ω ∗ = ωn − nPavg (28)
be transformed using (23) to the inverter’s local reference frame.
∗
The currents at this point of connection, iod and ioq , are referred voq = voq , n − mQavg (29)
to the global reference frame as part of the model modification
φ̇d = ωPLL − ω ∗ (30)
process. Their corresponding global reference frame quantities
∗
are iO D and iO Q . φ̇q = voq − voq (31)
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MUELLER et al.: MODEL MODIFICATION PROCESS FOR GRID-CONNECTED INVERTERS USED IN ISLANDED MICROGRIDS 5

Fig. 2. Grid-supporting inverter connected to a microgrid bus regulated by a grid-forming inverter. Currents and voltages shown are those used in model
combination equations, and are all in synchronous reference frame. Scope of grid-supporting inverter’s local reference frame is indicated by a dashed line.

where m and n are the droop constants. The control system bus may be included as
of the grid-forming inverter is identical to the grid-connected  
inverter in all areas except for the outer control loop; the grid- vB D = rn ifrmOD + id, line + id, ext − id, load (34)
forming control system is autonomous and requires no external  
references. Instead, PI controllers ensure that the voltage and vB Q = rn ifrmO Q + iq , line + iq , ext − iq , load (35)
frequency of the system follow references calculated using the
droop equations. These PI controllers contain integrator states where id, ext and iq , ext are the sum of currents from any grid-
φd and φq . supporting inverters connected to the bus. Fig. 2 shows the
Because the frequency and voltage references of this system currents involved in these equations for the case of a single
are set by the total output power, the model must contain some grid-supporting inverter. In this case, currents id, ext and iq , ext
way of representing the power contribution of the inverter in are equal to the output currents of the grid-supporting inverter
the context of the system as a whole. The inverter’s local bus referred to the global reference frame. These equations allow
voltage is calculated through the virtual resistor method [18], the model to be expressed as a set of nonlinear functions of state
[20], allowing the currents flowing to the loads and the rest of variables xfrm and input vector ufrm , consisting of the newly
the microgrid system to be modeled. This is shown in Fig. 2. added external currents id, ext and iq , ext . Again subscript frm is
Considering only the contribution of the grid-forming inverter, used to indicate that these vectors pertain to the grid-forming
the bus voltage equations are inverter.
This nonlinear model may be linearized in the same way as the
  grid-connected inverter model. During linearization, the inputs
O D + id, line − id, load
vB D = rn ifrm (32) are set to 0, such that the entirety of the load is supplied by the
  grid-forming inverter. Since the quantities relevant to the grid
O Q + iq , line − iq , load
vB Q = rn ifrm (33) variations are frequency and bus voltages, these are selected as
model outputs. The rotation transformation in (23) is applied to
refer these quantities to the global reference frame. The resulting
where rn is the virtual resistance, and ifrm frm
O D and iO Q are the small-signal model is
output currents of the grid-forming inverter. Superscript frm
is used to distinguish these currents from those of the grid- x̃˙ frm = Afrm x̃frm + Bfrm ũfrm (36)
supporting inverter, which are similarly named. All quantities
in (32) and (33) are in the global reference frame. By setting this ỹfrm = Cfrm x̃frm + Dfrm ũfrm . (37)
resistance to a large value, the sums of currents are driven close
to zero, which approximates Kirchhoff’s current law and retains Returning to the grid-connected model, the currents iO D and
all of the voltage and current variables needed in the model. iO Q may be chosen as system outputs. As mentioned previously,
The effect of the virtual resistor on the model’s dynamics is these quantities are the output currents referred to the global
negligible. As shown in [20], the virtual resistor contributes to a reference frame. The full model is expressed as
pair of eigenvalues located much further into the left-half plane
x̃˙ sup = Asup x̃sup + Bsup ũsup (38)
than any other system eigenvalues. For a value of rn = 104 Ω,
the negative real component of these eigenvalues is four orders ỹsup = Csup x̃sup + Dsup ũsup . (39)
of magnitude greater than the next fastest group. In the same
way that power flowing to and from the rest of the microgrid For clarity, subscripts sup in the aforementioned equations indi-
is represented by the currents at this point of connection, the cate that this model describes the grid-supporting unit, though
currents flowing to and from other inverters connected at this the equations themselves are the same as in (27).
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON ENERGY CONVERSION

B. Combination of Inverter Models TABLE I

STEADY-STATE SOLUTION SOLVER TIMES IN SECONDS
The grid-supporting and grid-forming models separately de-
scribe the inverters shown in Fig. 2. The models are derived and PC System ODE15s ODE23s ODE23t
linearized separately, but can be combined at this stage to form
a model capable of accurately predicting the dynamics of the A Grid Forming 0.3042 6.2736 0.3285
Grid Supporting 0.3208 9.1452 0.3325
grid-supporting inverter in a way that the original model could Combined 100.16 127.77 29.965
not. For this new model, the assumption of fixed grid voltage B Grid Forming 0.6500 15.704 0.6198
and frequency has been replaced with an accurate description Grid Supporting 0.5379 12.901 0.5261
Combined 165.19 244.37 0.9189
of the variations in these parameters as a response to changes in C Grid-Forming 1.4276 9.4921 0.6114
commanded inputs P ∗ and Q∗ . The grid-supporting unit’s small- Grid Supporting 0.5158 11.578 0.5337
signal inputs ṽB D , ṽB Q , and ω̃g are supplied by the outputs of Combined 42.076 95.284 7.0901

the grid-forming unit. In the same way, the inputs to the grid-
forming unit, ĩd, ext and ĩq , ext are supplied by the outputs of the
grid-supporting unit. This allows bus voltage and frequency to is much greater for the combined nonlinear system than for the
be calculated as linear combinations of states, rather than as in- sum of the times needed for the individual systems. Further-
puts. The remaining inputs are the exogenous command inputs. more, applied changes to the grid-supporting system commonly
The combination is accomplished using matrix operations necessary as part of a design process, e.g., changes to controller
  gains or physical parameter values, require the equilibrium so-
x̃sup lution to be recalculated. When combining the models after the
x̃full = (40)
x̃frm linearization step, only the linearization of the grid-supporting
    unit must be repeated, whereas when the system is derived as a
ũsup ũext whole, the entire set of equations must be solved again.
ũall = = (41)
ũfrm ũint The second advantage of the combination-based approach
  is scalability. The inputs to the grid-forming inverter may be
Bsup 0   defined to include not only the currents from a single grid-
Ball = = Bext Bint (42) supporting inverter, but any number of grid-supporting con-
0 Bfrm
  nected in parallel at the local bus. Moreover, changes made to
Dsup = Dext Dint (43) one grid-supporting inverter require only that a new operating
  point be found for that inverter, so as the number of inverters
Asup 0
Afull = (44) in the system increases so too do the benefits of this modeling
0 Afrm approach.
  Because the scope the combined model is limited to the in-
Dfrm Csup Cfrm
+ Bint (45) verters and loads present at a single point of connection, few
Csup Dint Cfrm assumptions are made on the topology and structure of the mi-
crogrid as a whole. The model combination process requires that
x̃˙ full = Afull x̃full + Bext ũext. (46) the microgrid is operating in islanded mode according to P/f
and Q/V droop equations. As a result, the model inherits some
Equation (45) is the final model of the two inverters.
assumptions made in the derivation of the grid-forming inverter.
This process of combination results in a new model that is
Specifically, the system is assumed to be stable and balanced,
much larger than the original grid-connected model. One may
and local loads are assumed to be linear. The structure of the
wonder about the practicality of this approach when a very sim-
system at the local bus is accounted for in (34) and (35), and
ilar model could simply be derived from the representation of
any additional loads or parallel grid-supporting inverters may
the system in Fig. 2. This is done in [21], with the objective
be represented in the d-axis and q-axis currents at this point of
of forming a full model, and then, reducing the model order
connection. Beyond the local bus, a procedure for modeling the
to simplify system-level simulations. The advantages of the
interactions between autonomously acting grid-forming invert-
present method are its efficiency and scalability. Because the
ers is required, such as that given in [20] and [21]. At this level,
two models are linearized before being combined, the lineariza-
the microgrid structure must be taken into account.
tions are much simpler in a computational sense. To determine
the equilibrium point around which the systems are perturbed,
the nonlinear equations must be solved. This is the most re- IV. VALIDATION OF COMBINED MODEL
source intensive step of the modeling process. Table I shows The objective of this approach to modeling a grid-supporting
the calculation time required to determine the equilibrium so- inverter is to accurately predict the dynamic response of the in-
lutions for the grid-forming and grid-supporting inverter sys- verter during changes in the commanded inputs. To demonstrate
tems, as well as an equivalent nonlinear system derived without the accuracy of the full model, its predictions were compared
the combination-based approach. The time is shown for three to the results of hardware experiments. The hardware results
MATLAB differential equation solvers executed on three PCs were gathered from a microgrid testbed consisting of two invert-
of different performance calibers. In all cases, the time required ers designed around Infineon BSM30GP60 IGBT modules and
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MUELLER et al.: MODEL MODIFICATION PROCESS FOR GRID-CONNECTED INVERTERS USED IN ISLANDED MICROGRIDS 7

grid frequency, and voq slightly increases due to the voltage drop
across the coupling inductor and its parasitic resistance.
Similar graphs are shown for the grid-supporting system con-
nected to a droop-controlled grid-forming unit in Figs. 6 and
7. The predictions of the model clearly match the response ob-
served in hardware. The frequency, which naturally increases as
a result of the P/f droop equation, matches both in its transient
response and in its final steady-state value. These results show
that the combined model preserves the accuracy of the grid-
connected model while also representing the variations in the
bus voltage and frequency of the microgrid.

V. APPLICATION
Fig. 3. Picture of partial experimental setup including inverters, control
boards, and filter hardware. As an example of the practical application of this approach,
consider the challenge of tuning the response of an inverter such
as the one described in this study for use in an existing micro-
grid system. To ensure the stability of the system as a whole,
controlled by TMS320F28335 DSPs. The experimental setup is the controller gains of the newly connected inverter must be set
shown in Fig. 3. appropriately. Specifically, the control system must be designed
The validation tests include step changes to the inputs P ∗ and such that it does not contribute to oscillations in bus voltage and
∗
Q for both the original grid-connected inverter model and the frequency. The reasons behind this requirement are illustrated
combined grid-supporting inverter model. The conditions at the by an eigenvalue analysis of a system of multiple grid-forming
start of each test are as follows. For grid-connected inverter tests, inverters, as given in [13], [20]. The low-frequency modes of
a single inverter is synchronized to the grid and regulates power the multiple inverter system are shown to be primarily related to
flow of 0 W and 0 VAR. When first activated, the inverter sys- the droop controller and current controller states. A participa-
tem includes series-connected current limiting resistors, which tion factor analysis reveals that the controllers’ integrator states
are then manually bypassed when the inverter reaches steady participate strongly in oscillating modes. Of the low-frequency
state. For grid-supporting tests, two inverters are synchronized modes, one relating to the d-axis current controller and P/f
to the grid in parallel. This is done in the same way as the grid- droop controller is the least damped, and is consequently labeled
connected tests. When both inverters reach steady state, the grid a problem mode. In the grid-forming inverter control system, the
is disconnected from the system and one inverter is allowed to d-axis current controller provides the commanded d-axis volt-
transition into grid-forming operation. The inverter still oper- age to the SVM. The PLL regulates the measured d-axis voltage
ating according to the grid-supporting control system serves as of the output filter to 0. In this way, oscillations in the d-axis
the inverter under test for the hardware experiment, while the current controller cause oscillations in the frequency calculated
grid-forming inverter autonomously regulates bus voltage and by the PLL. Since the grid-forming unit sets the frequency in the
frequency. The nominal grid frequency used was 60 Hz, and the microgrid, the frequency oscillations carry through the system,
nominal q-axis voltage was 85 V. The inductance of the filter affecting the impedance of reactive loads. This, in turn, causes
inductors used was 4.2 mH, filter capacitors were 15 μF, and variations in the reactive power output, which influences the
coupling inductors were 0.5 mH. Inverter switching frequencies bus voltage through the Q/V droop controller. In systems of
were 5 kHz. distributed generation sources, which are commonly connected
The response of the grid-connected system to step changes through low-voltage networks in which line impedance can be
in P ∗ and Q∗ are shown in Figs. 4 and 5, respectively. The primarily resistive, additional coupling exists between active
step changes occur at t = 1.5 s. These graphs show P , Q, vod , and reactive power, further contributing to these potential insta-
voq , and ωPLL . Similarly matching results can be shown for all bilities [18]. This problem mode represents a strong potential for
system states, but have been omitted due to space constraints. instabilities in a droop-controlled microgrid, and it is, therefore,
These results of these tests are not novel: it is well known that critical that the grid-supporting inverter be designed such that it
a linearized model is able to describe the dynamics of a grid- does not decrease the damping ratios of eigenvalues associated
connected inverter. The results are presented here to demonstrate with the grid-forming unit’s current controllers.
the accuracy of the derived grid-connected model in its precom- For the purposes of this example, we will assume a require-
bined form. Quantities shown are as recorded by the inverters. ment that the damping ratios of low-frequency eigenvalues be
Because the frequency is calculated through the PLL, which greater than 30%. This requirement is simply chosen to illus-
provides a reference angle used in the calculation of d-axis and trate the difference in performance resulting from the use of
q-axis quantities, the transients observed in grid voltage magni- each model. In order to meet the damping requirements, a sys-
tude voq and frequency ωPLL are related to the PLL’s dynamics. tem designer may choose between two options. The simplest
Once the initial transient has subsided, ωPLL returns to the orig- option is to tune the controller gains using the grid-connected
inal steady-state value, consistent with the assumption of fixed model such that the eigenvalues have reached locations in which
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON ENERGY CONVERSION

Fig. 4. Comparison of model and experimental P ∗ step response dynamics in grid-connected operation. All x-axis values given in seconds. (a) Active Power.
(b) Reactive Power. (c) Quadrature Axis Voltage. (d) Direct Axis Voltage. (e) Frequency.

Fig. 5. Comparison of model and experimental Q ∗ step response dynamics in grid-connected operation. All x-axis values given in seconds. (a) Active Power.
(b) Reactive Power. (c) Quadrature Axis Voltage. (d) Direct Axis Voltage. (e) Frequency.

Fig. 6. Comparison of model and experimental P ∗ step response dynamics in grid-supporting operation. All x-axis values given in seconds (a) Active Power.
(b) Reactive Power. (c) Quadrature Axis Voltage. (d) Direct Axis Voltage. (e) Frequency.

Fig. 7. Comparison of model and experimental Q ∗ step response dynamics in grid-supporting operation. All x-axis values given in seconds (a) Active Power.
(b) Reactive Power. (c) Quadrature Axis Voltage. (d) Direct Axis Voltage. (e) Frequency.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MUELLER et al.: MODEL MODIFICATION PROCESS FOR GRID-CONNECTED INVERTERS USED IN ISLANDED MICROGRIDS 9

TABLE II
EIGENVALUES (λ) AND DAMPING RATIOS (ζ) OF GRID-TIED AND
GRID-FORMING SYSTEMS BEFORE AND AFTER
MODEL COMBINATION

Grid-Tied (T) / Grid-Forming (F) Combined System

ID λ ζ (%) λ ζ (%)
T1 −93.591 ± j 141.06 55.3 −22.567 ± j 37.290 51.8
T2 −124.47 ± j 88.243 81.6 −35.342 ± j 10.168 96.1
T3 −10.124 ± j 8.4065 76.9 −10.909 ± j 7.9483 80.8
F1 −6.3118 ± j 15.743 37.2 0.2623 ± j 9.9210 −2.64
F2 −25.731 ± j 13.879 88.0 −3.5997 ± j 1.3249 93.8

Fig. 8. Comparison of P ∗ and Q ∗ step response, as measured in the hardware

testbed, with gains tuned using the grid-connected and combined models. All
x-axis values given in seconds. (a) Active Power. (b) Reactive Power.

the requirements are satisfied. However, since the assumption

of fixed grid voltage does not hold in a microgrid, the perfor-
mance predicted by the model is not guaranteed. Alternatively,
the combined grid-supporting model could be used to tune the
gains. Let gain sets “A” and “B” be the gains resulting from these
two methods. In both cases, the outer-loop controller gains are
selected based on the desired bandwidth of the full system, and
the gains of the inner-loop controllers are gradually increased
until the damping ratio requirement has been met.
A comparison of the active and reactive power responses
of the inverter for each set of gains is shown in Fig. 8. In this
experiment, commanded values P ∗ and Q∗ were simultaneously
changed. Although the same tuning process was applied to each
of the models, the responses are dissimilar. The experimental
Fig. 9. Comparison of simulated P ∗ step response to the responses predicted
response using the gains tuned on the grid-connected model by the grid-connected and combined models. Controllers in simulation and
(Set A) is underdamped, while the gains tuned on the combined models all use gain set C for this test. All x-axis values given in seconds.
model (Set B) result in an acceptably damped response. (a) Active Power. (b) Reactive Power. (c) Frequency. (d) Quadrature Axis
Voltage.
The error in the grid-connected model’s predictions is a re-
sult of the interactions between the inverter control systems,
which are beyond the scope of the individual models. This in- inverters are not represented by these modes. The combined
teractive behavior may have more severe consequences than model provides access to this information. The eigenvalues and
underdamped responses. Consider another set of gains, Set C, damping ratios of the combined model are given in the right-
designed using the grid-connected model in the same way as most columns of Table II. Using gain set C, the damping of
Set A. The low-frequency eigenvalues of the original grid-tied low-frequency modes has significantly decreased and the sys-
model using gain set C are given in Table II, along with their tem has become unstable. A participation factor analysis of the
associated damping ratios. These eigenvalue pairs are listed as combined system shows that the unstable mode is participated in
T1, T2, and T3. In the same way, the low-frequency eigenvalues heavily by the current controller states of both the grid-forming
of the grid-forming inverter model are found and given as F1 and grid-supporting inverters, as expected given the character-
and F2. Table II shows that the damping ratios of these low- istic sensitivities of the droop-controlled microgrid system.
frequency eigenvalues are all greater than 30%, consistent with A P ∗ step response of a grid-supporting inverter using gain
the design requirements. However, the interactions between the set C is shown in Fig. 9, along with the predictions of the
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON ENERGY CONVERSION

grid-connected and combined grid-supporting models whose [8] J. Rocabert, A. Luna, F. Blaabjerg, and P. Rodriguez, “Control of power
eigenvalues are shown in Table II. Due to the known system converters in ac microgrids,” IEEE Trans. Power Electron., vol. 27,
no. 11, pp. 4734–4749, Nov. 2012.
instability, this response was simulated rather than measured in [9] E. Planas, A. Gil-de Muro, J. Andreu, I. Kortabarria, and I. M. de Alegria,
the hardware testbed. The simulation was carried out in MAT- “Design and implementation of a droop control in d-q frame for islanded
LAB/Simulink using the PLECS blockset, and was designed to microgrids,” IET Renew. Power Gener., vol. 7, no. 5, pp. 458–474, Sep.
2013.
be as accurate as possible, incorporating inverter switching, non- [10] J. Guerrero, P. C. Loh, T.-L. Lee, and M. Chandorkar, “Advanced con-
linearities resulting from switching dead-time, and other non- trol architectures for intelligent microgrids part ii: Power quality, energy
ideal characteristics present in the hardware system. The plots in storage, and ac/dc microgrids,” IEEE Trans. Ind. Electron., vol. 60, no. 4,
pp. 1263–1270, Apr. 2013.
Fig. 9 show that the instability and oscillatory behavior of the mi- [11] A. Tsikalakis and N. Hatziargyriou, “Centralized control for optimiz-
crogrid system is well described by the grid-supporting model, ing microgrids operation,” IEEE Trans. Energy Convers., vol. 23, no. 1,
but is totally absent from the predictions of the grid-tied model. pp. 241–248, Mar. 2008.
[12] M. Rasheduzzaman, S. Bhaskara, and B. Chowdhury, “Implementation of
This result provides the clearest representation of the shortcom- a microgrid central controller in a laboratory microgrid network,” in Proc.
ings of inverter models derived with the assumption of fixed grid North Amer. Power Symp., Sep. 2012, pp. 1–6.
parameters. Connecting a grid-supporting inverter to an other- [13] J. Mueller, “Small-signal modeling of grid-supporting inverters in droop
controlled microgrids,” Master’s thesis, Elect. Comput. Eng., Missouri
wise stable microgrid bus may cause the whole microgrid to Univ. Sci. Techn., Rolla, MO, USA, 2014.
become unstable, even if the grid-supporting inverter is “well de- [14] N. Bottrell, M. Prodanovic, and T. Green, “Analysed small signal state-
signed” according to a grid-connected model. The model combi- space model of an active rectifier,” in Proc. 45th Int. Univ. Power Eng.
Conf., Aug. 2010, pp. 1–6.
nation approach outlined in this study provides a representation [15] N. Kroutikova, C. Hernandez-Aramburo, and T. Green, “State-space
of the interactions that occur between the inverters without re- model of grid-connected inverters under current control mode,” IET Elect.
quiring a full rederivation of the system’s governing equations. Power Appl., vol. 1, no. 3, pp. 329–338, May 2007.
[16] M. Prodanovic and T. Green, “Control and filter design of three-phase
inverters for high power quality grid connection,” IEEE Trans. Power
Electron., vol. 18, no. 1, pp. 373–380, Jan. 2003.
VI. CONCLUSION [17] E. Twining and D. Holmes, “Grid current regulation of a three-phase
The coupling between complex power, bus voltage, and fre- voltage source inverter with an LCL input filter,” IEEE Trans. Power
Electron., vol. 18, no. 3, pp. 888–895, May 2003.
quency that occurs in a droop-controlled microgrid is significant [18] N. Pogaku, M. Prodanovic, and T. Green, “Modeling, analysis and
enough that new modeling techniques are required to correctly testing of autonomous operation of an inverter-based microgrid,”
predict the responses of microgrid inverters. In this study, a IEEE Trans. Power Electron., vol. 22, no. 2, pp. 613–625, Mar.
2007.
method of modifying an existing small-signal model for use in [19] Y.-R. Mohamed and E. El-Saadany, “Adaptive decentralized droop con-
microgrid contexts has been proposed. The method of modifica- troller to preserve power sharing stability of paralleled inverters in dis-
tion involves combining the existing model with an autonomous tributed generation microgrids,” IEEE Trans. Power Electron., vol. 23,
no. 6, pp. 2806–2816, Nov. 2008.
grid-forming inverter model. The combined model preserves the [20] M. Rasheduzzaman, J. Mueller, and J. Kimball, “An accurate small-
functionality of the original small-signal model but is also ac- signal model of inverter-dominated islanded microgrids using dq refer-
curate in microgrids, where the bus voltage and frequency vary. ence frame,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 2, no. 4,
pp. 1070–1080, Dec. 2014.
The procedure is highly scalable and can be extended to systems [21] M. Rasheduzzaman, J. Mueller, and J. Kimball, “Reduced-order small-
of multiple inverters, facilitating a modular process of analysis signal model of microgrid systems,” IEEE Trans. Sustain. Energy,
and design of full islanded microgrids. doi: 10.1109/TSTE.2015.2433177, in press.
[22] F. Katiraei, M. Iravani, and P. Lehn, “Small-signal dynamic model of a
micro-grid including conventional and electronically interfaced distributed
resources,” IET Gener., Transm. Distrib., vol. 1, no. 3, pp. 369–378, May
REFERENCES 2007.
[1] T. Vandoorn, J. Vasquez, J. De Kooning, J. Guerrero, and L. Vandevelde, [23] X. Tang, W. Deng, and Z. Qi, “Investigation of the dynamic stability of
“Microgrids: Hierarchical control and an overview of the control and microgrid,” IEEE Trans. Power Syst., vol. 29, no. 2, pp. 698–706, Mar.
reserve management strategies,” IEEE Ind. Electron. Mag., vol. 7, no. 4, 2014.
pp. 42–55, Dec. 2013. [24] R. Majumder, A. Ghosh, G. Ledwich, and F. Zare, “Stability
[2] N. Hatziargyriou, H. Asano, R. Iravani, and C. Marnay, “Microgrids,” analysis and control of multiple converter based autonomous mi-
IEEE Power Energy Mag., vol. 5, no. 4, pp. 78–94, Jul. 2007. crogrid,” in Proc. IEEE Int. Conf. Control Autom., Dec. 2009,
[3] A. Mehrizi-Sani and R. Iravani, “Online set point adjustment for trajectory pp. 1663–1668.
shaping in microgrid applications,” IEEE Trans. Power Syst., vol. 27, [25] R. Majumder, “Some aspects of stability in microgrids,” IEEE
no. 1, pp. 216–223, Feb. 2012. Trans. Power Syst., vol. 28, no. 3, pp. 3243–3252, Aug.
[4] J. P. Lopes, C. Moreira, and A. Madureira, “Defining control strategies for 2013.
microgrids islanded operation,” IEEE Trans. Power Syst., vol. 21, no. 2, [26] Z. Liu, J. Liu, and Y. Zhao, “A unified control strategy for three-phase
pp. 916–924, May 2006. inverter in distributed generation,” IEEE Trans. Power Electron., vol. 29,
[5] M. Chandorkar, D. Divan, and R. Adapa, “Control of parallel connected no. 3, pp. 1176–1191, Mar. 2014.
inverters in standalone ac supply systems,” IEEE Trans. Ind. Appl., [27] P. Krause, O. Wasynczuk, and S. Sudhoff, Analysis of Electric Ma-
vol. 29, no. 1, pp. 136–143, Jan. 1993. chinery and Drive Systems, 3rd ed. Hoboken, NJ, USA: Wiley,
[6] K. De Brabandere, B. Bolsens, J. Van den Keybus, A. Woyte, J. Driesen, 2013.
and R. Belmans, “A voltage and frequency droop control method [28] J. Uudrill, “Dynamic stability calculations for an arbitrary
for parallel inverters,” IEEE Trans. Power Electron., vol. 22, no. 4, number of interconnected synchronous machines,” IEEE
pp. 1107–1115, Jul. 2007. Trans. Power App. Syst., vol. PAS-87, no. 3, pp. 835–844,
[7] J. Guerrero, J. Vasquez, J. Matas, L. de Vicua, and M. Castilla, “Hierar- Mar. 1968.
chical control of droop-controlled ac and dc microgrids: A general ap- [29] P. Shamsi and B. Fahimi, “Dynamic behavior of multiport power electronic
proach toward standardization,” IEEE Trans. Ind. Electron., vol. 58, no. 1, interface under source/load disturbances,” IEEE Trans. Ind. Electron.,
pp. 158–172, Jan. 2011. vol. 60, no. 10, pp. 4500–4511, Oct. 2013.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

MUELLER et al.: MODEL MODIFICATION PROCESS FOR GRID-CONNECTED INVERTERS USED IN ISLANDED MICROGRIDS 11

Jacob A. Mueller (S’12) received the B.S. and M.S. Jonathan W. Kimball (M’96–SM’05) received the
degrees in electrical and computer engineering from B.S. degree in electrical and computer engineering
the Missouri University of Science and Technology, from Carnegie Mellon University, Pittsburgh, PA,
Rolla, MO, USA, in 2012 and 2014, respectively, USA, in 1994, and the M.S. degree in electrical engi-
where he is currently working toward the Ph.D. de- neering and the Ph.D. degree in electrical and com-
gree. puter engineering from the University of Illinois at
His research interests include modeling and con- Urbana-Champaign, Champaign, IL, USA, in 1996
trol of power electronic converters, microgrid stabil- and 2007, respectively.
ity, and nonintrusive load monitoring. He was with Motorola, Phoenix, AZ, USA, from
1996 to 1998, where he was involved in designing in-
sulated gate bipolar transistor modules for industrial
applications. He then joined Baldor Electric, Fort Smith, AR, USA, where he
designed industrial adjustable-speed drives ranging 1150 hp. In 2003, he joined
the University of Illinois at Urbana-Champaign, as a Research Engineer, where
he then became a Senior Research Engineer. In 2003, he cofounded SmartSpark
Energy Systems, Inc., Champaign, where he was the Vice President of Engi-
Md. Rasheduzzaman (S’12) received the B.S. de- neering. In 2008, he joined the Missouri University of Science and Technology,
gree in electrical and electronic engineering from the Rolla, MO, USA, where he is currently an Associate Professor.
Chittagong University of Engineering and Technol- Dr. Kimball is a Member of Eta Kappa Nu, Tau Beta Pi, and Phi Kappa Phi.
ogy, Chittagong, Bangladesh; the M.S. degree in elec- He is a licensed Professional Engineer in the state of Illinois. He has served on
trical and computer engineering from Purdue Uni- the IEEE Power Electronics Society Administrative Committee, as a Member-
versity Calumet, Hammond, IN, USA; and the Ph.D. at-Large and as Education Chair, and is currently the Chairperson of the St.
degree in electrical engineering from the Missouri Louis, MO, Chapter of the Industry Applications Society.
University of Science and Technology, Rolla, MO,
USA, in 2006, 2010, and 2015 respectively.
He started working as a Visiting Assistant Pro-
fessor at Valparaiso University, Valparaiso, IN, USA,
in 2015. His research interests include microgrids, power systems, and power
electronics.