IEEE TRANSACTIONS ON ENERGY CONVERSION 1

Development of a MATLAB/Simulink Model of a

Single-Phase Grid-Connected Photovoltaic System
Michael E. Ropp, Member, IEEE, and Sigifredo Gonzalez

Abstract—Because of their deployment in dispersed locations on IEEE-1547, UL-1741, and IEC-62116, require that a grid-
the lowest voltage portions of the grid, photovoltaic (PV) systems connected PV inverter be able to detect and prevent unin-
pose unique challenges to power system engineers. Computer mod- tentional islanding [3]–[5]. The standards suggest the use
els that accurately simulate the relevant behavior of PV systems
would thus be of high value. However, most of today’s models either of a specific test to demonstrate that a PV inverter can
do not accurately model the dynamics of the maximum power point meet this requirement and be certified as a “nonislanding
trackers (MPPTs) or anti-islanding algorithms, or they involve ex- inverter,” but this test is difficult and somewhat expensive
cessive computational overhead for this application. To address this to perform in spite of its conceptual simplicity, and thus,
need, a MATLAB/Simulink model of a single-phase grid-connected modeling could also be of help in understanding the results
PV inverter has been developed and experimentally tested. The de-
velopment of the PV array model, the integration of the MPPT of anti-islanding tests.
with an averaged model of the power electronics, and the Simulink 2) Action of the maximum power point tracker (MPPT): The
implementation are described. It is experimentally demonstrated MPPT is effectively the amplitude controller for the in-
that the model works well in predicting the general behaviors of verter’s output, and its operation can strongly affect sys-
single-phase grid-connected PV systems. This paper concludes with tem behavior [6] and must also be properly simulated.
a discussion of the need for a full gradient-based MPPT model, as
opposed to a commonly used simplified MPPT model. In addition, the interaction between the MPPT and the
large buffer capacitor usually used at the dc inputs of PV
Index Terms—Inverters, islanding detection, modeling and sim- inverters is important and should be simulated.
ulation, photovoltaics (PVs).
3) Response to grid voltage and frequency fluctuations: For
PV inverters, this primarily means that the trips for abnor-
I. INTRODUCTION mal voltages or frequency must be simulated.
PV system models for use in power system simulations have
HOTOVOLTAIC (PV) systems produce electricity with-
P out producing CO2 . This property has led to worldwide
government policies aimed at increasing the deployment of
been reported by other authors (for example, [7]–[9]). Although
useful in some cases, most of these suffer from one of the
following problems.
grid-connected PV systems that are connected with, and can
1) They use full switching models of the converter that im-
export power to, utility power networks. PV systems are usu-
pose too high of a computational burden for the model to
ally deployed in a highly dispersed mode, in the lowest voltage
be useful in power system simulations.
portions of the grid (down to residential end-use voltages), and
2) They do not model the anti-islanding controls.
they thus present unusual challenges for power system engineers
3) They do not realistically capture the dynamics introduced
tasked with understanding how high penetration levels of PV in
by the PV array, input buffer capacitor, and MPPT.
industrial parks or “solar subdivisions” might impact system
This paper describes the work undertaken to produce a model
operations. A well-verified computer model would be of great
that meets this need. The paper has two goals. First, it presents
help in this task.
the model itself and describes the extensive experimental
In order to meet the needs described earlier, a number of
validation conducted in collaboration with the Distributed
important aspects of PV inverters need to be modeled with
Energy Technologies Laboratory (DETL) at Sandia National
reasonable accuracy. These include the following.
Laboratories, Albuquerque, NM. Second, after the model is
1) Islanding detection behavior: A suitable model must rea-
vetted, it is used to examine the need for the complex MPPT
sonably simulate the inverters’ means of islanding de-
model used here.
tection. Islanding detection is discussed thoroughly in
the literature [1], [2], and applicable standards, such as II. DESCRIPTION OF THE MODEL
A. Overview
Manuscript received May 30, 2007; revised February 20, 2008. This work
was supported in part by the National Science Foundation under Grant The system being modeled is shown in Fig. 1. It consists of
ECS-0238533. Paper no. TEC-00205-2007. a PV array feeding an H-bridge inverter (switches S1 –S4 ) that
M. E. Ropp is with the Department of Electrical Engineering, South Dakota feeds current into the utility voltage Vu through an equivalent
State University, Brookings, SD 57007 USA, on leave from Sandia National
Laboratories, Albuquerque, NM 87185 USA (e-mail: [email protected]). interconnecting inductance Leq that includes effects from out-
S. Gonzalez is with the Distributed Energy Test Laboratory, Sandia National put filters, transformers, and the utility source impedance. The
Laboratories, Albuquerque, NM 87185 USA (e-mail: [email protected]). power output of single-phase inverters oscillates at twice the
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. line frequency, and thus, a large buffer capacitor Cb is included
Digital Object Identifier 10.1109/TEC.2008.2003206 between the PV and the H-bridge.

0885-8969/$25.00 © 2009 IEEE

2 IEEE TRANSACTIONS ON ENERGY CONVERSION

Fig. 1. Schematic of the system being modeled.

The model includes:

1) six islanding detection algorithms:
a) Sandia frequency shift (SFS) [10];
b) the classical linear instability (CLI) method [11],
[12];
c) Sandia voltage shift (SVS) [2];
d) rate of change of frequency (RoCoF) [13];
e) impedance detection [1], [2]; Fig. 2. Plot of the “unit I–V curve” used in the model.
f) frequency bias [10];
2) two I–V curve based-models of a PV array;
present power reading from the previous, to get the algebraic
3) a perturb-and-observe (P&O) MPPT [14];
sign of the change in power. That sign is used to set a flag in
4) an s-domain representation of an L–C output filter.
the block labeled “Convert CtZ block output”; if the change in
This model is unique in its inclusion of CLI and in its ability
power is positive (output of “Compare To Zero” block is 1), the
to use any of the anti-islanding methods in combination (which
flag is set to +1, and if negative or zero (output of “Compare To
is common in practice), except for CLI and SFS.
Zero” block is 0), the flag is set to −1. The flag then multiplies
a user-settable MPPT increment, which, in turn, is added to the
B. MPPT present PV + capacitor current to increment or decrement it as
The PV array, capacitor Cb , and MPPT are usually modeled appropriate.
in one of two ways for power system simulations: as a Thevenin
equivalent or as a constant-power source regulated by a classical C. PV Array Model
feedback loop. These models have two major advantages: high Modeling the PV array in this context is always a challenge.
computational efficiency and the lack of a need for a detailed PV The physical I–V relationship based on the Shockley equation
model. However, virtually all real PV inverters use MPPT algo- requires parameters that are rarely available to power electronics
rithms that belong to the family of gradient-based optimization or power system engineers and is thus unsuitable. Mathematical
methods [15]. The most common of these is the P&O method, curve fits are available, but most of these are cumbersome to use
which is a variant of the simple 1-D gradient-based minimiza- in this application.
tion technique. Modeling this type of MPPT using a simplified To achieve accuracy while minimizing computational bur-
model might be problematic for two main reasons. den and maintaining simplicity for the user, this model gives
1) The dynamics of gradient-based controls and the linear the user two options for modeling the PV array. The first is
controller are inherently different [15]. a highly simplified approach based on the “unit” I–V curve
2) The linear controller MPPT model does not model the PV shown graphically in Fig. 2 and in tabular form in Table I, de-
array and capacitor Cb , and thus, cannot properly simulate veloped by the authors of this paper to approximately represent
the energy-limited nature of the source. the I–V curve of a single generic PV cell, except for the fill
Thus, in this paper, a P&O MPPT model along with a model factor of 0.682, which is more representative of an array.
of the PV array and capacitor was implemented. The PV + This “unit” curve is scaled to any arbitrary system size by en-
capacitor model is covered in the next section. A block diagram tering the open-circuit voltage at standard test conditions (STC)
of the P&O MPPT model developed here is shown in Fig. 3. and the nominal STC power of the array, so the user need know
This model implements a basic 1-D gradient search with a fixed only these two parameters. The I–V curve can be changed to
step size [15]. The instantaneous PV inverter ac output current more closely represent a particular array by changing the vectors
(i3 in Fig. 1 and IPV in Fig. 3) and the point of common coupling used in the lookup table.
voltage (vu in Fig. 1 and VPCC in Fig. 3) are multiplied to obtain This approach is useful for most system-level applications,
the inverter output power, which is averaged over a time window but it has one important drawback: it fixes the fill factor (FF),
of one-tenth of the user-defined MPPT stepping interval. The which in reality depends on irradiance and temperature. Because
PV power measurement passes through a zero-order sampling the FF affects MPPT performance, a second PV array modeling
block. The “Integer Delay” block holds the sampled power for option is included, i.e., of accepting I–V curve data from a
one stepping interval, so that the “Subtract” block subtracts the program such as “Sandia IVTracer2” that generates accurate
ROPP AND GONZALEZ: DEVELOPMENT OF A MATLAB/SIMULINK MODEL OF A SINGLE-PHASE GRID-CONNECTED PV SYSTEM 3

Fig. 3. Perturb-and-observe maximum power point tracker. Signal propagation is from right to left (the inputs are at the right).

TABLE I E. Modeling of the Power Stage

VALUES USED IN THE “UNIT I–V CURVE” LOOKUP TABLE
The model would not be useful for system simulations if a
full switching model of the power stage were used, because
simulation times and data file sizes would be excessive. In-
stead, an average model of the power stage was used, and for
the hysteresis-controlled H-bridge, this model is merely a gain.
However, one challenge that had to be overcome in this research
was how to integrate this averaged model with the MPPT model
described earlier.
The approach developed here is as follows. Output power
control is discussed here because all of the converter’s dynamics
are in the MPPT feedback loop, making it the more complex
case (input power control is also available). Since the output
current (i3 in Fig. 1) feeds into an essentially constant utility
voltage source, increasing i3 will always result in an increase in
output power. Thus, the MPPT will command ever-increasing
currents until the process becomes limited by the gain of the
power stage. At that point, even though the MPPT commands
an increase in current, the output current will actually decrease,
the output power will fall, and the P&O MPPT will reverse
direction. The gain of the power stage is clearly a function of
I–V curves from an extensive database of measured module voltage v1 in Fig. 1, and that functional dependency must be
parameters. The Simulink model that implements this latter PV modeled. There are two ways to approach this problem. The
array model for three different irradiances is shown in Fig. 4. first is to recognize that the H-bridge inverter is a buck-type
The user copies and pastes into the I and V blocks the current converter. If one considers a PWM buck dc–dc converter, the
and voltage vectors at different irradiances from IVTracer2. The relationship of output current to input voltage can be determined
“Subtract” block implements Kirchhoff’s current law at the node as follows:
at the top of capacitor Cb in Fig. 1; it subtracts the load current
from the PV current (i1 –i2 ). The resulting capacitor current is vo Zio
=D=
used to compute the capacitor voltage via the 1/capacitance and vi vi
“Integration by the capacitor” blocks. The capacitor and PV D
voltage are the same (v1 in Fig. 1), so the capacitor voltage is io = vi (1)
Z
fed back into the I–V lookup table.
where the input voltage vi is the converter’s input voltage, vo is
the converter’s output voltage, io is the output current, Z is the
D. Anti-Islanding Methods impedance of the load fed by the converter, and D is the duty
One distinguishing feature of this model is that it combines ratio. The output current is proportional to the input voltage. In
the realistic MPPT model with models of six anti-islanding the present case, the load is not simply an impedance but rather
methods. Fig. 5 shows the implementation of one of these, SFS a nonideal (Thevenin) voltage source, the determination of Z is
and frequency bias. This implementation is somewhat novel in nontrivial, and D would not be constant.
that it uses the phase-locked loop (PLL) that produces the sine Another approach would be to consider that in the hysteresis-
wave reference for the inverter’s output current, which modestly controlled H-bridge inverter, the current i3 in Fig. 1 is controlled
improves the computational speed over lookup-table-based ver- using the H-bridge to apply ±v1 at v2 , thereby controlling the
sions. Over/undervoltage and over/underfrequency relaying are time rate of change of the current through Leq . Consider the case
also modeled. in which the PV array is nearing maximum power operation after
4 IEEE TRANSACTIONS ON ENERGY CONVERSION

Fig. 4. Block diagram of the PV array model. Note that “Io u t ” is the inverter output current and is an input here.

Fig. 5. SFS waveform generator. Signal propagation is from right to left (the input is at the right).

starting from zero current. The current i3 is increasing, and the where ω is the frequency of vU (in radians per second) and I3m
voltage v1 has fallen to the point at which the converter cannot is the maximum value of i3 (in Amperes). The approximation is
further increase the amplitude of the current. This case will acceptable near the beginning of the cycle of the sine wave, and
appear approximately as shown in normalized form in Fig. 6, as long as the hysteresis bandwidth is small. Combining these
which shows the sine wave reference used by the hysteresis equations and considering the case of t = 0 allows a calculation
controller, the hysteresis boundaries, and a linear approximation of the maximum amplitude of the current as a function of the
to the sine wave at t = 0. The PV voltage becomes so low that PV voltage
the current through Leq “ramps” at approximately the same rate
vPV
as the reference sine wave, and the bridge does not switch at all I3m = . (3)
during the first portion of the sine wave. Under this condition, ωLeq
near the beginning of the sine wave (t = 0), the derivative of i3 This calculated value of I3m will actually be above the true
can be expressed as value, but the relationship again supports the notion that it is
di3 v2 − v U physically reasonable to make the converter gain proportional
= ≈ ωI3m cos ωt (2)
dt Leq to v1 , with the constant of proportionality being somewhat less
ROPP AND GONZALEZ: DEVELOPMENT OF A MATLAB/SIMULINK MODEL OF A SINGLE-PHASE GRID-CONNECTED PV SYSTEM 5

Fig. 7. Experimentally measured MPPT power ripple.

Fig. 6. Plot of normalized reference current waveform for hysteresis control

(solid light line), hysteresis bands (dashed light lines), and straight-line approx-
imation to the reference at t = 0 (heavy solid line).

than that given by (3). The approach followed in the model

being discussed here was to normalize I3m by its maximum
value, which would be attained if the PV array reached its open-
circuit voltage VOC , and multiply by a gain constant K:
v1
I3m ,norm = K . (4)
VOC
Now, explicit knowledge of Leq is not required. A value of
K = 3 has proven to be reasonable for most system-level work,
but if higher accuracy is desired, the value of K can be adjusted
until the maximum power increment specified by the user is of
appropriate size, near the beginning of a simulation started from
zero initial conditions.

III. MODEL VALIDATION

Fig. 8. Effect of MPPT decision rate on PV output power, for fixed capacitor
One feature that sets this model apart from others in the lit- size and MPPT increment.
erature is the level to which it has been experimentally vetted.
The model was extensively tested by comparing model results
against experimentally obtained data from the Distributed En-
ergy Technologies Laboratory (DETL) at Sandia National Lab-
The model was tested to see whether it correctly pre-
oratories. A number of test cases have been examined, a few of
dicted this behavior. A representative example is shown in
which will be presented here.
Fig. 8. In this simulation, the PV array’s maximum power was
1265 W, irradiance was fixed, and the size of Cb was 40 mF.
A. Effect of MPPT Decision Rate on MPPT Stability The heavy trace was produced using an MPPT stepping rate of
It is known that some PV inverters exhibit a considerable 2 Hz (MPPT stepping time interval of 0.5 s), and the lighter
low-frequency ripple because of the time lag required for the trace was produced using an MPPT stepping rate of 5 Hz. If the
PV array + capacitor to reach its new operating point when MPPT steps too quickly, the low-frequency oscillation evident
the MPPT perturbs the operating state. This can reduce steady- in the lighter trace in Fig. 8 begins to appear, i.e., the model
state MPPT efficiency and lead to undesired shutdowns during predicts the onset of undesired oscillation as the MPPT step-
changing irradiance conditions, but it can also reduce run-on ping time interval is increased. If the rate increases too far—in
times during islanding tests. Fig. 7 shows an experimentally this case, 10 Hz was sufficient—the MPPT becomes unstable.
measured example; the vertical axis indicates the power output Similar behavior could be observed if the MPPT stepping time
as a percent of nameplate rating. There is approximately a 0.3% interval was fixed but the buffer capacitor size was increased.
MPPT ripple in this example, at a frequency of about 0.15 Hz. This is consistent with known inverter behavior, suggesting the
The general downward slope is due to changing irradiance. MPPT model is realistic.
6 IEEE TRANSACTIONS ON ENERGY CONVERSION

Fig. 9. Comparison of the NDZ for SFS predicted by the phase criterion Fig. 10. Average run-on times for four SFS inverters at full power (experi-
(“theoretical,” marked by dashed lines) and by the model (heavy lines with mental result).
squares).

C. Using the Model to Understand an Observed Interaction

Between Islanding Prevention and MPPT Functions
B. Anti-Islanding Method Nondetection Zones and Behavior as
a Function of Load Quality Factor This test was described in an earlier paper [9]. The model was
used to help explain an experimental observation that a partic-
The modeled islanding prevention schemes have been thor- ular commercially available PV inverter did not island when its
oughly tested by comparison with both earlier theoretical work MPPT was active, but islanded readily for certain loads when
and laboratory results. In general, the anti-islanding models ap- the inverter deactivated its MPPT due to excessive tempera-
pear to work well, predicting general behaviors and nondetection ture or input power. The model demonstrated that the reason
zones (NDZs) reasonably accurately. was that the MPPT was marginally stable, so that when the
Fig. 9 shows the results of one set of tests to determine the utility was lost, the actual mechanism by which islanding was
NDZs predicted by theory (phase criteria [16]) and the model. detected was that the perturbation caused by the anti-islanding
The purpose of the test was to see how accurately the model algorithm destabilized the MPPT, eventually leading to an input
predicted the location of the edge of the NDZ. The SFS gain low-voltage cutoff.
was 0.05, the load was a parallel RLC with R = 11.8 Ω, and
the MPPT was turned “off”. Fig. 9 shows generally reason-
able agreement, but there are two discrepancies apparent: the D. Comparison of Modeled and Experimental Results in a
Multiinverter Test
model NDZ is slightly shifted downward from the theoreti-
cal one, and the trend in the model’s lower boundary appears Here, we describe a use of the model to address a question
to be inverted from expectations. The downward shift is at- that to the best of our knowledge has not been addressed in
tributed to the fact that the model’s simulated inverter output the literature to date. One issue of concern to utility planners
current lags the voltage by about 2◦ . That phase shift comes is the high-penetration case, where the total power rating of a
from the inverter’s output low-pass filter, which was left as-is number of small independent PV systems connected to a given
because it and its phase shift are representative of “real-world” portion of the utility system is a significant fraction of the power
inverters. capacity of that section of the utility. One worry about the high-
The reason for the lower boundary discrepancy appears to penetration case is that the loss-of-mains detection systems of
lie within the PLL used to synchronize the inverter’s cur- the various PV systems may interfere with each other in such a
rent reference wave to the utility voltage. The Q-factors of way that they lose effectiveness.
loads toward the left side of Fig. 9 are very high, resulting To address this, a high-penetration experiment was carried
in an inevitable phase “glitch” upon loss of utility. The PLL out at the Distributed Energy Test Laboratory at Sandia National
senses this phase jump and adjusts its frequency in an at- Laboratories. This test involved four 2.5-kW PV inverters using
tempt to maintain phase lock, resulting in a frequency transient both SFS and SVS and feeding a common RLC load with a Q
that includes an overshoot. For L-values of 300 and 100 µH of 2.5, tuned to resonate at 60 Hz. The results of the full-power
(the leftmost two values in Fig. 9), the overshoot for Cnorm - tests are shown in Fig. 10.
values less than 0.9875 is large enough to result in an over- In each individual test, the run-on times of the four invert-
frequency trip. The same effect does not appear for the upper ers were tightly clustered, but between tests, there was a very
boundary because of the asymmetry of the frequency trip lim- broad variability. In some cases, the inverters shut down within
its; the PLL transient exceeds the 60.5-Hz limit on the high ten line cycles, but in one case, the inverters ran on for over
side but does not quite reach the 59.3-Hz limit on the low 100 cycles and came close to exceeding the 2-s limit allowed
side. by IEEE-1547. (In no case did they actually exceed that limit.)
ROPP AND GONZALEZ: DEVELOPMENT OF A MATLAB/SIMULINK MODEL OF A SINGLE-PHASE GRID-CONNECTED PV SYSTEM 7

Fig. 11. Results of simulations of four inverters in which the utility cutoff
time was varied, with all inverter and load parameters fixed. Fig. 12. Simulations of two SVS anti-islanding tests: one using a linear
controller-based MPPT simulation (heavy dashed trace) and one using the
gradient-based MPPT model described here (lighter solid trace).
This variability is an indication of how difficult this test is to
control completely, and also of the complexity of the interac-
tions between the two anti-islanding methods and the MPPT.
However, interpretation of the results is difficult and controver-
sial. Which of the variety of run-on times is most representative
of the inverters’ performance in this situation?
This four-inverter case was simulated using the model pre-
sented in this paper. The inverters were connected in parallel,
without interconnecting impedances, and the parameters were
set as follows:
Vu ,norm = 240 Vrms
R = 5.76 Ω L = 6.115 mH C = 1.1513 mF
MPPT stepping interval = 0.25 s MPPT increment = 2
SFS gain = 0.0185 SVS gain = 0.01.

Then, a number of simulations were run to study the effects of

varying several parameters. One of the most interesting results Fig. 13. Two simulated inverters with SVS during a utility voltage transient.
was obtained when varying the time at which the utility was One inverter uses a linear controller MPPT simulation (heavy dashed trace); the
disconnected over the range of 2–4 sec. The results of this set of other uses the gradient-based MPPT model described here (lighter solid trace).
simulations are shown in Fig. 11. The horizontal-axis variable
in Fig. 11 is the time at which the utility was disconnected, with understanding of the behavior in Fig. 10, because it is caused
t = 0 being the start of the simulation. A number of closely by interactions that are difficult to isolate experimentally.
spaced simulations were run near 3 s to investigate the effect
of cutting the utility off at different points in the sine wave.
IV. COMPARISON OF MPPT MODELING METHODS
Fig. 11 suggests that this factor made only a small difference,
but when a larger time step was used, corresponding to different After the model was well verified, it was used to examine
locations within the interacting cycles of SFS, SVS, and the the aforementioned concern about potential differences between
MPPT, it was found that the run-on time variation was only due the linear controllers typically used to model MPPTs in large
to changing the utility cutoff time varied by almost a factor of system simulations and the actual gradient-based MPPTs in
3, from 0.65 to 1.93 s. The maximum run-on time and range real converters. In many cases, the linear MPPT model actu-
of run-on times predicted by the model are reasonably close ally worked well, but many problems also became apparent.
to those measured in the experiments. Figs. 10 and 11 suggest Although anomalies were found for many combinations of in-
that all of the run-on times found during the experiment are verter functions, simulations involving the SVS method were
characteristic of the combination of inverters, and supports the particularly problematic. Two representative results are shown
test protocols specified in UL-1741, IEEE-1547, and IEC-62116 in Figs. 12 and 13. Fig. 12 shows two SVS anti-islanding simula-
that use as their benchmark the longest run-on time obtained over tions, one with the linear controller (heavy dashed trace) and one
several tests. This case also demonstrates the value of the model; with the gradient-based MPPT (lighter solid trace). The utility
without this “virtual laboratory,” it would be difficult to gain an is lost at t = 12 s. The linear controller runs on indefinitely, in
8 IEEE TRANSACTIONS ON ENERGY CONVERSION

part because it does not model the energy-limited nature of the [5] International Electrotechnical Commission TC82 working document,
source and in part because its output does not contain the MPPT IEC-62116 “Test Procedure of Islanding Prevention Measures for Utility-
Interconnected Photovoltaic Inverters” revised, Sep. 29, 2006.
perturbations, but the gradient-based inverter shuts down fairly [6] M. Ropp, J. Ginn, J. Stevens, W. Bower, and S. Gonzalez, “Simulation
quickly. In this case, the linear MPPT can lead to erroneous and experimental study of the impedance detection anti-islanding method
conclusions. in the single-inverter case,” in Proc. 4th World Conf. Photovolt. Energy
Convers., May 2006, pp. 2379–2382.
Fig. 13 shows simulations of the same two inverters during [7] F. Fernandez-Bernal, L. Rouco, P. Centeno, M. Gonzalez, and M. Alonao,
a utility voltage transient, with the SVS anti-islanding method “Modelling of photovoltaic plants for power system dynamic studies,” in
active. At time t = 12 s, the utility voltage undergoes a step Proc. 5th IEEE Int. Conf. Power Syst. Manage. Control, 17–19 Apr. 2002,
pp. 341–346.
decrease from 120 to 110 Vrm s . The inverters are identical [8] M. Ciobotaru, T. Kerekes, R. Teodorescu, and A. Bouscayrol, “PV inverter
except for the MPPT model. Because the energy constraints on simulation using MATLAB/Simulink graphical environment and PLECS
the PV + capacitor source are not modeled, the simulated SVS blockset,” in Proc. 32nd IEEE Ind. Electron. Conf., Paris, France, Nov.
2006, pp. 5313–5318.
using the linear MPPT model increases the output voltage much [9] Z. Ye, R. Walling, L. Garces, R. Zhou, L. Li, and T. Wang, “Study and
more than it would be able to in reality. The response of the development of anti-islanding control for grid-connected inverters,” Nat.
energy-constrained gradient-based MPPT, which does model Renewable Energy Lab, Golden, CO, May 2004, Rep. SR-560-36243,
May 2004.
the PV + capacitor, is smaller and in the opposite direction. The [10] M. Ropp, M. Begovic, and A. Rohatgi, “Analysis and performance assess-
model without the energy constraint could lead to erroneous ment of the active frequency drift method of islanding prevention,” IEEE
conclusions about such things as the impact of high penetrations Trans. Energy Convers., vol. 14, no. 3, pp. 810–816, Sep. 1999.
[11] R. A. Jones and T. R. Sims, “Investigation of potential islanding of dis-
of PV during system transients. tributed photovoltaic systems,” Sandia Nat. Lab., Albuquerque, NM, Rep.
SAND-87-7024, 1988.
V. CONCLUSION [12] A. Cocconi, S. Çuk, R. D. Middlebrook, and T. S. Key, “Design and
development of a 4-kW/20 kHz DC-isolated power conditioner for util-
This paper presents an accurate behavioral model of a grid- ity interactive photovoltaic applications,” Sandia Nat. Lab., Albuquerque,
NM, Rep. SAND-87-0353, 1983.
tied PV system suitable for system-level investigations. Simpli- [13] C. M. Alfonso, W. Freitas, W. Xu, and L. C. P. da Silva, “Performance of
fied means for modeling the PV array and integrating a gradient- ROCOF relays for embedded generation applications,” Inst. Electr. Eng.
based MPPT into a very simple averaged model of the power Proc. Gener., Transmiss., Distrib., vol. 152, no. 1, pp. 109–114, Jan. 2005.
[14] D. Hohm and M. Ropp, “Comparative study of maximum power point
converter are developed, and the model has been experimentally tracking algorithms,” Prog. Photovolt., vol. 11, pp. 47–62, Jan. 2003.
vetted. Finally, it has been shown that MPPT models based on [15] J. S. Arora, Introduction to Optimum Design. New York: McGraw-Hill,
linear controllers can be inadequate for this application. 1989.
[16] M. E. Ropp, M. Begovic, A. Rohatgi, G. A. Kern, R. Bonn, and S. Gonza-
lez, “Determining the relative effectiveness of islanding prevention tech-
ACKNOWLEDGMENT niques using phase criteria and nondetection zones,” IEEE Trans. Energy
Convers., vol. 15, no. 3, pp. 290–296, Sep. 2000.
The authors wish to acknowledge the following for their valu-
able contributions to this paper: SDSU students R. Bhandari,
E. Grebel, and D. Schutz; C. Thompson of Xantrex, Inc.; and S.
Atcitty, W. Bower, B. Boysen, D. King, and J. Stevens of San-
dia National Laboratories. Sandia is a multiprogram laboratory Michael E. Ropp (M’99) was born in Rapid City, SD, in 1967. He received
the B.Sc. degree in music from the University of Nebraska-Lincoln, Lincoln, in
operated by Sandia Corporation, a Lockheed Martin Company, 1992, and the M.S. and Ph.D. degrees in electrical engineering from the Georgia
for the United States Department of Energy’s National Nuclear Institute of Technology, Atlanta, in 1996 and 1998, respectively.
Security Administration under contract DE-AC04-94AL85000. He is an Associate Professor of electrical engineering at South Dakota State
University, Brookings. His current research interests include issues relating to
interfacing renewable energy systems to utility grids, power electronics for
REFERENCES photovoltaics, distributed controls, and computer modeling of power systems.

[1] M. E. Ropp, M. Begovic, and A. Rohatgi, “Prevention of islanding in

grid-connected photovoltaic systems,” Prog. Photovolt., vol. 7, no. 1,
pp. 39–59, 1999.
[2] W. Bower and M. Ropp, “Evaluation of islanding detection methods for
utility-interactive inverters in photovoltaic systems,” Sandia Nat. Lab.,
Albuquerque, NM, Rep. SAND2002-3591, Nov. 2002. Sigifredo Gonzalez received the M.S.E.E. degree from New Mexico State
[3] IEEE Standard for Interconnecting Distributed Resources with Electric University, Las Cruces, in 1996.
Power Systems, IEEE Standard 1547-2003, Jul. 28, 2003. He has been an Electrical Engineer of the Technical Staff, Sandia National
[4] Inverters, Converters, and Controllers for Use in Independent Power Sys- Laboratories, Albuquerque, NM, for 11 years. His current research interests
tems, Underwriters’ Laboratories Standard UL-1741, Standard for Safety, include performance, reliability, and utility interconnected standards compliance
May 1999. evaluations of photovoltaic dc/ac inverters.