Академический Документы
Профессиональный Документы
Культура Документы
3, September 1999
810
I. INTRODUCTION
Islanding of a grid-connected independent generator such
as a photovoltaic (PV) system, occurs when a section of a
utility system containing such generators is disconnected
from the main utility voltage source, but the independent
generators continue to energize the utility lines in the isolated
section. Islanding of PV systems is a concern primarily
because it poses a safety hazard to utility repair and
maintenance personnel [l]. There is also some concern that
islanding of PV systems may disrupt the arc-clearing function
of utility reclosers [2], or that out-of-phase reclosure may
occur and lead to damage to the PV system or utility
equipment.
PE-288-EC-0-07-1998 A paper recommended and approved by the
IEEE Energy Development and Power Generation Committee of the
IEEE Power Engineering Society for publication in the IEEE
arg{ R-
+ (jmL)- + j d }-I
=0
(3)
811
I \
Grid
PV array
Power
conditioning
breaker
(reclaser)
..
..
..
.
..
..
.
..
..
..
..
..
..
..
..
..
..
.
.
1s
A. An analytical approach
In order to characterize the performance of AFD, we
utilized a simple analytical approach based on the describing
function analysis technique [6,7]. We first noted that the
system can be separated into linear and nonlinear subsystems
as shown in Figure 3. The nonlinear subsystem is the PV
system, and the linear subsystem is the parallel RLC load.
We then assumed that, for the purposes of this simplified
analysis, the higher-order harmonics in the AFD waveform of
Figure 2 can be neglected. This assumption is reasonable
812
(Y
44
arg{R-
+ (jd)-+ j
d } = 0 . 5 ~t ., = 0 . 5 ~cf.
(4)
where V,,,,, and Va,+ are the lower and upper voltage trip set
points respectively. Throughout this paper and for purposes
of illustration, we have assumed the frequency trip limits to
be k0.5 Hz from the nominal line frequency, the
Time (sec)
813
AFD. If 61 lies outside the frequency trip set points or I
1.20
Equation ( 5 ) is not satisfied, then the load lies outside the
AFD NDZ. MATLAB was used to carry out this long
sequence of calculations. Figure 6 shows an example of the
P
results of this procedure for R = 14.4 R, cf= 1%, 5% and
1.10
lo%, and several values of L and C. (Note: R = 14.4 C2
.'1
corresponds to a 1-kW load at 120 Vms.) The light
.E 1.05
horizontal dashed lines indicate the NDZ of a PV system
N
a
using only the OFWUFR, without AFD. Recall that our
1.00
purpose is to examine whether the NDZ of AFD overlaps
with that of the O F W F R . Figure 6 shows that this does not
0.95 1
happen for relatively large L and small C, but that the two
1.00E-04
1.00E-03
1.00E-02
1.00E-01
NDZs do overlap for small L and large C. Therefore, AFD
Inductance
(L)
fails to detect islanding for these loads. However, note that
the width of the NDZ decreases rapidly for inductances
smaller than about 1 mH. This is because of the sharpness of Figure 6. The NDZ of AFD found by the simplified
the resonant peak of the load's magnitude response in that analytical model for several values of cf. In these
range; with large C and small L, small deviations from simulations, R = 14.4 R. The shaded region is the
nominal frequency (which are caused by AFD) lead to large OFRAJFR NDZ.
changes in the magnitude of the load impedance, leading to a
detectable voltage deviation (assuming the magnitude of iPv
fixed) and PV system shutdown. With large L and small C,
30~
the magnitude response of the load is much flatter, and
25.therefore the voltage trips do not play a role in that range.
20.From Figure 6, it is clear that cfmust be fairly large (greater
than 1%) for AFD to be effective over a wide range of
$
15.resonant inductance-capacitance values. Unfortunately, it is
clear from Figure 2 that this can only be done at the expense
10of increased total harmonic distortion of the PV system's
output current (THD,). This distortion can be quantified by
5,
calculating the Fourier coefficients of the AFD waveform for
0
several values of cf: This was done, and the results, shown in
0
0.05
0.1
0.15
0.2
0.2:
Figure 7, indicate that the THD, is linear with CJ The
"Choppingfraction";percent of zero time per half-cycle
allowable limit of THD, therefore sets an upper limit on the
maximum usable value of cJ
Figure 7. Plot of the THD, vs. chopping fraction for the
Figure 8 shows the results of varying R on the NDZ of waveform in Figure 2.
AFD for a fixed c j Again, the light horizontal dashed lines
indicate the NDZ of the OFIUUFR. We see that increasing R
1.2
(that is, decreasing the load real power demand) causes a
larger overlap region between the NDZs of AFD and the
O F W F R , thereby decreasing the effectiveness of AFD,
1.1s
whereas the effectiveness of the method improves for heavier
loads. The NDZ becomes centered around unity-dpf loads at
1.1
higher C because the larger capacitors are more capable of
supporting the voltage v, during the zero portions of ipv,
1.05
reducing the effect of the AFD.
As a first-cut validation of the analytical model, we
1
investigated the case of a resistive load with a small amount
of capacitance added. This case was chosen because
0.95
experimental observations are available [SI and show that
1.00E-04
1.00E-03
1.00E-02
1.00E-01
AFD fails for such loads. The analytical model described
Inductance (H)
here predicts this as well; for a 14.4 R load and cf= 5%, it
predicts that the addition of between 14.4 and 14.6 pF will Figure S. The NDZ of AFD found by the simple
cause AFD to fail, establishing agreement between the analytical model for several values of R. In these
models and the behavior noted in the experiments.
simulations, cf= 5%. The shaded region is the OFWUFR
NDZ.
~
.
-
--a
E
814
B. A simulation approach
As a second approach to verifying the analysis just
described, we have modeled the system shown in Figure 1
using MATLAB. The model contains three sections: a PV
system model, a loadlv, model, and a zero crossing detector
(ZCD) model. The PV system is simply modeled as a
programmable current source whose output is the AFD
waveform in Figure 2. One idealization in this model which
bears mentioning is that it is assumed that the PV system
current jumps to zero and begins a new cycle on each rising
zero crossing of the utility, regardless of the previous value
of the output current. Under most circumstances, this
assumption makes no difference because the values of cf
being used are small.
The equations for the loadv, model are derived by
discretizing the differential equations of the load branch
currents, summing at node a, and solving for v,.
The ZCD model simply detects the rising zero crossing of
v, by detecting the appropriate change in polarity. It also
measures the period of v, by subtracting the time at which the
previous rising zero crossing occurred from the time of the
present zero crossing. The measured period of v, is updated
every cycle.
As before, to verify the model, we examined the cases of
a purely resistive load and a resistive load with a small
amount of added capacitance. Again, the model behavior is
consistent with the previously mentioned experimental
findings; AFD always drives the frequency of the resistive
load beyond the trip setpoints within one line cycle, but the
addition of a small amount of capacitance (a few microfarads
for a 14.4 R load) results in nondetection.
For nearly-resonant loads, the results agree well with
those found using the simplified analytical model. Figure 9
shows a plot of the frequency measured at each rising zero
crossing of v, as a function of time for the case R = 14.4 a, L
= I mH, cf = 5%, and C varying over a narrow range about
the value C = 7.036 mF which resonates with the given L at
60 Hz. The range of power factors of these loads is about
80% lagging to about 75% leading. The frequency of v,
increases or decreases initially, but then levels off, and for C
= 7.0 and 7.1 mF the final frequency is within the k0.5 Hz
trip thresholds.
The capacitances at which the final
frequency coincides with the trip thresholds are the NDZ
boundary capacitances. Continuing the process used to
obtain Figure 9 for several values of L, we have determined
these boundary capacitances, and therefore the NDZ of AFD,
using this model for cf = 5% and compared these results with
those found using the simplified analytical model. The
results are shown in Figure 10. The agreement between the
simplified model and the simulations is very good, within 5%
over the entire range of L and C considered. The reason for
the deviation of the simplified analytical model from the full
simulation at higher L values (smaller C values) i s that the
assumption that v, is only a function of the fundamental of ipv
is less valid there.
60.
C = 6.9 mF
BO.
60.4
Jl
C = 7.2 mF
0.05
0.1
0.15
0.2
Time Isad
0.25
0.3
= C f k - , + F(A@R)
(6)
815
~_
_
_
~
_
_
l.E-05
l.E-04
___
1.603
l.E-02
Inductance (H)
1.E-01
61.5,
...................................................................
816
knowledge of the load, and have verified this procedure both
by simulation and by using the currently-available
experimental data. Finally, to correct the shortcomings of
AFD which were discovered during this analysis, we have
presented AFDPF, an improvement to AFD which enhances
its effectiveness in many ways, including reducing the size of
its NDZ and the size of the region of overlap of the NDZs of
AFD and the OFWUFR, allowing PV systems so equipped to
reinforce negative frequency deviations, and reducing the
loss of effectiveness in the multiple inverter case. Future
work will include: 1) further experimental validation of the
modeling results; 2) investigation of the effects of nonlinear
load components, motor loads, and constant-power loads; 3)
optimization of AFDPF by selecting the optimal F that
minimizes the size of the N D Z but avoids such problems as
steady-state instability and subharmonic production; and 4)
detailed analysis of AFDPF in the multi inverter case.
V. ACKNOWLEDGMENTS
The authors gratefully acknowledge the contributions of
the following people to this work: Richard Bass, Georgia
Institute of Technology; Ward Bower and John Stevens,
Sandia National Laboratories; Greg Kern, Ascension
Technology; Rob Wills, Advanced Energy Systems.
VII. REFERENCES
[I] J. Stevens, Utility Intertied Photovoltaic System Islanding
Experiments, Proceedings of the 17th IEEE Photovoltaic
Specialists Conference (1987), p, 1134-1 138.
[2] P. Longrigg, Effects on Electrical Distribution Networks of
Dispersed Power Generation at High Levels of Connection
Penetration, International Journal of Ambient Energy 3(4)
October 1992, p. 199-214.
[3] W. Mombauer, K.-H. Weck, Load Modelling for Harmonic
Flow Calculations, European Transactions on Electrical
Power Engineering 3(6) NovlDec 1993, p. 453-460.
[4] A. Kitamura, M. Okamoto, F. Yamamoto, K. Nakaji, H.
Matsuda, K. Hotta, Islanding Phenomenon Elimination Study
at Rokko Test Center, Proceedings of the 1st IEEE World
Conference on Photovoltaic Energy Conversion (1994) pt. I , p.
759-762.
[5] G. Kern, SunSine300: Utility Interactive AC Module AntiIslanding Test Results, Proceedings of the 1997 IEEE
Photovoltaic Specialists Conference, Anaheim, CA.
[6] S. J. Ranade, N. R. Prasad, S. Omick, L. F. Kazda, A Study of
Islanding in Utility-Connected Residential Photovoltaic
Systems, Part I: Models and Analytical Methods, IEEE
Transactions on Energy Conversion 4(3) Sept 1989, p. 436445.
[7] J-J. E. Slotine, W. Li, Applied Nonlinear Control, Prentice-Hall
1991.
Michael E. Ropp (S94) earned the B.S. degree from the University
of Nebraska in 1991 and the M,S,E,E, degree from the Georgia
Institute of Technology in 1996. He is currently a Ph.D. candidate
at Georgia Tech, His research interests include photovoltaic
systems; power electronics; and communications.