Documente Academic
Documente Profesional
Documente Cultură
2, APRIL 2013
AbstractThis paper presents simulation results for a taxonomy and are best interpreted as indicating that steady-state voltage
of typical distribution feeders with various levels of photovoltaic and current do not limit PV penetration in the relevant cases,
(PV) penetration. For each of the 16 feeders simulated, the max- not that very high PV penetrations are necessarily achievable.
imum PV penetration that did not result in a steady-state voltage
or current violation is presented for several PV location scenarios: Previous reports have looked at various penetrations of PV on
clustered near the feeder source, clustered near the midpoint of a single feeder in steady state, concluded that reverse power flow
the feeder, clustered near the end of the feeder, randomly located, can cause overvoltages, and examined methods of resolving this
and evenly distributed. In addition, the maximum level of PV is problem [4], [5]. In [6], reverse power flow due to evenly dis-
presented for single, large PV systems at each location. Maximum tributed generation was shown to exceed transformer thermal
PV penetration was determined by requiring that feeder voltages
stay within ANSI Range A and that feeder currents stay within the ratings on a sample feeder. Many other papers have focused on
ranges determined by overcurrent protection devices. Generation power curtailment or provision of reactive power to alleviate
ramp rates, protection and coordination, and other factors that feeder voltage problems that arise with high PV penetrations
may impact maximum PV penetrations are not considered here. on single representative feeders, e.g., [7][9]. Other works have
Simulations were run in GridLAB-D using hourly time steps over a examined various PV penetrations in specific locations, e.g.,
year with randomized load profiles based on utility data and typical
meteorological year weather data. For 86% of the 336 cases simu- [10], [11], or in simplified feeder models with only a few buses
lated, maximum PV penetration was at least 30% of peak load. [12]. The study described here appears to be the first to evaluate
maximum steady-state PV penetrations at several locations on
Index TermsDistributed power generation, photovoltaic (PV)
systems, power distribution, power system simulation. a range of circuits designed to be representative of all U.S. ra-
dial feeders. This work does not consider the use of PV-based
reactive power or other PV-based means of regulating voltage.
I. INTRODUCTION A commonly used rule of thumb in the U.S. allows distributed
PV systems with peak powers up to 15% of the peak load on a
feeder (or line section) to be installed without a detailed dis-
II. METHODOLOGY
In this study, PV penetration is defined as the ratio of total
peak PV power to peak load apparent power on the feeder Fig. 2. Linear fit of nominal load to maximum feeder load found by simulation.
(1)
load given in [14].1 Fig. 2 shows an analysis of the corre-
lation between maximum scaled feeder loads as found by
simulation and nominal feeder loads. While overall corre-
A. Feeder Loading lation between simulated maximum load and nominal load
The taxonomy feeders used for this study do not incor- is relatively strong, certain individual feeder peak loads di-
porate time-varying load information. Annual load profiles verge significantly from nominal loads because the same
(8760 hourly loads) were developed for each load location loading algorithm was applied to all feeders. This has a sig-
using three steps: nificant impact on PV penetration results and contributes to
1) An appropriate averaged load profile for each load class some anomalous results discussed later in this paper; for
(commercial and residential) was obtained from a utility example, feeder R1-12.47-3 has relatively low load so its
in the feeders geographical region, as defined in [14]. For maximum PV penetration as calculated from (1) is skewed
region 5, comprising the far Southeast near the Gulf of high.
Mexico, no load profile information was publicly avail- 3) To reflect the stochastic nature of real loads, the load pro-
able, so load profile data from neighboring region 4 was files were then randomized. Each of the 8760 scaled annual
used. Commercial loads were assigned a load profile from load values for each load point was multiplied by a factor
the utilitys commercial or industrial load class, and resi- sampled from a Gaussian (normal) distribution. The stan-
dential loads were assigned a load profile from the residen- dard deviation of the Gaussian distribution was 0.2, and its
tial load class. mean value was 1.0. This resulted in a load diversity factor
2) The load profiles were then scaled by a feeder-specific of about 1.3 for each load class. Each transformer feeds
scale factor and by the transformer capacity at each exactly one load, indicating that the creators of the feeders
load point. In this study, is defined as the ratio of peak may have consolidated the feeder loads. To account for this
apparent load power to total rated load transformer ap- consolidation, a nominal residential load size of 5 kVA was
parent power. For each feeder, was determined by run- assumed, and randomization was applied to each 5-kVA in-
ning multiple yearly simulations with one-hour time steps, crement for residential transformers. No typical load size
varying . A bisection search was used to find the max- was assumed for commercial transformers because com-
imum for each feeder that maintained load voltages mercial loads vary widely in size. Hence, the amplitude of
within ANSI Range A, line currents below their respective random hour-to-hour fluctuations in commercial loads is
continuous current ratings, and fuse currents below fuse much larger than for residential loads, and also larger than
amperage ratings. Fig. 1 shows the maximum scaled load 1The nominal feeder loads used here are as given in [14, Table 10] and are
found by simulation for each feeder alongside its nominal the peak planning loads [19].
352 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 4, NO. 2, APRIL 2013
TABLE I
MAXIMUM PV PENETRATION FOR ALL DISTRIBUTED PV SCENARIOS (IN % OF PEAK LOAD)
9) single PV system at a randomly selected point. where the PV penetration reached higher than 100% of peak
To site the single PV system in scenarios 6, 7, and 8, all load, the step size was increased to 50%. A bisection search was
three-phase nodes on the circuit were sorted by distance from then conducted to define the maximum PV penetration more
the feeder source and the PV system and its transformer were narrowly. The resolution of the bisection search for PV pen-
installed at the location that best fit the desired scenario. For etrations below 150% was 5 percentage points. Above 150%
example, to produce scenario 8, a single PV system and trans- penetration, the resolution began rising based on the assumption
former were installed at the farthest three-phase node from the that, for these very high penetration cases, the exact maximum
source. For scenario 6, the node where the PV system was in- penetration is not as relevant as the insight that maximum PV
stalled was required to be at least 300 feet from the source to penetration is not limited by steady-state voltage or current. The
ensure that the system was downstream from all substation com- maximum search resolution was 30 percentage points.
ponents. The maximum PV penetration results listed here have been
To produce scenario 9, the PV system and transformer were spot-checked for power flow solution errors using hand anal-
placed at a randomly selected three-phase node, again with the ysis and automated error-checking. No errors were found. While
requirement that the node be at least 300 feet from the feeder it would be ideal to verify each result via theoretical analysis,
source. doing so is not practical given the complexity of the feeders,
For each of these scenarios, the PV system was sized so that which contain hundreds of nodes and thousands of components
it alone would meet the desired PV penetration level. The trans- simulated at 8760 time points for each scenario. Efforts are un-
former was sized to have a rated apparent power 20% larger derway to experimentally verify the effects of high PV pene-
than the PV system peak rated power. During simulation of trations, determine maximum allowable penetration levels, and
scenarios 69, current was monitored in the line immediately verify models in high penetration scenarios [20][23]. These
upstream from the PV system and compared to the continuous studies focus on specific geographic locations or feeders. In
summer current rating of the line. Voltage was monitored at the physical experiments, it is not practical to test the wide variety
PV system point of common coupling (and as in all scenarios, of PV location and penetration level scenarios tested here. We,
the voltages at all loads and the states of all fuses were moni- therefore, must rely on the accuracy of GridLAB-Ds power
tored). flow solutions and the correct construction of the taxonomy
All PV systems simulated were installed at locations chosen feeders, which have been thoroughly verified [14], [15].
by a computer algorithm that, to avoid introducing bias, did
not evaluate the suitability of the locations for PV. For the dis- III. DISTRIBUTED PV RESULTS AND ANALYSIS
tributed PV cases all load points were considered, and for the
single PV system simulations all three-phase nodes were con- This section describes simulation results for the distributed
sidered. On real feeders distribution engineers typically avoid PV location scenarios (scenarios 15). Scenarios 1 through 4
allowing significant PV interconnection in poor locations, so were simulated once for each feeder, and the random location
some of the lowest maximum penetration limits found in this scenario was simulated five times for each feeder.
study may be overly conservative. Table I lists the maximum PV penetration results for the var-
ious distributed PV scenarios. The number preceded by R
D. PV Penetration Levels in each feeder name indicates its geographical region. Values
of maximum PV penetration obtained vary widely among the
For each of the nine scenarios and 16 feeders, GridLAB-D feeders, reflecting the diversity of physical feeder configura-
simulations were performed at PV penetrations of 0% (the base tions. Maximum PV penetration also varied widely from one
case), 15%, 30%, and so on, with 15% steps until an overvoltage location to another on most feeders. All feeders tolerated at least
or overcurrent was detected at any point during the year. In cases 75% PV penetration in at least one case.
354 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 4, NO. 2, APRIL 2013
Fig. 5. Maximum PV penetration for all nonrandom, distributed location scenarios. Columns with an asterisk (*) above indicate cases where PV penetration was
limited by overvoltage; otherwise PV penetration was limited by overcurrent. One result over 300% is truncated.
TABLE II
MAXIMUM PV PENETRATION FOR ALL SINGLE-SYSTEM SCENARIOS (IN % OF PEAK LOAD)
Fig. 7. Maximum PV penetrations for the 16 taxonomy feeders with single PV systems. Columns with an asterisk (*) above indicate cases where PV penetration
was limited by overvoltage; otherwise PV penetration was limited by overcurrent. Results over 500% are truncated.
Fig. 8. Maximum PV penetrations plotted against conductor distance from PV system to feeder source, by geographic region.
PV systems becomes too large when any of the many affected requirement that the system be installed at a three-phase node
points on the feeder experiences overvoltage or overcurrent; the increases the probability of selecting a node near the trunk of
variation from one group to another is smaller due to aggrega- the feeder where voltage is less likely to be an issue.
tion effects. Simulations with single PV systems provide an opportunity
Compared to the distributed PV scenarios, more feeders to examine the relationship between maximum PV penetration
showed minimum PV tolerances of less than 30%. This is and distance from the PV system to the feeder source. This
expected; putting a single PV system sized at more than 30% relationship is shown in Fig. 8, where distance was measured
of the entire feeders peak load on a small branch will cause by following the conductors back to the source. This shows a
voltage and/or current problems. definite trend of decreasing PV tolerance with distance. There
The relationship of maximum PV penetration to maximum are, however, many exceptions to the trend. These exceptions
baseline meter voltage was analyzed (Fig. 6). Again, the trend are largely because some PV systems were installed on smaller
is for PV tolerance to decrease with increasing voltage, although feeder branches by the intentionally unintelligent PV location
there are more outliers than in Fig. 4, reflecting the high case-to- algorithm.
case variation in single PV system scenarios.
Fig. 7 shows the maximum PV penetrations for each feeder V. CONCLUSION
under each nonrandom location scenario. Results over 500% are In this paper, several trends have been noted when consid-
truncated to provide better resolution for more realistic pene- ering maximum PV penetration relative to steady-state voltage
trations. For single PV systems, more scenarios are limited by and overcurrent: For distributed PV systems, maximum PV pen-
current than in the distributed PV scenarios. This is because the etration was nearly always above 50% unless the feeder already
HOKE et al.: STEADY-STATE ANALYSIS OF MAXIMUM PV PENETRATION LEVELS 357