Content uploaded by Jan Kleissl
Author content
All content in this area was uploaded by Jan Kleissl on Oct 18, 2018
Content may be subject to copyright.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
1
High PV Penetration Impacts on Five Local
Distribution Networks Using High Resolution Solar
Resource Assessment with Sky Imager and Quasi-
Steady State Distribution System Simulations
Andu Nguyena1, Maxime Velayb, Jens Schoenec, Vadim Zheglovc, Ben Kurtza, Keenan Murraya,
Bill Torrea, Jan Kleissla
a UC San Diego, 9500 Gilman Drive, San Diego, CA 92093, USA.
b Grenoble Institute of Technology, ENSE3, Grenoble, France.
c Enernex,620 Mabry Hood Rd NW # 300, Knoxville, TN 37932, USA.
Abstract
Some potential adverse impacts of high photovoltaics (PV) penetration on the power grid are an increasing number of tap
operations, over-voltages, and large and frequent voltage fluctuations and PV power ramps. The ability to create realistic PV
input profiles with high spatial and temporal resolution is crucial to assess these impacts. This paper proposes a unique method to
improve the accuracy of feeder hosting capacity studies using (1) high resolution PV generation profiles from sky imagers, (2)
quasi-steady state distribution system simulation, and (3) distribution models created from utility data. Solar penetration levels,
defined as ratio of peak PV output to peak load demand, from 0% to 200% and various cloud conditions are considered. Three
conclusions were drawn: (1) the impacts of high PV penetration depend on feeder topology and characteristics; (2) the use of a
single PV generation profile overestimates the tap operation number up to 260% resulting from an overestimation of power ramp
rates and magnitudes- therefore, multiple realistic profiles should be used; and (3) distributed PV resources increase the feeder
hosting capacity significantly compared to a centralized setup.
Keywords: PV impact; high PV penetration; solar resource assessment; power system simulation
1. Introduction
A common concern with renewable resources is the effect of weather on power production. For systems with natural resource
storage, such as large hydroelectric plants, this concern is for time scales of weeks and months. For PV, the concerns and effects
are on much shorter timeframes due to intermittent disturbances by passing clouds. While it is possible to forecast weather with
reasonable accuracy and predict average power production across a region quite accurately, individual clouds can cause sudden
disturbances in PV production. For solar PV interconnected at the distribution feeder level, addressing the intermittent
disturbance in solar power output is a key technical challenge for ensuring power quality and further increasing the penetration
level of PV. Recent studies [1, 2] have reviewed the potential issues with high PV penetration.
Although there is significant interest in the potential issues of high PV penetration, investigations with high resolution and
realistic PV generation profiles are lacking. Most studies either use a single solar generation profile for all PV systems or low
resolution profiles [3, 4, 5], assuming the same and coincident power ramps. However, short term PV ramp rates have been
shown to become uncorrelated over distances as short as a few hundred meters [6]. Therefore, applying a single or a few PV
profiles to all systems leads to unrealistic results and does not reflect the nature of the distributed generation and its geographic
diversity effect. This leads to erroneous assessments of wear-and-tear on voltage regulation equipment caused by solar power
variability.
* Corresponding author. Tel.: +1-202-361-9049.
E-mail address: andunguyen@ucsd.edu.
*Unmarked Revised Manuscript
Click here to view linked References
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
2
We propose a unique approach to overcome this limitation using (1) solar resource assessment with a sky imager to provide
high temporal and spatial resolution of PV generation profiles, (2) realistic and topology-diverse distribution models created
from data provided by a host utility in the U.S., and (3) quasi-steady state distribution power simulation with OpenDSS [7]. The
generation levels of the PV generators are individually determined from solar resource assessment with sky imagers and
irradiance data taken from field measurements. This technique for generating high-resolution individual PV profiles is a
significant improvement over typical PV system impact studies.
For our simulation work, OpenDSS was selected due to its exceptional capabilities in performing quasi-steady state (QSS)
simulation for distribution systems. Some alternative tools are GridLab-D, Matlab Simulink, CYME, and EDD. OpenDSS and
Gridlab-D are two tools that allow scripting for running many different time series simulations with minimal computing time.
Between OpenDSS and GridLab-D, OpenDSS was chosen due to the easier integration with Matlab and our solar forecasting
engine.
This paper is organized in six sections. After the introduction, the feeder models and data used are described in Section 2.
Section 3 provides detail on the approach for achieving realistic PV generation profiles. Section 4 provides a summary of the
scenarios that were run for investigating various impacts of high PV penetration levels and the benefits of using such PV
profiles. Section 5 discusses the results of the simulations and their implications. It is then followed by concluding remarks in
Section 6.
2. Feeder Models and Data
2.1. Five Distribution Feeder Models
System information for five distribution feeders were provided by the host utility in SynerGEE including: (1) network
topology: information on buses, equipment settings, and the equipment connected to the buses, (2) equipment characteristics:
specifications of capacitors, voltage regulators, and other utility equipment, and (3) the locations and characteristics of the loads.
This information was used to create equivalent distribution models in OpenDSS. Short circuit currents and power flow results
from SynerGEE simulations were provided and used as a benchmark to validate our OpenDSS models. Table I summarizes the
characteristics of these feeders. Feeders A, D, and E are in rural areas. Feeders B and C are in urban areas. Figs. 1-5 show the
topologies of these five feeders. It should be noted that the five feeders investigated here are specific to the San Diego area and
might not be representative of the feeders in other US regions. Despite that fact, we believe that the conclusions drawn especially
for line losses and overvoltages are generalizable, while the number of voltage regulator actions is coupled to the feeder
hardware and operating procedures of a specific utility.
Among the five feeders, Feeder A is the longest feeder. In addition, it has two large PV systems at the end of the circuit and
the largest peak load demand. Thus, it is the most challenging feeder to accommodate high PV penetration and is investigated in
more detail in Sections 4 and 5. Fig. 6 shows the large drop in voltage along Feeder A at max load demand and no PV. This
violates the lower ANSI limit of 0.95 p.u. [8]. The large voltage drop is partly due to the peak loading condition in this case
which maxed out the voltage regulators’ capability to regulate the voltage levels, limiting their ability to reduce the volta ge drop.
This would not occur in practice since it is unlikely that all loads operate at max demand simultaneously. At a lighter loading
condition where the voltage regulators’ capability is not maxed out, the voltage drop is less severe. The daily load demand is
usually between 30% and 70% max load without PV as observed from the data given at the substation of these feeders over the
period of one year.
TABLE I
CHARACTERISTICS OF FIVE FEEDERS
Circuit
A
B
C
D
E
Feeder length (km)
177.8
39.6
34.9
51.5
115.7
Number of Loads
2246
3761
1466
471
1169
Total peak load (MW)
11.1
8.3
4.8
3.7
6.7
Number of Capacitor banks
2
2
2
1
2
Num of Transformers & VRs
7
3
1
1
2
2012 Number of PV systems
45
85
28
19
43
2012 Peak PV output (MWAC)
2.3
1.3
2.1
0.2
0.3
2012 PV penetration (%)
18
15
58
4
4
Total capacity of large PV
systems >0.5MW (MWAC)
2.0
0
2.0
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
3
Fig. 1. Feeder A with 11 MWDC peak load and 45 PV systems totaling 2.3 MWDC peak production. There are two large 1-MWDC PV sites at
the end of the feeder, which cause severe issues, for example large voltage drop, high voltage fluctuations, increased tap operations, etc. The
size (with regards to rated output power) of the PV sites can be gleaned from (1) the color, based on the logarithmic color scale in the figure,
and (2) the size of the circle with the area of the circle being proportional to the log of the rated output power.
Fig. 2. Topology of the feeder B.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
4
Fig. 3. Topology of the feeder C.
Fig. 4. Topology of the feeder D.
Fig. 5. Topology of the feeder E.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
5
Fig. 6. Normalized unbalanced 3-phase voltage profile of feeder A simulated at 11.1 MWAC max load demand. X-axis shows distance (in
terms of the line length) from the investigated buses to the substation (shown in Fig. 1).
2.2. Variable PV Penetration
Existing commercial (10 kW-1 MW) and residential (1 kW-30 kW) sized PV generators were provided from the host utility in
the form of one-line diagrams with PV site locations and PV ratings. The PV locations indicated in the respective one-line
diagram were mapped to the locations of the buses specified in the five feeders. Fig. 1 illustrates feeder A with 45 existing PV
systems. The locations of the mapped PV sites are depicted by circles. Feeder A has two large 1 MW PV plants and Feeder D has
a 1.13 MW PV plant. Feeder B has 8 commercial-sized PV systems ranging from 70 kW to 174 kW, totaling 909 kW. The
remaining PV systems are residential with an average size of 6 kW. Table I provides some information on the existing PV
systems on these feeders. The locations and sizes of PV systems are shown in Figs. 1-5.
PV penetration level is calculated as:
(1),
where
and
are the rated peak total production of the PV systems and rated peak total consumption
of the loads in the feeder, respectively. Other definitions for PV penetration level that could be used are:
(2),
and
(3),
where
and are maximum daily PV power production and maximum daily load’s
power consumption; while and are PV’s total daily energy production and
load’s total daily energy consumption respectively. Consequently, and are subjected to change from one day
to another while is a constant for a feeder. is dependent on the total energy produced by PV and consumed by
loads during the day while depends only on the maximum power produced by PV and consumed by loads on that day. No
matter what definition is used, the conclusions made from the results are valid for all cases. In this work, the first definition of
in (1) is used.
As of 2012, based on California Solar Initiative incentive data, the PV penetration of the five feeders ranges from 4% to 18%
except from feeder C with 58%. While no simulations with the real 2012 fleet were conducted, the 2012 fleet served as the initial
population of systems to create the different PV penetrations scenarios. To simulate future growth of PV penetration levels for
feeder A, additional PV systems were created by duplicating the specifications of the 43 small (<0.5 MW) existing systems and
moving them randomly to non-PV loads. Fig. 7 shows the existing as well as simulated PV systems on Feeder A. The locations
of the artificial PV systems are randomized. Large and medium-sized PV generators were connected three-phase, while the
typical connection for the small-sized PV generators was single phase. Furthermore, the small-sized, mostly rooftop-mounted PV
sites connect directly to a bus that has a residential load connected.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
6
Fig. 7. Feeder A with 45 original PV systems and 387 additional virtual distributed PV systems created for the simulation.
2.3. Solar Irradiance Data
A network of irradiance sensors or pyranometers at UCSD collects 1 sec irradiance data. This data is normalized by expected
clear sky irradiance to yield clear sky indices, kt, that are proportional to cloud optical depth [9]. Cloud conditions are classified
into three representative categories (Fig. 8): clear sky (large generation, low variability, Dec 19th), partly cloudy (medium
generation, large variability, Dec 14th) and overcast (low generation, low variability, Dec 18th). We investigated the impacts of
these cloud conditions on the distribution feeders. Videos showing sky conditions of these three days are at [10, 11, 12] . Since
local measurements at each feeder were not available, the sky images and pyranometer measurements from La Jolla, CA were
spatially translated to the feeder locations. While strong spatial gradients in climate exist in coastal California, winter conditions
are usually determined by synoptic weather patterns and are spatially homogeneous. Thus, utilizing the same irradiance data on
all feeders provides better insights into the relative importance of feeder characteristics on PV penetration impacts.
Fig. 8. Global Horizontal Irradiance (GHI) profiles for the three investigated days in 2012 with different cloud conditions measured in La
Jolla, California.
2.4. Load Shape Profile
The net power consumption at the substations of all five feeders in 2011 and 2012 were provided by the host utility at 15 min
time steps. The 15 minute data is interpolated to provide 30 second load profiles as simulation input. Since the focus of this paper
is on the PV generation profiles and impacts and distribution feeder topologies, a single representative normalized daily load
profile of feeder B on Dec 18th 2012 was used in all simulations (Fig. 9). The Dec 18 load profile peaks at 70% peak load which
is also the overall peak load of the feeder observed in the two year net load data set. The profile represents the typical duck-shape
profile of California wintertime residential power consumption, resulting from large power consumption by residential heating,
cooking, and lighting in the early evening. The load profile of the overcast day Dec 18 is chosen to minimize the impact of PV
output on the net load profile so as to mimic a true load profile at 0% PV penetration. The effect of load disaggregation or
variation is not considered to limit feeder impacts to different PV penetration levels. Load variations were investigated in [13].
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
7
Fig. 9. A typical normalized daily load profile on the urban Feeder B was used for all simulations on all feeders. This load profile is the
aggregated load measured at the substation on the overcast day Dec 18th 2012 and was provided by the host utility. The impact of PV output
can be seen at noon despite the overcast condition.
3. High Resolution Distributed PV Generation Profiles Using Sky Imagers
3.1. Cloud Map and Shadows
The UCSD sky imager (USI), a high resolution fisheye-lens sky camera, captures an image of the sky every 30 seconds [14].
Each image is processed to determine the location of clouds within the image, which are then mapped into a sky grid. The sky
image covers a circular sky area with a radius ranging from less than a kilometer to tens of kilometers, depending on the height
of the clouds. Cloud height obtained using multiple sky-imagers [15] is utilized.
The shadows of the clouds are determined using ray-tracing based on the sun location to create a shadow map, which shows
shaded and unshaded regions on the feeder. Fig. 10 shows a snapshot of high resolution cloud shadows on feeder A on Dec 14,
2012. A video of cloud motion for the whole day is at [11]. Note that in the snapshot more than half of the feeder area is covered
by clouds, while the remaining area is clear. Due to the fish-eye nature of the camera lens, the spatial resolution of a given area
on the shadow maps depends on its distance to the center of the sky image. The resolution is approximately 10 meters near the
center and approximately 100 meters near the edge of the sky image. More analysis on this can be found in [15].
The ray tracing technique has been adopted due to its simplicity and easy implementation. Other techniques could be
employed. However, since the ray tracing technique takes only less than 1% of the total computing time for the whole
forecasting/ resource assessment process, we did not feel the need to improve the ray tracing technique further nor looked for an
alternative method. The cloud detection and movement analysis are the most computationally expensive operations as well as
being most prone to errors. They are currently being worked on for further improvements in speed and accuracy. Future papers
will be published on these topics upon completion.
Combining the shadow maps with the known locations of the PV systems on the feeder allows us to determine every 30
seconds whether or not a given PV generator is shaded by a cloud. Unshaded areas on the shadow map receive a high level of
irradiance and cloudy areas receive a lower level of irradiance that depends on the optical depth of the cloud. Unshaded areas do
not necessarily receive 100% of clear sky irradiance due to atmospheric effects such as aerosol scattering, while shaded areas can
receive considerably more than zero irradiance due to considerable diffuse horizontal irradiance even on cloudy days. Fig. 10
shows clouds covering half of the feeder. Although Fig. 10 only shows clear or cloudy levels for visualization purpose, we
actually consider three levels of cloudiness: clear sky, thin cloud, and thick cloud. More detail on these levels is explained in the
next section. We applied a complex scheme to estimate the opacity of the clouds and detail can be found in [16].
Because significant spatial differences in irradiance exist, it is crucial to determine the unique solar irradiance profile for each PV
site to analyze the impact of high PV penetration.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
8
Fig. 10. A snapshot of high resolution cloud shadows over Feeder A on Dec 14. More than half of the feeder area is covered by clouds. Full
video is at [17].
3.2. PV Profile Generation
Solar resource assessment and forecasting using sky imagers was detailed in our previous works in [9] and [16]. Each pixel in
a sky image is categorized into one of three types according to three typical cloud conditions: no cloud (or clear sky), thin cloud,
and thick cloud. This categorization is used to convert the shadow map to irradiance levels by calculating the clear sky index, kt,
defined as the amount of irradiation that reaches the ground for each cloud condition. For a given time step, kt is calculated from
a histogram of measured irradiance data obtained directly from GHI sensors over the past two hours. The measured value is
normalized with respect to the clear sky irradiance level, which is determined using a clear sky irradiance model for the
atmospheric condition at the location of interest. From the histogram, three values of irradiance at three peaks are chosen as
representative of clear sky, a thin cloud, and a thick cloud.
A global horizontal irradiance (GHI) profile yields information about the irradiance level at a given location as a function of
time. We calculated GHI profiles for each PV location from the series of cloud maps captured by the sky camera and the
irradiance data measured at the irradiance sensor locations. Specifically, we calculated the ) level at each time step as
follows:
Determine the cloud condition (clear/thin/thick) for each ground pixel using the cloud map information. While Fig. 10
shows only clear or cloudy levels for visualization purpose we actually consider three levels of cloudiness: clear sky, thin
clouds, and thick clouds [11].
Calculate the clear sky index, , from histograms of the preceding 2-hours of GHI measurements. is assigned based on
the peaks in the histogram within three ranges: clear sky (0.9-1.1), thin cloud (0.6-0.9), and thick cloud (0.1-0.5) [18]. If
some cloud conditions are not present during the previous two hours (e.g., if the sky has been clear during the last two
hours), then default clear sky indices are used- 1.06 for clear sky, 0.7 for thin cloud and 0.42 for thick cloud. The kt value in
clear conditions is larger than 1 due to due to a low bias in the clear sky model and irradiance enhancement.
Calculate the weighted clear sky index () for a PV system as a spatial average of values of all pixels in the footprint of
that PV system as
, where n is the total number of pixels constituting the footprint of the PV system.
Calculate GHI for each PV as
, where
is the clear sky GHI at time and was determined from the
clear sky model described in [19].
Transpose
to the global irradiance at the plane-of-array of the PV system considering tilt and azimuth angles . The
plane-of-array clear sky global irradiance GIclear is then determined by computing GHIclear using the Ineichen-modified
Kasten clear sky model [19], breaking up the GHI into direct and diffuse following Boland et al. [20], and then transposing to
the plane-of-array using the Muneer transposition model as described by Page [21]. PV system tilt and azimuth used to
calculate are given from California Solar Incentive Data recorded on Dec. 31st 2014. Tracking systems did not exist on
the feeders, but could easily be simulated with our approach by varying tilt and/or azimuth dynamically.
The representativeness of the sky imager GHI profiles was validated using the variability index (VI) metric [22]. In a
population of 90 days from December 2014 to March 2015, the daily aggregate VI from the sky imager GHI compared to VI of
30 sec average GHI from ground stations produced statistics of correlation coefficient 0.91, mean bias of 0.12, and mean
absolute error of 2.0. A rigorous assessment on the accuracy of our solar forecasting method using sky imagers was previously
conducted in [16].
Finally, the power output of the PV system
is obtained from a power conversion model [23]. The model considers cell
temperature effects on PV panel efficiency as a function of wind speed and ambient air temperature. Further, inverter efficiency
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
9
is expressed as a function of the power factor [23]. Hourly wind speed and ambient temperature data are from NOAA’s Quality
Controlled Local Climatological Data stations closest to each feeder.
Fig. 11. Comparison of the irradiance measurement (black) of one ground station and the modeled irradiance (green) using sky imagers on Dec
14th.
Fig. 11 demonstrates that the model based on sky imager cloud maps yields realistic GI time series. The GI measured by the
ground sensor is slightly more variable for three reasons: (1) Our resource assessment method uses three discrete levels of
irradiance output, whereas the range of actual irradiance output measured by the sensors is continuous and kt can exceed 1.1
during short-lived cloud enhancement events that are of little relevance to PV integration issues; (2) errors in cloud detection
algorithms cause, for instance, a thin cloud to be classified as clear sky leading the forecast to miss a ramp; and (3) the irradiance
sensor covers a ground area of only about 10-4 m2, while the finest resolution achieved by our sky imager method is 10 m by 10
m. Thus, our method averages the generation output on a scale commensurate with the typical footprint of an individual
distributed PV system. For small PV systems the sky imager solar resource assessment directly yields an appropriate solar
irradiance. For large PV systems the sky imager solar irradiance is averaged over the spatial extent of the system to obtain a
smoother signal. Fig. 12 shows a two-hour GHI profile for three different PV sites. The three GHI profiles are quite different due
to their different locations demonstrating that the geographic diversity effect leads to smoothing of aggregate power ramps.
Fig. 12. GHI profiles at 3 different sites (see Fig. 1 for locations) on Dec 12th.
3.3. Clouds Shading on Part of a PV Array
The ground area occupied by the PV system depends on its DC power rating; that is, PV systems with a higher rating are
generally made up of more PV panels that cover a larger area as compared to those systems with a smaller rating. At a given
time, some area of a PV array may be shaded by clouds while another part of the same array may be unshaded, a scenario that is
more likely to occur on large PV systems due to their larger footprint. The output of a large PV system will not be interrupted
immediately when a small cloud begins to pass over and gradually shades the array. Instead, there will be an averaging effect and
the cloud-caused variation of the PV power output will be smooth rather than sharp or step-like. On the other hand, the time
frame between maximum generation (clear sky) and low generation (array completely shaded by clouds) tends to be much
shorter for arrays with a smaller area as it takes less time for a cloud shadow to cover the entire footprint. The finest resolution of
the cloud maps is 10 m by 10 m for each pixel in the map. The site with the two 1 MW PV systems located at close proximity
has a combined estimated footprint of 9 pixels by 18 pixels and is the largest on the feeder. Partial shading of this and other sites
was considered in each simulation time step using the following methodology:
estimating the size of the PV area based on its DC rating,
establishing the footprint of the PV system based on its size and location,
determining the shaded and unshaded areas of the PV footprint based on the cloud map,
calculating the power output for the shaded and unshaded areas of the PV footprint based on the GHI profile for this
location, and
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
10
calculating the total power output for each PV system by summing up the power outputs from the PV system's shaded
and unshaded areas.
The power generated at a PV site with a large footprint will be smoothed somewhat, but the area of the PV site is small
compared to the area of the whole feeder. Consequently the smoothing effect will be more significant when considering the
effect of clouds on the total feeder PV generation [6].
4. Simulation Scenarios and Setup
The resource assessment technique outlined in Section 3 was used to produce solar power data for simulations of the five
distribution models in Section 2.1 with different scenarios:
(1) 5 Feeders: Simulations with all five feeders to study the dependence of PV impacts on feeder topology. Multiple
individual PV profiles are used.
(2) PV penetration levels ranging from 0% to 200% at 25% increments are simulated.
(3) Daily Simulations with Different Cloud Conditions: For each of the above setups, 24 hour simulations with 30
second time step were simulated with three cloud conditions (Section 2.3).
(4) ‘X-single’ and ‘X-multiple’ PV profiles: To investigate the impact of using multiple PV profiles versus a single
common profile for all PV systems, two special configurations were set up: (1) ‘X-single’ configuration (with X as one
of the five feeders, A-E) with a single PV profile; and (2) ‘X-multiple’ configuration with multiple PV generation
profiles applied specifically to each system using the method of Section 3. The difference between the single versus
multiple PV profiles manifests in the substation net power variability (Fig. 13).
(5) Special ‘A-centralized’ and ‘A-distributed’ configurations: To investigate the impact of a large, centralized PV
allocation versus small, distributed PV systems, two special configurations were set up for feeder A: (1) the ‘A-
centralized’ configuration with 45 existing PV systems including two 1 MW sites (Fig. 1) and (2) a ‘A-distributed’
setup with 430 small PV systems, but the same PV penetration (Fig. 7). The simulation for the ‘A-centralized’
configuration does not converge when the PV penetration exceeds 100% due to the thermal limitation of the line
conductor so results above 100% penetration are not shown. Multiple PV profiles are used for both setups.
Fig.13. Net power consumption at the substation for A-single and A-multiple on Dec 12th. The use of multiple PV profiles reduces the
frequency and magnitude of net power fluctuations.
Thus, the total number of simulated scenarios is (5 + 2 + 6) scenarios x 5 penetration levels x 3 days = 195. The following
clarifies the configurations used for feeder A. The ‘A-mixed’ configuration is used in scenarios 1, 2, and 3 with 430 small and
two 1 MW PV systems (Fig. 7). Since 87% of the original 2012 PV capacity is concentrated in two co-located 1 MW systems,
the original configuration is not conducive to comparing single and multiple solar resource inputs. Thus for ‘A-single’ and ‘A-
multiple’ configurations, only 430 small PV systems are simulated. The ‘A-centralized’ configuration consists of the original 43
small and two 1 MW PV systems while the ‘A-distributed’ configuration uses the 430 small PV systems like in the ‘A-single’
and ‘A-multiple’ cases.
5. Results and Discussion
5.1. Dependence on Feeder Topology and Characteristics
Comparing simulation results of all five feeders at different PV penetrations from 0 to 200% yields the following insights:
Distribution Line losses (Fig. 14): As PV penetration increases from 0% to 100%, the line losses decrease as local energy
demand is met by local PV output. As PV penetration passes 100%, the losses increases again because the PV power output
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
11
exceeds local demand, starts to feed loads further away, and then feeds back to the substation. For all feeders, the loss at 200%
PV penetration is larger than at 0% due to large amount of reverse power flow from excess PV power output.
Fig. 14. Normalized line losses with increasing PV penetration relative to the base case of 0% PV penetration on the clear day Dec 19th, 2012.
Voltage excursions: For this section, we focus the discussions on daylight hours when PV impact is present. Fig. 15 shows
the max and min voltages of all the buses on the feeder at every time step for the 0% and 200% PV penetration cases. It can be
observed that PV active power output has raised and shaped the min and max voltage lines. The trough in voltage during night
time in Fig. 15 was due to large residential loading in the evening and the fixed substation voltage of 1.0 p.u. used in this
particular simulation. It should be noted that the fixed 1.0. p.u. substation voltage setting is used only for demonstration of the
impact of different loading conditions on feeder voltage. In reality and the ensuing simulations the voltage regulators are
permitted to change taps to keep the voltage levels in the feeder within the ANSI limit range. For example, Fig. 18 shows that
when PV generation is significant, the voltage regulator will step down the voltages while stepping up at night time and large
loading conditions.
Fig. 16 shows the max and min voltages throughout the day for all feeders at different penetration levels. The voltage is as
high as 1.15 p.u. for 200% PV penetration for feeder A which is much higher than the normal ANSI limit of 1.05 p.u. Feeder A’s
voltage exceeds this limit even at 100% penetration. On the other hand, voltage ranges for feeders B, C, and D are always within
the allowable range of 0.95-1.05 p.u. Out of all factors that could contribute to the differences in voltage spread, which is
defined as the difference between the max and min voltages, among the feeders, the total feeder length has the strongest
correlation with the voltage spread (Fig. 17). This is the result of power flow causing larger voltage drop over a longer line when
impedance is similar. Therefore, the feeder hosting capacity, defined as the maximum level of PV penetration at which the
voltage spread is in a permissible range, is expected to correlate strongly with feeder length.
Fig. 15. Maximum and minimum voltages recorded on feeder A-mixed during the partly cloudy day (Dec 14) with fixed voltage setting at
substation.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
12
Fig. 16. Voltage ranges during daylight on the partly cloudy day Dec 14 between 0700 to 1600 PT for all feeders. The timestamps above and
below the bars show when the extremes occur.
Fig. 17. Difference between max and min voltage in Fig. 14 versus feeder length (Table 1). There is strong correlation between feeder length
and voltage range with a correlation coefficient of 0.97.
Tap operations: Transformers and voltage regulators are programmed to change their tap position to keep voltage levels
along the feeder within permissible limits. The voltage regulators are simulated in OpenDSS with a voltage bandwidth of 2 volts
and a time delay of 45 seconds. Other parameters are kept at default. Fig. 18 shows that transformer #2 decreases the voltage
during the daytime and increases the voltage in the evening during the net load peak. Fig. 19 shows the change in the number of
tap operations for each feeder on the partly cloudy day. Feeders A and E experienced high increases in the number of tap
operations due to large voltage ranges (Fig. 16). There is no change in tap operations on feeder B due to its small voltage range.
For feeder D, the only change is from 80% to 100% PV penetration and three extra tap operations are sufficient to manage 200%
penetration.
Fig. 18. Normalized voltage step down ratio for secondary transformer #2 (Fig. 1) on rural Feeder A on Dec 14th, 2012.
Fig. 19. The tap operation number on the partly cloudy day normalized by the 0% PV case. At 0% PV, feeders A, B, C, D, and E have 219, 0,
7, 0, and 15 operations. Feeder D shows 3 tap operations for penetration levels >= 100%.
5.2. Single vs. Multiple PV Irradiance Profiles
The use of the single profile causes PV systems over the entire feeder to ramp synchronously, which leads to more severe
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
13
power fluctuations. This is unrealistic, especially for feeders with large footprints as the irradiance correlation between sites
decreases exponentially with distance. To quantify and compare the fluctuation in voltages between the two configurations, we
calculate the voltage volatility as standard deviation for all five feeders (Fig. 20). Except for feeder A, the voltage is consistently
more volatile when using a single PV profile. The abnormality of volatility on feeder A is due to the large number of voltage
regulators, which could operate differently in response to various voltage fluctuating scenarios. The single PV profile
configuration also overestimates the line losses (Fig. 21), except for feeder A at 25-100% penetrations. The overestimation
amount is significant at high PV penetration levels, e.g. 3% overestimation on average at 200% penetration.
Fig. 20. Standard deviation of substation voltages of five feeders with single and multiple PV profile configurations on the partly cloudy day.
Fig. 21 Normalized line losses relative to the 0% penetration case for all five feeders with single and multiple PV profiles configurations on
the partly cloudy day.
Peak voltages generally occur at minimum net load around noon time at high PV penetrations. Since the partly cloudy day
does not contain a long clear sky period, the use of single PV profiles overestimates the maximum voltages for all five feeders
(Fig. 22). The overestimations for max voltages at 200% PV for the five feeders are 0.008, 0.0004, 0.001, 0.001, and 0.004 p.u.
respectively.
The primary effect of distributed PV profiles is the reduction of PV power ramp rates and ramp magnitudes which leads to a
reduction in voltage regulator operations and impacts the lifetime of voltage regulators and transformers. The tap operation
number on feeders A, C, and E (Fig. 23) decreases substantially. Feeders B and D require very few tap operations (0 and 3
respectively) even at 200% PV penetration so the different PV profiles do not make any difference. The use of a single profile
overestimates significantly the number of tap changes for days with clouds (i.e. partly cloudy and overcast conditions), but not
on clear days, as expected (Fig. 24). The overestimation for feeder A was up to 260% on the partly cloudy day. This shows the
unique feature of our approach in which we model the PV production output more realistically and estimate the actual tap
changing operations more accurately, especially in cloudy conditions.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
14
Fig. 22. Max voltage recorded from five feeders with single and multiple PV profiles configurations on the partly cloudy day.
Fig. 23. Normalized tap operations of five feeders with single and multiple PV profiles configurations on the partly cloudy day.
Fig. 24. Normalized tap operations of feeder A with single and multiple PV profiles configurations on three days with different cloud
conditions.
5.3. Distributed versus Centralized PV
The use of distributed PV systems reduces the power ramps significantly in both frequency and magnitude in comparison to a
centralized PV setup (Fig. 25). The distributed PV power output is smoothed out by summing the power of many small PV
systems that do not fluctuate synchronously. Thus, centralized PV on the feeder is much more prone to cause over-voltages due
to large power generation, in particular if voltage regulators are not strategically placed to accommodate the heterogeneous
concentration of PV on the feeder.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
15
Fig. 25. Net power consumption recorded at the substation at 100% PV penetration for ‘A-centralized’ and ‘A’ (distributed) configurations.
The higher the PV penetration level, the more severe is the voltage fluctuation caused by a centralized setup compared to a
distributed one (Fig. 26). The upper voltage limit of 1.05 p.u. is violated by the A-centralized configuration even at 100% PV
penetration while the A-distributed configuration does not violate this limit even at 200% PV penetration.
Fig. 26. Voltage range at different PV penetration levels of the centralized and distributed configurations of feeder A.
Furthermore, the A-centralized configuration significantly increases the number of tap operations in comparison to A-
distributed (Fig. 27). The difference is up to 230% at 50% PV. Lastly, Fig. 28 shows that while line losses decrease with
increasing PV penetration in the A-distributed configuration, it is not the same for the A-centralized case. The line losses
decrease initially but increase again after PV penetration reaches 50% in the A-centralized setup. Therefore, a distributed setup
would be preferable to minimize line losses.
Fig. 27. Comparison in number of tap operations between A-centralized and A-distributed configurations at different PV penetration during
daylight hours on Dec 14th.
Fig. 28. Total line loss during daylight hours of the centralized and distributed PV configurations at different PV penetration on Dec 14th.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
16
6. Conclusions
In this paper, we have proposed a new approach to study the impact of high PV penetration on a distribution network and its
hosting capacity. The proposed method combines high resolution resource assessment using sky imagery with power system
simulation on real distribution models to study the impacts of up to 200% PV penetration level on voltage excursions, line losses,
and tap changing operations. We capture the unique temporal variation in power generated by each PV system in a highly
realistic fashion.
We conclude that there are three main factors that are crucial in studying the PV impacts on the hosting capacity of a
distribution network:
(1) The topology and characteristics of the feeder, in particular the total feeder length. While feeder A has a hosting capacity
of 75% PV penetration, feeder E can host up to 200% and feeders B, C, and D can host even higher level of PV penetration. The
topology and characteristics, especially the feeder length, of a feeder strongly affect its hosting capacity (Section 5.1).
(2) The use of a single PV profile versus multiple profiles: Configurations with a single PV generation profile overestimate the
voltage volatility or deviation, maximum voltage levels, and especially the number of tap operations and power ramp rate and
magnitude. Thus, multiple PV generation profiles, which can be achieved either from real measurement data (which is not
always available due to large and expensive required infrastructure) or from solar resource assessment using sky imagery, should
be used (Section 5.2).
(3) The allocation of distributed resources: The concentration of PV systems on the feeder has a severe impact on increasing
power ramp rate and magnitude, voltage range, tap changing operations, and line losses. Thus, for planning purposes and to
increase the hosting ability of PV resource in a distribution network, the distributed allocation strategy is preferable to a
centralized one (Section 5.3).
For future work, we are designing various schemes utilizing solar forecasting, control of PV inverters, storage systems, and
optimization to mitigate the impacts of high penetration of distributed resources and to maximize the benefits of solar energy.
REFERENCES
[1]
M. ElNozahy and M. Salama, "Technical impacts of grid-connected photovoltaic systems on electrical networks—A
review," Journal of Renewable and Sustainable Energy, vol. 5, no. 3, p. 032702, 2013.
[2]
R. Passey, T. Spooner, I. MacGill, M. Watt and K. Syngellakis, "The potential impacts of grid-connected distributed
generation and how to address them: A review of technical and non-technical factors," Energy Policy, vol. 39, no. 10, pp.
6280-6290, 2011.
[3]
M. Thomson and D. Infield, "Impact of widespread photovoltaics generation on distribution systems," Renewable Power
Generation, IET, vol. 1, no. 1, pp. 33-40, 2007.
[4]
S. Ali, N. Pearsall and G. Putrus, "Impact of high penetration level of grid-connected photovoltaic systems on the UK low
voltage distribution network," in International Conference on Renewable Energies and Power Quality, 2012.
[5]
J. H. Enslin, "Network impacts of high penetration of photovoltaic solar power systems," in Power and Energy Society
General Meeting, 2010 IEEE, 2010.
[6]
M. Lave, J. Kleissl and E. Arias-Castro, "High-frequency irradiance fluctuations and geographic smoothing," Solar Energy,
vol. 86, no. 8, pp. 2190-2199, 2012.
[7]
R. C. Dugan and T. E. McDermott, "An open source platform for collaborating on smart grid research," in Power and
Energy Society General Meeting, 2011 IEEE, 2011.
[8]
A. N. S. Institute, "American National Standard For Electric Power Systems and Equipment - Voltage Ratings (60 Hertz),"
National Electrical Manufacturers Association, Virginia, 2011.
[9]
C. W. Chow, B. Urquhart, M. Lave, A. Dominguez, J. Kleissl, J. Shields and B. Washom, "Intra-hour forecasting with a
total sky imager at the UC San Diego solar energy testbed," Solar Energy, vol. 85, no. 11, pp. 2881-2893, 2011.
[10]
S. Lab, "Video of sky condition on December 14th 2012," July 2014. [Online]. Available:
http://solar.ucsd.edu/nguyen/20121214.mp4.
[11]
S. Lab, "Video of sky condition on December 18th 2012," July 2014. [Online]. Available:
http://solar.ucsd.edu/nguyen/20121218.mp4.
[12]
S. Lab, "Video of sky condition on December 19th 2012," July 2014. [Online]. Available:
http://solar.ucsd.edu/nguyen/20121219.mp4.
[13]
J. Schoene, J. Kleissl, V. Zheglov, J. Ning, A. Nguyen and B. Kurtz., "Improving economics of solar power through
analysis, forecasting, and dynamic system modeling.," Enernex Corp., Tennessee, 2013.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
17
[14]
B. Urquhart, B. Kurtz, E. Dahlin, M. Ghonima, J. Shields and J. Kleissl, "Development of a sky imaging system for short-
term solar power forecasting," Atmospheric Measurement Techniques, vol. 8, pp. 875-890, 2015.
[15]
D. A. Nguyen and J. Kleissl, "Stereographic methods for cloud base height determination using two sky imagers," Solar
Energy, vol. 107, pp. 495-509, 2014.
[16]
H. Yang, B. Kurtz, D. Nguyen, B. Urquhart, C. W. Chow, M. Ghonima and J. Kleissl, "Solar irradiance forecasting using a
ground-based sky imager developed at UC San Diego," Solar Energy, vol. 103, pp. 502-524, 2014.
[17]
A. Nguyen, "Demonstration of cloud overlay on feeder a on Dec 14th 2012.," July 2014. [Online]. Available:
http://solar.ucsd.edu/nguyen/cloudOnFallbrookFeeder520.mp4.
[18]
M. Ghonima, B. Urquhart, C. Chow, J. Shields, A. Cazorla and J. Kleissl, "A method for cloud detection and opacity
classification based on ground based sky imagery," Atmospheric Measurement Techniques, vol. 5, no. 11, pp. 2881-2892,
2012.
[19]
P. Ineichen and R. Perez, "A new airmass independent formulation for the Linke turbidity coefficient," Solar Energy, vol.
73, pp. 151-157, 2002.
[20]
J. Boland, Huang J and B. Ridley, "Decomposing global solar radiation into its direct and diffuse components," Renewable
and Sustainable Energy Review, vol. 28, pp. 749-756, 2013.
[21]
J. Page, in Practical Handbook of Photovoltaics: Fundamentals and Applications, Elsevier, 2003, pp. 5-66.
[22]
J. S. Stein, C. W. Hansen and M. J. Reno, "The variability index: a new and novel metric for quantifying irradiance and PV
output variability," in Word Renewable Energy Forum, Denver, CO, 2012.
[23]
M. Jamaly, J. Bosch and J. Kleissl, "A Power Conversion Model for Distributed PV Systems in California," Dept. of
Mechanical and Aerospace Eng., UCSD., San Diego, 2013.