You are on page 1of 13

Received: 30 September 2019 Revised: 20 December 2019 Accepted: 11 February 2020

DOI: 10.1002/pip.3259

RESEARCH ARTICLE

Estimating and parameterizing mismatch power loss in bifacial


photovoltaic systems

Chris Deline1 | Silvana Ayala Pelaez1 | Sara MacAlpine2 | Carlos Olalla3

1
National Renewable Energy Laboratory,
Golden, Colorado Abstract
2
Juwi Solar Americas, 1710 29th St., Boulder, Nonuniform irradiance on the rear side of bifacial photovoltaic (PV) systems can
Colorado
cause additional mismatch loss, which may not be appropriately captured in PV
3
Department of Electrical, Electronic, and
Automatic Control Engineering, Rovira i Virgili energy production estimates and software. We evaluated several scenarios including
University, Tarragona, Spain horizontal single-axis tracking (HSAT) over natural ground-cover and rooftop-
Correspondence mounted systems over high albedo reflective roofs. We found mismatch losses of up
Chris Deline, National Renewable Energy to 1.5% annual loss for very close-mounted (0.15 m) rooftop systems, but losses for
Laboratory, 15013 Denver West Pkwy,
Golden, CO. HSAT systems and high-ground-clearance rooftop systems were lower (<0.5%). A
Email: chris.deline@nrel.gov simplified empirical relationship was found that links the spatial variation of

Funding information irradiance (specifically, the mean absolute difference of irradiance) to the resulting
Solar Energy Technologies Office, Grant/ mismatch loss factor, with an R2 better than .99. This approximate relationship was
Award Numbers: 34910, 30286
experimentally validated on mismatched PV modules, and it provides a basis for
rapidly estimating bifacial mismatch loss factors for use in hourly PV performance
simulations such as PVSyst or SAM. Additional parameters investigated include
(a) climate dependence and module orientation, which were not found to have a
strong impact on bifacial mismatch losses and (b) PV module fill factor and bifaciality
ratio, which did have a strong linear impact on mismatch losses.

KEYWORDS

bifacial photovoltaics, PV systems

1 | I N T RO DU CT I O N with monofacial modules,4,5 thereby reducing the levelized cost of


electricity (LCOE)6,7 and improving the energy generated per unit area
Deployments of bifacial photovoltaic (PV) systems are quickly by making better use of the available solar resource.
proliferating because of their enhanced energy yield made possible by When deployed in a real PV installation, the irradiance incident
their ability to capture rear-side incident irradiance. More than 2 GWp on the rear of the bifacial PV module Grear depends on mounting
cumulative bifacial PV capacity has been deployed worldwide through conditions such as clearance height (H), row-to-row spacing, tilt, and
2018,1,2 and bifacial modules and systems are projected to reach 35% the site's albedo.1,8-10 The system energy generation therefore has
of the global market share by 2027.3 Bifacial PV modules are sensitive similar dependence. Furthermore, the uniformity of Grear throughout
to light incident on both sides, producing additional energy compared the module plane can be greatly influenced by self-shading of individ-
ual modules and by mutual shading of neighboring modules and
Alliance for Sustainable Energy, LLC (Alliance) is the manager and operator of the National rows,11-13 not to mention rear-racking obstructions.14 The impact of
Renewable Energy Laboratory (NREL) Employees of Alliance for Sustainable Energy, LLC,
under Contract No. DE-AC36-08GO28308 with the U.S. Dept. of Energy, have authored this
this nonuniform irradiance is reduction of short-circuit current (Isc) for
work. The United States Government retains and the publisher, by accepting the article for individual cells, and consequently, a nonconvex shape to the module
publication, acknowledges that the United States Government retains a non-exclusive, paid-
current-voltage (I-V) curve and reduction in the module's maximum
up, irrevocable, worldwide license to publish or reproduce the published form of this work, or
allow others to do so, for United States Government purposes. power point (Pmp). An example is provided in Figure 1 for the center

Prog Photovolt Res Appl. 2020;1–13. wileyonlinelibrary.com/journal/pip © 2020 John Wiley & Sons, Ltd. 1
2 DELINE ET AL.

F I G U R E 1 A, Font + rear-
irradiance distribution of a
bifacial module depends on the
mounting and site conditions.
The example shown is for Cairo,
21 June, at 2 PM (10o-tilt rooftop
system, 0.15-m H, 0.62
albedo); B, irradiance
distributions cause additional
module loss relative to uniform
assumption [Colour figure can be
viewed at wileyonlinelibrary.com]

 
PArray
module in a large bifacial rooftop array, assuming a 1-landscape Mismatch loss% = M½% = 1− × 100%: ð1Þ
ΣPCells
configuration of a 1 × 2 m module at 10o tilt angle, 0.15-m clearance
height H, and 0.62 ground albedo. For this example scenario, direct- Here, we establish the convention of reporting values in percentage
beam irradiance reflecting from the high-reflectance rooftop leads to rather than fractional terms. This mismatch loss is illustrated in
a brightening at the bottom of the module that is reduced toward the Figure 1B by comparing two calculated module I-V curves resulting
middle of the module. Irradiance is increased at the top of the module from the same bifacial optical and electrical simulations. One curve is
because of a greater field of view to the surrounding diffuse sky. This calculated by assuming that all cells in the module receive the same
type of nonuniform rear-irradiance profile has been reported multiple average irradiance over the entire module—effectively summing all
times,8-13 and it helps to motivate the current study because typical available power from each cell and averaging through the spatial vari-
PV energy-yield software (eg, PVSyst15 and SAM16) currently account ation in irradiance. The second curve results from a detailed treatment
for mismatch within the system as a single annual power loss factor, of the actual irradiance incident on each cell and resulting current mis-
not a detailed hourly value that depends on irradiance variability match. This detailed cell-level calculation of the array power accounts
throughout the system. for mismatch losses, whereas the first curve does not. The difference
Previous studies have investigated the uniformity of incident between these two peak-power levels defines the electrical mismatch
front-side irradiance and subsequent mismatch loss of conventional loss in Equation (1), and it must be calculated for each time point in an
(monofacial) PV systems, which can experience losses because of annual energy-yield simulation.
direct-beam shading.17-19 In these previous studies of direct-beam Although cell-level electrical models are accurate at calculating
shading, the irradiance differential within the system can be significant PArray, the intensive computation time makes them inappropriate for
when comparing shaded vs unshaded portions of the system. A large inclusion in fast (1-3 s) PV performance models such as PVSyst15 or
4× to 10× difference in irradiance within the PV module or system SAM.16 Furthermore, the rear-irradiance nonuniformity is a compli-
can lead to bypass-diode turn-on and large nonlinear losses if distrib- cated function of ground albedo, height, tilt angle, and more. There-
uted power electronics are not employed.20,21 However, the relatively fore, a simplified empirical representation of M based only on
modest variation in rear irradiance that develops within a bifacial sys- parameters of irradiance, and its spatial distribution would be
tem does not rise to the level of bypass diode turn on22; rather, it extremely valuable. This paper describes a successful reduced-order
more closely resembles the moderate shading losses resulting from fit to a series of hourly mismatch factors, based on detailed
diffuse inter-row front view-factor shading.23,24 Here, we specifically simulations of multiple bifacial system configurations. This extends
look at the influence of rear-irradiance distributions, which can be previously presented work on bifacial mismatch losses28 by proposing
nonuniform throughout the PV module or system and can therefore a universal parameterization of mismatch loss. Relative to prior
introduce additional mismatch losses for bifacial PV installations. This publications of a similar fashion in Janssen et al,22 we have improved
submission is an enhanced and expanded version of the conference the fit, conducted experimental validation, and evaluated different
20
paper recently presented in Deline et al, introducing improved sensitivities to climate and mounting configuration.
ray-tracing simulations, single-axis tracking scenarios and accounting In particular, M was found in all cases to depend on the spatial
for additional module parameters. distribution of total (front + rear) irradiance for cell i:
Our computation of mismatch loss in bifacial systems follows the
approach used previously25-27 for inter-row shading and arbitrary
Gtotal,i = Gfront,i + ϕBifi Grear,i , ð2Þ
shading of monofacial PV systems. Namely, a detailed cell-level
performance model is used to identify the actual array output PArray,
which includes mismatch losses. This is compared with the ideal where ϕBifi is the module's bifaciality at standard test conditions
output ΣPCells, which is the sum of individual cell-level power, (STC).12 The consideration of front-irradiance variability is still of high
excluding mismatch losses. Mismatch power loss (at a single point in importance for bifacial PV systems because the magnitude of front
time) is calculated for the array by irradiance outweighs that of rear irradiance in almost all real-life
DELINE ET AL. 3

bifacial system configurations. Very small front-side variability can considerably by looking up example curves and by quickly conducting
cause effects equivalent to larger rear-side irradiance variability, and it a bilinear interpolation between stored points, as described in Marion
should be considered, as well. et al.36 PV curves under specific environmental conditions are then
Through inspection, two scalar representations of irradiance dis- able to be calculated more rapidly than solving the equations exactly
tribution were found to have high correlation with M over multiple for each time point.
system configurations: the standard deviation of Gtotal,i and the mean The second step of the cell-level system analysis is to repre-
29
absolute difference of Gtotal,i. These are represented in relative sent a network of PV elements under different conditions and to
 total :
terms by dividing by average total irradiance G solve for the system's maximum operating point. This can be done
by several methods including matrix inversion,26 SPICE circuit
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  models,37,38 or most typically, direct summing of calculated I-V
1 Σ Gtotal,i − G  total 2
σ ½% =  x 100% ð3Þ curves.31,32 This is also the approach taken in the open-source soft-
Gtotal n
ware toolkit PVMismatch,39,40 which is the main method considered
and in this study. A brief comparison of this model was conducted with
1 X
n X
n   two additional electrical mismatch models26,36,39 for the rooftop
Δ½% = Gtotal,i −Gtotal,j  x 100%: ð4Þ
 total
n2 G configuration shown in Table 1. Because computation time was
i=1 j=1
similar for each model, and annual mismatch results for the three
The mean absolute difference Δ in Equation (4) has been shown previ- methods were within M < 0.1%, we will follow the method of
ously to accurately represent variability in non-normally distributed Meyers and Mikofski39 because of its open-source nature and
populations,30 which we have also found to be the case here. We also version control.
use the population formulations of each metric above because we are
representing the full population of irradiance inputs to a mismatch
model. It is important to note the distinction between mean absolute 2.2 | Bifacial optical simulations
difference and the similar-sounding (but distinct) mean absolute devi-
ation and median absolute difference, which are calculated differently Front and rear-irradiance values for use in the cell-level electrical
and do not have as strong a correlation with our measured results. model are given by annual hourly simulations using a previously
described bifacial view-factor model “bifacialvf,”41,42 which has been
validated against other rear-irradiance models and field data in Ayala
2 | E L E C T R I C A L A N D OP T I C A L M O D E L Pelaez et al.10,14 Single-axis-tracking simulations, particularly those
DETAILS including complicated geometry such as the presence of
rear-shading obstructions including torque tubes are performed in
2.1 | Single-cell electrical models RADIANCE43 using the “bifacial_radiance” open-source ray-tracing
software.44 The two-dimensional (2D) bifacialvf approach calculates
To accurately estimate electrical impacts of nonuniform irradiance, Gfront and Grear hourly across the collector width (CW) of the module
one must calculate energy at the cell level. In large series-connected assuming a semi-infinite array of n = 6 points for modules in land-
strings of PV cells and modules, current must be continuous through scape and n = 12 for modules in portrait mode, corresponding to
each electrical element. Therefore, mismatched irradiance prevents one point per cell along the axis of a 72-cell module. The
optimum operation of certain cells and modules in the system.31 Solv- bifacial_radiance simulation considers a PV array of seven rows,
ing for the overall global optimum operating point of a PV system with 20 modules per row, and irradiance calculated in an n = 100-point
irradiance gradients requires numerical simulation and is often done in scan along the center module in the center row. This array size
two steps. represents a large-scale field of modules big enough to reach the
In the first step, a I-V characteristic is calculated for each semi-infinite array steady state in rear irradiance because of nearby
independent PV element being simulated in the network. This can be rows and modules shadings.14 Bifacial_radiance simulations were
based on a lumped-element PV model using constant solar cell param- run on National Renewable Energy Laboratory's (NREL) high-
eters and variable environmental conditions (irradiance, temperature), performance “Eagle” computer (HPC), which enabled full hourly
as done in, for example, Bishop,31 Quaschning and Hanitsch,32 Singh simulations for multiple array configurations over a year of climate
et al,33 and De Soto et al.34 Advanced methods have also been conditions. Using 365 cores distributed into 11 nodes, each
proposed to explicitly solve the transcendental equations underlying
photovoltaic I-V curves35 or to otherwise reduce computation time. TABLE 1 Bifacial optical model configurations
Another example evaluated here computes a series of photovoltaic
System Height, m Albedo GCR Tilt # modules
I-V curves at several predetermined temperature and irradiance values
Rooftop 0.15-1 0.62 0.67 10o 1-landscape
(based on one of the above methods) and stores results in memory for
o
Tracking 1-2.4 0.2 0.33 0 1-portrait
future use. When simulating the many repeated conditions seen in
hourly simulations for a year, this method speeds up calculation Abbreviation: GCR, ground-coverage ratio.
4 DELINE ET AL.

simulation configuration time was reduced from 4 hours on a regu- 3 | S I M U L A T I O N RE S U LT S A N D R E D U C E D -


lar desktop computer to 1 minute using the Eagle HPC. ORDER MODEL

Initial simulations were conducted for the fixed-tilt rooftop scenarios


2.3 | System configuration and description of runs and single-axis-tracking scenarios described in Section 2. Cumulative
annual bifacial_radiance results for several scenarios are provided
To focus on scenarios that are of practical interest to the below, including the annual G  total (excluding rack shading or
community and to include conditions that may produce large mismatch):
amounts of mismatch loss, two typical configurations are considered
here: a fixed-tilt rooftop system deployed over a high reflective  rear  ϕ
G Bifi
BGG ½% =  front x 100% ð5Þ
white rooftop and a single-axis tracking system over a typical gro- G
und surface albedo. Both systems utilize 1 × 2-m modules, with
additional details shown below in Table 1. For the rooftop case, five where BGG is the bifacial irradiance gain or rear-irradiance ratio.
ground-clearance values are considered between H = 0.15 and 1 m. Specific findings and conclusions are given for each scenario below.
For the single-axis tracking case, H will vary throughout the day as
the tilt angle changes; hence, hub_height or module clearance at 0o
tilt is documented. Tracker rows are oriented north-south and mod- 3.1 | Cumulative annual mismatch loss
ules are deployed in 1-up portrait configuration, for a total CW of
2 m. Rear-irradiance nonuniformity from shading obstructions is For fixed-tilt simulations, BGG ranges from 10.5% to 19% depending
considered by modeling each hub height simulation either with or primarily on H but also on climate, as shown in Table 2. The single-
without a torque tube. Including obstructions can significantly axis-tracking simulations over natural ground albedo (0.2) resulted in
increase the accuracy between measured and modeled GRear.45 The lower values of BGG of 9% to 12% because of the reduced albedo and
torque tube considered was a black (absorbing) round tube of improved front-side irradiance capture. Fixed-tilt results match previ-
10-cm diameter, with a torque tube-to-module spacing (zgap) of ously reported values in Ayala Pelaez et al,10 which also showed BGG
5 cm. Figure 2A shows an illustration of the rooftop configuration of 10% to 18% for similar high-albedo rooftop configurations at dif-
with parameters highlighted, and the tracked system is represented ferent heights.
in Figure 2B with the torque tube included. Investigations into rear-irradiance uniformity indicate that Grear
Values of row spacing are defined by the ground-coverage ratio uniformity—and therefore, M—also changes with H. The total annual
(GCR)—the ratio of module CW to inter-row spacing distance mismatch loss MYear based on Equation (1) totaled over a year is
(Table 1). These values are chosen to eliminate front-side inter-row shown in Figure 3 at different heights. The largest loss MYear on the
shading, and in the case of the single-axis-tracking simulations, order of 1.5% annual mismatch loss occurs at lowest H, because of
back-tracking algorithms are assumed.46 The electrical model uses PV increased rear-irradiance nonuniformity at this low height.12 The
module source characteristics based on a 350-W, 72-cell n-PERT values shown here are in the same range as was previously reported
module with ϕBifi = 1.0 and fill factor (FF) of 0.79. The impact of mod- in Bailey and Jaubert28 for similar fixed-tilt conditions. The main take-
ule bifaciality is investigated later on, so the high ϕBifi value is used to away here is that the impact of mismatch loss in bifacial systems is
establish a reference baseline condition. Annual hourly simulations are sensitive to H, with lower bifacial mismatch losses at high clearance H.
conducted for multiple typical-year climate conditions for a typical US But for the more typical case of close-mounted rooftop systems,
climate (Richmond, VA), a sunny climate (Cairo, Egypt) and cloudy mismatch in rear irradiance can have a meaningful effect. For the
climate (Shanghai, China). lowest-height case considered here of 0.15 m, an initial rear-

(A) (B)

F I G U R E 2 System scenarios
considered: (A) fixed-tilt, south-facing
modules in 1-landscape, (B) north-south
horizontal single-axis-tracking (HSAT)
field of 20 × 7 modules in 1-portrait
[Colour figure can be viewed at
wileyonlinelibrary.com]
DELINE ET AL. 5

PN
TABLE 2 Simulation annual results, bifacial_radiance optical Grear ðwith tubeÞ
model Grear shading factor = Pn =N0 : ð6Þ
n = 0 Grear ðno tubeÞ

Height, Annual GTotal, BGG [%]


Climate m System kWh/m2 (Equation 5) This separate optical loss factor can be included as a rear-shading loss
Cairo 0.15 Fixed 2229 10.5 term48 (not mismatch loss) in annual simulations such as PVSyst.
Cairo 1 Fixed 2383 18 We have also investigated the climate dependence of mismatch
Richmond 1 Fixed 1795 17.5 loss by conducting the same fixed-tilt and HSAT simulations in three

Shanghai 1 Fixed 1734 19 climate zones: sunny Cairo, cloudy Shanghai, and temperate Rich-
mond. A slight climate dependence is visible for the high-albedo roof-
Cairo 1.5 HSAT 2552 9
top simulations in Figure 3. Here, high direct-normal irradiance (DNI)
Cairo 2 HSAT 2579 10
climates lead to greater ground-reflected irradiance nonuniformity.
Shanghai 2 HSAT 1784 12
MYear is therefore highest in Cairo and lowest in Shanghai. For HSAT
Abbreviation: HSAT, horizontal single-axis tracking. simulations, the effect of climate was similarly modest (not shown in
Figure 3), but the slight climate variability trend is reversed from what
was shown with high-albedo rooftop simulations—for the natural
ground-cover situation, the greatest mismatch power loss was seen
with the Shanghai simulations. This is possibly because of the greater
influence of diffuse irradiance in a low-albedo tracking scenario. The
results given by McIntosh et al47 echo this trend, indicating that mis-
match loss values are higher under more diffuse climate conditions for
HSAT simulations over natural ground cover.

3.2 | Fill factor, bifaciality factor, and albedo


dependence

Several additional parameters were investigated for their effect on


F I G U R E 3 Annual mismatch loss MYear for cloudy and sunny
system mismatch loss in the fixed-tilt rooftop scenario: module fill fac-
climates, for different clearance heights H of a low (fixed)-tilt bifacial
rooftop array, albedo = 0.62, and single-axis-tracking system over 0.2 tor (FF), bifaciality factor ϕBifi, and ground albedo. As shown in
albedo. Simulations use the bifacial_radiance optical model [Colour Figure 4A, modules exhibiting lower FF are susceptible to reduced
figure can be viewed at wileyonlinelibrary.com] mismatch loss. This is expected because of the smaller slope -dP/dI in
low-FF modules. When low-irradiance cells or modules are forced to
irradiance gain of 11% is reduced by the 1.5% annual mismatch loss pass current greater than their local Imp, the result is power loss. Low-
to provide an effective annual bifacial energy gain of 9% (excluding FF modules have a less steep power vs current slope between Imp and
module bifaciality). Isc and, therefore, lose less power for a given amount of current mis-
For HSAT simulations, we show two shading cases in Figure 3— match. The change in M was found to be directly proportional to the
including torque tube rear shading (dash-dot line) and without rear difference in FF, shown in Equation (7).
shading (dashed line). Accounting for torque tube rear-shading
obstructions requires the use of bifacial_radiance optical simulations, FF 1
M½%FF1 = M½%FF0 ð7Þ
FF 0
and the simulations shown in Figure 3 were all conducted using the
bifacial_radiance ray-tracing optical model on NREL's Eagle HPC
cluster. This linear relationship is illustrated in Figure 4A by scaling the original
For a given normalized height (H/CW), HSAT mismatch losses are FF = 79 curve to match the lower FF = 65 curve. By multiplying M by
slightly higher than for fixed-tilt simulations. In the case of no rear- the linear ratio of the two FF values, the lower FF curve is recreated.
shading from the torque tube, annual mismatch loss is 0.3% to 0.6%, This suggests that sample mismatch curves calculated for fixed values
matching the results of McIntosh,47 which showed values of 0.3% to of FF can be applied more generally to other module types by using
0.4% under similar conditions. With the effect of rear shading from Equation (7).
the torque tube, this value is increased by 0.1%. This is comparable Modules with lower ϕBifi and albedo also show a reduced loss fac-
with previously reported values of M in Bailey and Jaubert,28 where tor M, shown in Figure 4B. This is because modules with lower
cell shading from rack obstructions add an additional 0.1% to 0.2% to ϕBifi and systems with lower albedo experience lower values and dis-
M relative to the unshaded case depending on the extent of shadow. persion of Gtotal. As opposed to fill factor, however, there is no direct
For these cases, Grear is reduced by 8% between the no-shading and scaling factor for M because these parameters impact GTotal (per Equa-
torque-tube-shading simulations: tion (2)) and M only indirectly.
6 DELINE ET AL.

F I G U R E 4 Annual mismatch loss


MYear scaling vs FF, ϕBifi, and albedo for
the 1-up landscape rooftop scenario in
Richmond, VA. Reference scenario (black
line): FF = 79, albedo = 0.62, and
ϕBifi = 1. (A), Lower FF = 65 results (dots)
can be replicated by scaling reference
curve by Equation (7). (B), ϕBifi and albedo
affect GTotal dispersion directly and MYear
indirectly. Thus MYear does not scale
linearly with ϕBifi or albedo changes
[Colour figure can be viewed at
wileyonlinelibrary.com]

3.3 | Single-axis-tracking spatial distribution of Because commercially relevant array configurations use tracker rows
irradiance several times larger than the example here, this 1% power increase
from edge brightening likely represents an upper bound for commer-
An additional contribution to rear-side irradiance in real bifacial PV cial systems and remains a topic for further investigation.
systems is edge brightening.13,14 For unobscured modules at the end
of a tracker row, fewer surrounding obstructions lead to more
ground-reflected irradiance and increased GRear. This effect is a conse- 3.4 | Fixed-tilt reduced-order model development
quence of the breakdown of typical assumptions for PV system per-
formance modeling that assumes a semi-infinite array extent. The annual results of Figure 3 indicate relatively low overall mismatch
Although this localized irradiance “boost” may theoretically lead to loss, particularly for high clearance conditions and single-axis-tracking
increased energy capture by the system, some will be lost to electrical simulations. However, individual hourly mismatch loss values can be
mismatch loss. much higher—in excess of 10%.
Figure 5 shows cumulative annual GTotal for the tracker scenario By looking at hourly mismatch loss as a function of irradiance
given in Section 2.3 with H = 1.5 m including rear torque tube shading. standard deviation (σ) and mean absolute difference (Δ) from Equa-
The enhanced GTotal for the southernmost and northernmost modules tions (3) and (4), we can identify consistent performance trends. The
in the system are shown, as is the shading loss because of center tor- annual simulation data for each hour of the “rooftop” simulations in
que tube. These cumulative annual results indicate a maximum 5.8% Table 1 are compiled and considered together, using both bifacialvf
GTotal brightening on the south edge and a torque tube GRear shading and bifacial_radiance optical simulations. Data with low irradiance
optical loss14 of 8.3%. (GFront < 100 Wm−2 or GRear < 15 Wm−2) are excluded, leaving
Compared with the results given in Figure 3, where the 1-axis 130,300 remaining hourly points.
tracking system plus torque tube does not account for the finite edge Initial analysis shown in Figure 6 indicates that mismatch loss
effects of a 20-module tracker row, GTotal is 1.0% higher annually M trends closely with standard deviation of total irradiance σ[%],
when the edge brightening is averaged over the entire system. To Equation (3), with a wider irradiance distribution leading to increased
calculate the increase of GTotal over the system because of edge mismatch, as expected. The majority of data points indicate low power
brightening, we sum GTotal for each cell in the system and compare mismatch loss M < 3% when the sun is in front of the array plane (solar
with a scan over just the center of the 20-module tracker row. Mis- angle of incidence (AOI) < 90o, light blue points in Figure 6). However,
match losses are also slightly higher, at MYear = 0.6% for the higher mismatch conditions result when the sun is behind the plane of
20-module row instead of MYear = 0.45% for the tracker case in the array (black points in Figure 6, solar AOI >90o) and direct shading
Figure 3 that excludes edge effects (but accounts for torque tube can occur on the rear of the module. Conceptually, the highest spread
shading). The increase of GRear and M from edge effects depends on in rear irradiance occurs when direct irradiance, not just diffuse and
tracker row length and the number of interior vs edge modules. ground-reflected irradiance, is shining on the rear of the array.

F I G U R E 5 Year cumulative Gtotal [Whm−2]


for an interior row of 20 modules × 7 rows HSAT
at 1.5-m hub height, considering torque tube
[Colour figure can be viewed at wileyonlinelibrary.
com]
DELINE ET AL. 7

3.5 | Fixed-tilt reduced-order model—MAD

Close inspection of Figure 6 indicates that the results are split into
two populations, as described above. This suggests that Equation (3)
may not be the most appropriate way to describe the bifacial
rear-irradiance distributions. Standard deviation σ is typically used to
describe normally distributed populations, but Grear does not neces-
sarily follow a normal distribution across the module. Mean absolute
difference (Δ, Equation (4)) more appropriately represents variance in
non-normal distributions.30 Indeed, Equation (11) is used below in
Figure 7, which has the same form as Equation (10) but substituting
Δ instead of σ to plot the same data of Figure 6. It clearly results in
a tighter single population of data for all incidence angles. These
F I G U R E 6 Hourly simulation showing instantaneous power data are fit to a high degree of accuracy (R2 = .995) with a second-
mismatch M[%] as a function of Gtotal spatial standard deviation. Log-
order fit:
log Fit #1 from Equation (8). Comparison fit from Janssen et al22
shown as reference [Colour figure can be viewed at
wileyonlinelibrary.com] M½%Fit2 = 0:142 Δ½% + 0:032 Δ½%2 : ð11Þ

What remaining variability does exist shows a dependence on


rear-irradiance fraction (Equation (5)). The highest rear-irradiance con-
Two best-fit curves were found through the fixed-tilt data with ditions in red exhibit slightly higher M than the lower rear-irradiance
similar goodness of fit: a log-log fit, and a second-order polynomial fit. conditions shown in Figure 7B.
The log-log empirical-fit parameters are given in Equation (8), with an The accuracy of our two reduced-order empirical models in
R2 of .977. The curve is shown in Figure 6 in orange. Equations (8) and (11) are compared against the full cell-level model
(reference case) by summing the total mismatch power lost over each
height and climate scenario. This indicates the degree of increased
M½%Fit1 = e1:57*lnðσ½%Þ − 2 ð8Þ
error by moving from a detailed model to a simple empirical one.
Although our two optical models (bifacialvf and bifacial_radiance)
M½%Ref22 = 0:33 σ ½% + 0:0745 σ ½%2 ð9Þ
provide different values of overall annual mismatch because of differ-
ences in the optical results, the reduced-order models provide a very
M½%Fit1b = 0:15 σ ½% + 0:027 σ ½%2 ð10Þ good fit to both. The results in Table 3 indicate that both empirical
models provide a very close match, with the Log-log Fit #1 showing
slightly better agreement for low levels of annual mismatch, and the
The second fit is of the same form as was published previously second-order polynomial fit showing better agreement under high
in Janssen et al,22
although the empirical parameters were found to mismatch conditions. For both empirical fits, either Δ or σ can be
be substantially different for our scenario. The previously published substituted for the other with only a modest increase in average
best-fit equation is reproduced here in Equation (9), and the M error, in all cases <0.1%. Equations (8) and (11) can therefore be
updated best fit given our data is shown in Equation (10). For the used with either mean absolute difference (MAD) or standard devia-
fixed-tilt data in Figure 6, the curve of Equation (10) is indistinguish- tion as the independent input parameter with good overall accuracy in
able from that of Equation (8) with an R2 of .983, and therefore, the annual mismatch power loss estimation.
data are not shown. The original published curve of Equation (9) is An important question related to the configuration of the system
shown as a dashed line in Figure 6, indicating a poor fit to our data. is whether mismatch loss is significantly different when modules are
The reasons for the differences between these two studies is not oriented in portrait vs landscape. Additional fixed-tilt simulations were
yet fully understood. By careful inspection, however, it can be seen conducted where the 1 × 2 m modules are rotated in the portrait
that a handful of points from our study do follow the line of direction, and H and inter-row spacing is doubled to keep geometry
Equation (9). If additional low-irradiance points are included in our constant relative to the module collector width. The results in
−2
study (Gfront < 100 Wm ), then some of these outlier points also Figure 8 show good agreement between the two orientations,
follow this originally published trend line. Therefore, it is possible suggesting that module orientation should not significantly impact
that the best-fit equation of Equation (9) is based on an unfiltered bifacial mismatch loss. This follows logically from the fact that
dataset influenced mainly by low-irradiance, high-mismatch outliers. mismatch here is never so great that bypass diodes are activated.
Given the low power contribution of these low-irradiance points, Therefore, the main difference between landscape and portrait
we focus here on lower-mismatch, high-irradiance data instead, as modules—the bypass-diode configuration—should be expected to
was done in Equations (8) and (10). have no impact when shading and mismatch levels are quite modest.
8 DELINE ET AL.

F I G U R E 7 (A), Hourly simulation


showing power mismatch M as a function
of Gtotal mean absolute difference along
with Fit #2 (Equation (11)). (B), Remaining
variability with respect to Grear/Gfront
[Colour figure can be viewed at
wileyonlinelibrary.com]

TABLE 3 Annual mismatch loss simulated by the complete model vs the reduced-order model listed by equation number

bifacialvf bifacial_radiance

Full model, Fit #1 Fit #2 Full model, Fit #1 Fit #2


Height, m % Equation (8), % Equation (11), % Height, m % Equation (8), % Equation (11), %
0.15 2.04 2.04 2.01 0.15 1.57 1.56 1.56
0.25 1.50 1.43 1.48 0.25 1.07 1.04 1.09
0.5 0.44 0.42 0.49 0.5 0.30 0.29 0.36
1 0.05 0.04 0.08 1 0.05 0.06 0.10
Avg M N/A ±0.03 ±0.03 Avg M ERROR N/A ±0.01 ±0.03
ERROR

Note. Three annual climate scenarios at four fixed-tilt heights, two optical models.

not fall directly in line with the curves established for fixed-tilt simula-
tions. Although the form of the curve is similar, the empirical-fit
parameters are different. We use the general form of Fit #2
(Equation (11)) to calculate new best-fit parameters specifically for the
case of the tracking system with no torque tube, shown in
Equation (12).

M½%Fit3 = 0:054 Δ½% + 0:068 Δ½%2 ð12Þ

We should note that although the different scenarios begin to


diverge at very large Δ, agreement is much better at Δ < 4%. This is
relevant because the majority of the data and the bulk of the energy
generation is associated with low Δ. To identify the overall quality of
F I G U R E 8 Portrait and landscape fixed-tilt configurations show the empirical fits, we must compare the energy generation and calcu-
comparable mismatch. Portrait clearance height H and row spacing lated MYear for all of the scenarios considered in this paper. Although
are scaled 2× to keep H/CW and GCR constant [Colour figure can be the plots shown in Figures 7–10 are filtered for clarity with
viewed at wileyonlinelibrary.com] Gfront > 100 Wm−2, we now consider all unfiltered data for the overall
quality of model fits. Cumulative values of MYear are considered in
Table 4 for five groups of scenarios:
3.6 | Single-axis-tracking reduced-order model
1. Fixed, vf: All fixed-tilt scenarios, calculated using the bifacialvf opti-
Hourly mismatch power-loss values are next calculated for the case of cal model and PVMismatch electrical model. This includes three cli-
a north-south zero-tilt single-axis-tracking system with several differ- mate zones (Cairo, Shanghai, and Richmond) and four values of
ent configurations. The system parameters are listed in Section 2.3 H for both landscape and portrait module orientation.
and include four hub heights, three climates, and two rear-shading Albedo = 0.62.
cases: no rack shading and with a center torque tube included. Plots 2. Fixed, RAD: All fixed-tilt landscape scenarios as above, calculated
of hourly results are given in Figure 9, along with the previously using the bifacial_radiance optical model and PVMismatch electri-
shown fixed-tilt simulations. For the two tracking scenarios, results do cal model.
DELINE ET AL. 9

F I G U R E 9 Hourly simulation for HSAT. Three climate conditions


and two shading conditions—with and without rear torque tube
F I G U R E 1 0 Photo of experiment, showing the layout of
shading. Power mismatch M as a function of Gtotal mean absolute
transmission filters for each string in the bottom module, and top
difference along with Fit #2 (Equation (11)) and Fit #3 (equation (12))
module connected for four-wire measurement on a Spire 4600 long-
for tracking [Colour figure can be viewed at wileyonlinelibrary.com]
pulse flash I-V curve simulator [Colour figure can be viewed at
wileyonlinelibrary.com]

3. Tracking, vf: Single-axis-tracking scenarios calculated using the that the time required to calculate hourly M using Equations (8) or
bifacialvf optical model and PVMismatch electrical model. Three (11) is a fraction of a second, compared with the ~1 minute required
climates and four values of H are considered, in 2-up landscape ori- for the conventional full cell-level simulation. This speed-up of
entation, Albedo = 0.2. 3 orders of magnitude is compelling, given the accuracy of the
4. Tracking, rad: Single-axis-tracking scenarios calculated using the reduced-order model.
bifacial_radiance optical model and PVMismatch electrical model.
Three climate zones and four values of H are considered, in 2-up
landscape orientation with no rear-shading structure. Albedo = 0.2. 4 | EXPERIMENTAL VALIDATION
5. Tracking + TT: Single-axis-tracking scenarios calculated using the
bifacial_radiance optical model and PVMismatch electrical model. Field validation of Grear inhomogeneity has already been conducted by
Three climate zones and four values of H are considered, in 2-up several authors8-13 and is not the focus here. Instead, we are con-
landscape orientation. A 10-cm torque tube with an absorptive cerned with confirming the simplified electrical mismatch model
matte black surface (0.01 reflectance) is held 5 cm from the rear of described in Section 3. Reliance on modeled data is insufficient to give
the modules leading to a Grear shading optical loss factor of 8%. confidence in our empirical model's ability to describe field conditions.
Albedo = 0.2. I-V curve measurements of PV modules were collected under artifi-
cially applied mismatch conditions using an indoor solar simulator to
For each group of scenarios, we summarize the MYear value calcu- verify the applicability of these models to more general conditions.
lated by the full electrical model and three empirical fits given by Unlike previously published experiments such as Alonso-Garcia
Equations (8), (11), and (12). The average MYear value sums modeled et al49 or Sinapis et al,29 where a large amount of opaque shading is
power across all 295,000 hourly points. applied to cells, the amount of mismatch considered here is quite
Table 4 shows a clear winner among the various fit equations. Fit modest. Previous experiments have shown the response of PV sys-
#2 (Equation (11)) was initially modeled from fixed-tilt scenarios only, tems to partial shading to be significant and nonlinear. However, for
but it shows the best overall accuracy. Even for tracking scenarios, Fit bifacial systems, typically occurring irradiance distributions are not
#2 shows good agreement, better in fact than Fit #3, which was likely to result in cell reverse bias or bypass diode turn on.22 This is
established specifically from the single-axis-tracking data. This sug- because regardless of the amount of rear shadowing or inherent gra-
gests that the high Δ data, which influenced the least-squares fit to dient on the back of the module, Gfront will be much more uniform and
the data, may not be providing the best overall annual result. It is the greater in magnitude than Grear.
authors' suggestion that Fit #2, which is based on calculated mean
average difference in Gtotal, be used for all bifacial system scenarios
because it provides the best overall annual match to the full cell-level 4.1 | Experiment configuration
simulation.
A final consideration for the use of reduced-order empirical Here, the effect of irradiance spatial variability was investigated
models is the vastly improved computation time. We can note here experimentally on PV modules with small amounts of artificial partial
10 DELINE ET AL.

TABLE 4 Annual mismatch loss MYear for all simulation scenarios presented thus far

Simulation case Full model, % Fit #1 Equation (8), % Fit #2 Equation (11), % Fit #3 Equation (12), %
Fixed, vf 0.85 0.80 0.875 1.14
Fixed, rad 0.60 0.60 0.64 0.77
Tracking vf 0.18 0.18 0.25 0.22
Tracking rad 0.37 0.26 0.36 0.39
Tracking + TT 0.48 0.38 0.50 0.55
Avg M ERROR N/A 0.06 0.03 0.15

Note. Full cell-level model is listed along with the reduced-order model listed by equation number.

shading applied. A Spire 4600 long-pulse flash I-V curve simulator was Mismatch loss M from Equation (1) is calculated by comparing the
used, along with two Siemens SM55 36-cell monocrystalline modules module power at its terminals with the sum of each 12-cell substring
(monofacial, FF = 0.7). Monofacial modules are compatible with power. A four-wire measurement is used to remove interconnection
single-sided flash testing, and they can validate our empirical models series losses, which would accumulate over six separate substring
here because Gtotal distributions do not depend on whether irradiance measurement.
variability is on the front or rear of the module. Each SM55 module The standard deviation σ and MAD Δ for total irradiance is not
has two junction boxes—one on each end of the module—that house calculated directly but inferred from the measured Isc of each of the
the module's bypass diodes and interconnect terminals. In addition, six substrings. Some inherent mismatch in the modules is visible from
each module has ribbon tabs accessible, allowing each group of the no-shading Test 0 case where σ Isc=1.5% even without externally
12 cells to be individually contacted. By placing the two SM55 mod- applied shading. Other measured Isc values for each of the other shad-
ules side by side on the flash simulator and connecting the module ing tests are shown in Figure 11. Calculated σ Isc and ΔIsc have similar
terminals in series, we obtain a single 72-cell series connection of values and range up to 18% for the most extensive shade condition.
monocrystalline cells—and in which every 12 cells is individually con- Note that Figure 11 orders the substrings from lowest to highest mea-
tacted and measured and compared with the entire 72-cell series- sured Isc; this does not imply that the first substring of the module
connection measurement (Figure 10). was always the one to receive the heaviest shading.
The class AAA rating of the Spire simulator indicates that irradi-
ance spatial uniformity is within 2% when no artificial shading is
applied. To introduce artificial irradiance distributions, a series of 4.2 | Experiment results
semitransparent films are used, with spectrally normalized transmis-
sion between %T = 84% (3 M CG3300) and %T = 90% (3 M CG6000). Example module-level I-V curves from Test #3 are shown in Figure 12
Transmission was measured by monitoring the reduction in Isc when a under the artificial shading conditions given in Table 5. The multiple
single cell was repeatedly flashed under shaded conditions. After an steps of the curve indicate sequential shading of four of the six mod-
initial baseline measurement with no artificial shading applied, five ule substrings, leading to a measured variation in Isc of σ Isc = 10.9%
additional test configurations are applied with increasing levels of and ΔIsc = 12.1%. Under the combined module-level measurement,
shading. The final two measurement conditions use multiple layers of PModule = 81.4 W, whereas the six individual submodule measurements
P
semitransparent film to achieve higher amounts of shading loss. The resulted in a totaled PCells = 87.1 W. Applying Equation (1), we
amount of shading film applied to each module substring is given in arrive at a mismatch power loss of M = 6.5%. The measured mismatch
Table 5. To minimize ambiguous results and multiple I-V curve peaks, curve of Figure 12 allows us to illustrate an important point related to
the shading is applied uniformly across an entire row of 12 cells. the operation of bypass diodes. For multiply shaded PV strings and

TABLE 5 Filter transparency %T applied to module substrings for each test condition

Test # M1 str1, % M1 str2, % M1 str3, % M2 str1, % M2 str2, % M2 str3, %


0 N/A N/A N/A N/A N/A N/A
1 90 N/A 84 N/A N/A N/A
2 90 N/A 84 N/A N/A 90
3 90 7 84 N/A N/A 90
4 90 75 84 70 N/A 90
5 64 75 84 70 N/A 90

Note. N/A, no artificial shading applied.


DELINE ET AL. 11

F I G U R E 1 3 Measured mismatch loss M for each of the shading


F I G U R E 1 1 Measured substring Isc for each of the shading tests, tests vs σ Isc (squares) or ΔIsc (red circles). Empirical models shown for
ordered from lowest to highest Isc. σ Isc and ΔIsc each vary from 1.5% to comparison using the same coefficients given in Equations (8) and (11)
18% [Colour figure can be viewed at wileyonlinelibrary.com] adjusted for FF by Equation (7) [Colour figure can be viewed at
wileyonlinelibrary.com]

underlying Figure 8 indicates that >90% of annual energy is produced


during hours when Gtotal Δ < 4%. It is thefore the goodness of fit to
low Δ conditions that should carry the most weight. Thankfully both
models provide high accuracy for annual simulations (Table 3)
suggesting good agreement at smaller Δ values as well.

5 | A P P L I C A T I O N TO P E R F O R M A N C E
MODE LS O F ENE R GY Y I EL D
F I G U R E 1 2 Module-level indoor simulator measurement from
Test #3 showing current and power vs voltage. Isc variation:
σ Isc = 10.9%, ΔIsc = 12.1% [Colour figure can be viewed at A main consideration in developing these mismatch models is the
wileyonlinelibrary.com] application to typical PV yield simulation software—for example,
PVSyst and NREL's System Advisor Model (SAM). As of this writing,
modules, opaque shading can lead to high amounts of Isc mismatch, mismatch loss for bifacial simulations are considered by single annual
during which time the system Pmp shifts to a lower operating voltage loss factors. However, as we have seen above, mismatch loss can vary
through bypass diode turn on. Here, we can see that this is not occur- significantly from hour to hour, and it should ideally be applied hourly
ring in our case of modest irradiance nonuniformity. based on calculated Gtotal mismatch conditions and assuming spatial
Overall M values calculated for each of the five shading tests plus irradiance variability calculated as a part of the yield simulation. In this
the initial reference measurement are shown in Figure 13, expressed case, reduced-order models using Equation (11) and adjusting for PV
in terms of both σ Isc and ΔIsc. The empirical shading mismatch models module parameters with Equation (7) can quickly calculate hourly loss
developed from fixed-tilt simulations in Equations (8) and (11) are also parameters arising from rear-irradiance variability.
shown, along with 1% error bars based on simulator uniformity and A separate consideration for PVSyst users is that mismatch loss is
measurement repeatability. In Figure 13, the modeled values use expected to be applied to Grear only, rather than applying to combined
Equation (7) to account for the lower FF of the modules used in this DC power as we have reported above with M. A conversion between
experiment (0.7 vs 0.79 assumed in simulations). Although the FF cor- system power losses reported by this paper and rear-irradiance loss
rection is slight, it does lead to slightly better agreement with mea- terms required for simulations can be stated below.48 Starting with
sured data than if it were neglected. Equations (1) and (2), PArray can be recast in terms of array efficiency η
Figure 13 indicates a good match between indoor measurements and front and rear-average irradiance values:
and empirical fits based on simulated bifacial system production and
 
 front + G
 rear  ϕ
our assessment of mismatch loss' evolution with fill factor. Both loga- PArray = G Bifi  η  ð100%−MYear ½%Þ: ð13Þ
rithmic Fit #1 (Equation (8)) and second-order polynomial Fit #2
(Equation (11)) have good agreement even out to very large levels of In this formulation, consistent with reported values above, the
Gtotal Δ and M power loss. Although this is satisfying and indicates total mismatch factor MYear is applied to both front and rear photo-
appropriate mechanisms of mismatch loss are considered out to very generated power. Instead, we wish to develop a form that applies loss
high levels of mismatch, Sections 3.1 and 3.2 have shown us that typi- factors only to rear irradiance by introducing a new rear-irradiance
cal annual M values are 2% or less. In fact, inspection of the data loss factor Lrear:
12 DELINE ET AL.

 
 front + ð100%− Lrear ½%ÞG
 rear  ϕ
PArray = G Bifi  η, ð14Þ AC KNOWLEDG EME NT S
This work was authored, in part, by the National Renewable Energy
where the loss factor Lrear represents the electrical mismatch loss, Laboratory, operated by Alliance for Sustainable Energy, LLC, for the
 rear . Equations (13) and (14) can be
applied only as a factor reducing G US Department of Energy (DOE) under Contract DE-
used to solve for Lrear in terms of the rear-irradiance fraction of the AC36-08GO28308. Funding was provided by the US Department of
system (Equation (5)): Energy's Office of Energy Efficiency and Renewable Energy (EERE)
  under Solar Energy Technologies Office (SETO) Agreement 30286
100%
Lrear ½% = MYear ½% 1 + : ð15Þ
BGG ½% and 34910. The views expressed in the article do not necessarily
represent the views of the DOE or the US Government. A portion of
As an example to illustrate the use of the above equations, we assume the research was performed using computational resources sponsored
a bifacial system with a rear-irradiance fraction of 10% and system by the Department of Energy's Office of Energy Efficiency and
mismatch loss of MYear = 1%. Given these factors, we calculate that Renewable Energy and located at the National Renewable Energy
when used in software tools such as PVSyst, the appropriate rear- Laboratory.
irradiance mismatch loss would be Lrear = 11%. This is a significant
OR CID
increase in the mismatch loss factor when stated in these terms, and it
Chris Deline https://orcid.org/0000-0002-9867-8930
suggests that these factors should not be neglected.
Silvana Ayala Pelaez https://orcid.org/0000-0003-0180-728X
Carlos Olalla https://orcid.org/0000-0001-8333-9840

6 | C O N CL U S I O N S
RE FE RE NCE S
1. Libal J, Kopecek R. Bifacial photovoltaics: technology, applications and
Here, we have used NREL's HPC cluster to evaluate several bifacial
economics. Stevenage: The Institution of Engineering and Technology;
system scenarios on an hourly basis to determine electrical mismatch 2018 (IET Energy Engineering, Vol. 107).
losses M because of nonuniform front- and rear-side irradiance. Most 2. Kopecek R, Libal J. Towards large-scale deployment of bifacial
system configurations have annual MYear values below 1%, although photovoltaics. Nat Energy. 2018;3(6):443-446.
3. International Technology Roadmap for Photovoltaic (ITRPV) results
mismatch power loss is higher for individual hourly time periods
2017 including maturity report 2018, 9th edition, September 2018.
experiencing high irradiance gradients at low sun angles, or for sys- 2018.
tems with high albedo and low clearance height. Although these 4. Cuevas A, Luque A, Eguren J, del Alamo J. 50% more output power
values of M are low, system loss values must be converted to a value from an albedo-collecting flat panel using bifacial solar cells. Sol
Energy. 1982;29(5):419-420.
~10× greater to be of use in PV energy simulation tools (eg, PVSyst)
5. Sun X, Khan MR, Deline C, Alam MA. Optimization and performance
because these tools currently expect rear mismatch terms to apply of bifacial solar modules: a global perspective. Appl Energy. 2018;
only to Grear rather than Gtotal. 212(September 2017):1601-1610.
Single-axis-tracking systems were evaluated with and without 6. Darling SB, You F, Veselka T, Velosa A. Assumptions and the levelized
torque-tube rear-shading loss using the ray-tracing software cost of energy for photovoltaics. Energ Environ Sci. 2011;4(9):3133-
3139.
bifacial_radiance. Although the rear-shading obstructions were found
7. Bazilian M, Onyeji I, Liebreich M, et al. Re-considering the economics
to only modestly increase the level of mismatch loss, rear optical shad- of photovoltaic power. Renew Energy. 2013;53:329-338.
ing losses were found to be 8%. Spatial uniformity of irradiance was 8. Yusufoglu UA, Pletzer TM, Koduvelikulathu LJ, Comparotto C,
also found to be affected by edge effects from our assumed Kopecek R, Kurz H. Analysis of the annual performance of bifacial
modules and optimization methods. IEEE J Photovoltaics. 2015;5(1):
20-module length row. When placed entirely in series, these edge
320-328.
modules yielded an additional 1% of irradiance gain in the system, 9. Liang TS, Pravettoni M, Deline C, et al. A review of crystalline silicon
with only slightly increased M. bifacial photovoltaic performance characterization and simulation.
Based on detailed cell-level electrical mismatch models, we have Energ Environ Sci. 2019;12(1):116-148.
10. Ayala Pelaez S, Deline C, Macalpine S, Marion B, Stein JS, Kostuk RK.
developed an empirical relationship that links the spatial variation of
Comparison of bifacial solar irradiance model predictions with field
irradiance (mean absolute difference Δ) to M with an R2 better than validation. IEEE J Photovoltaics. 2019;9(1):82-88.
.99. When substituting the empirical fit for the full cell model in our 11. Kreinin L, Bordin N, Karsenty A, Drori A, Grobgeld D, Eisenberg N.
mismatch studies, annual mismatch loss values could be recovered PV module power gain due to bifacial design. Preliminary
experimental and simulation data. In: 35th IEEE Photovoltaic
within MYear = 0.1%. This is further validated by experiment using an
Specialists Conference. 2010. p. 002171–5.
indoor flash-test simulator and artificially applied shading to portions 12. Deline C, MacAlpine S, Marion B, Toor F, Asgharzadeh A, Stein JS.
of a module. Additional parameters that were established to have a Assessment of bifacial photovoltaic module power rating
linear impact on M include fill factor and module bifaciality ratio. We methodologies—inside and out. IEEE J Photovoltaics. 2017;7(2):
575-580.
have therefore developed a basis for fast estimation of bifacial mis-
13. Lindsay A, Chiodetti M, Binesti D, et al. Modelling of single-axis track-
match loss factors that can be used in hourly PV performance simula-
ing gain for bifacial PV systems. In: 32nd European Photovoltaic Solar
tion software directly or as a postprocessing tool to identify Energy Conference and Exhibition, Munich, Germany. 2016.
appropriate input parameters for such models. p. 1610–1617.
DELINE ET AL. 13

14. Ayala Pelaez S, Deline C, Greenberg P, Stein JS, Kostuk RK. Model 33. Singh JP, Aberle AG, Walsh TM. Electrical characterization method
and validation of single-axis tracking with bifacial PV. IEEE J for bifacial photovoltaic modules. Sol Energy Mater Sol Cells. 2014;
Photovoltaics. 2019;9(3):715-721. 127:136-142.
15. Wittmer B, Mermoud A. Yield simulations for horizontal axis trackers 34. De Soto W, Klein SA, Beckman WA. Improvement and validation of a
with bifacial PV modules in PVSyst. In: 35th European Photovoltaic model for photovoltaic array performance. Sol Energy. 2006;80(1):
Solar Energy Conference and Exhibition, Brussels, Belgium. 2018. 78-88.
16. Blair N, Dobos AP, Freeman J, et al. System Advisor Model, SAM 35. Petrone G, Spagnuolo G, Vitelli M. Analytical model of mismatched
2014.1. 14: General description. National Renewable Energy photovoltaic fields by means of Lambert W-function. Sol Energy Mater
Laboratory Golden, CO; 2014. Sol Cells. 2007;91(18):1652-1657.
17. Woyte A, Nijs J, Belmans R. Partial shadowing of photovoltaic arrays 36. Marion B, Rummel S, Anderberg A. Current-voltage curve translation
with different system configurations: literature review and field test by bilinear interpolation. Prog Photovoltaics Res Appl. 2004;12(8):
results. Sol Energy. 2003;74(3):217-233. 593-607.
18. Karatepe E, Boztepe M, Colak M. Development of a suitable model 37. Zekry A, Al-Mazroo AY. A distributed SPICE-model of a solar cell.
for characterizing photovoltaic arrays with shaded solar cells. Sol IEEE Trans Electron Devices. 1996;43(5):691-700.
Energy. 2007;81(8):977-992. 38. Paraskevadaki EV, Papathanassiou SA. Evaluation of MPP voltage
19. Appelbaum J, Bany J. Shadow effect of adjacent solar collectors in and power of mc-Si PV modules in partial shading conditions. IEEE
large scale systems. Sol Energy. 1979;23(6):497-507. Trans Energy Convers. 2011;26(3):923-932.
20. Deline C, Ayala Pelaez S, MacAlpine S, Olalla C. Bifacial PV mismatch 39. Meyers B, Mikofski M. Accurate modeling of partially shaded PV
loss estimation and parameterization. In: 36th EU PVSEC, Marseille, arrays. In: 2017 IEEE 44th Photovoltaic Specialist Conference (PVSC),
France. 2019. p. 1–5. Washington, DC. 2017. p. 3354–3359.
21. Femia N, Lisi G, Petrone G, Spagnuolo G, Vitelli M. Distributed 40. SunPower. PVMismatch [Internet]. Github. [cited 2019 Aug 21].
maximum power point tracking of photovoltaic arrays: novel Available from: https://github.com/SunPower/PVMismatch
approach and system analysis. IEEE Trans Ind Electron. 2008;55(7): 41. Marion B, MacAlpine S, Deline C, et al. A practical irradiance model
2610-2621. for bifacial PV modules. In: 44th IEEE Photovoltaic Specialists
22. Janssen GJM, Gali RSR, de Groot K, Carr AJ, Van Aken BB, Conference (PVSC), Washington, DC. 2017.
Romijin IG. Impact of inhomogeneous irradiance at the rear of bifacial 42. NREL. bifacialVF [Internet]. GitHub; 2019 [cited 2019 Aug 21].
panels on modelled energy yield. In: 33rd European PV Solar Energy Available from: http://github.com/NREL/bifacialvf
Conference and Exhibition (EU PVSEC), Amsterdam, Netherlands. 43. Ward GJ. The RADIANCE lighting simulation and rendering system.
2017. p. 1–6. In: 21st Annual Conference on Computer Graphics and Interactive
23. Passias D, Källbäck B. Shading effects in rows of solar cell panels. Sol Techniques. 1994. p. 459–72.
Cells. 1984;11(3):281-291. 44. NREL. bifacial_radiance [Internet]. GitHub; 2019 [cited 2019 Aug 21].
24. Van Schalkwijk M, Kil AJ, Van Der Weiden TCJ. Dependence of Available from: http://github.com/NREL/bifacial_radiance
diffuse light blocking on the ground cover ratio for stationary PV 45. Capelle T, Araya F, Haffner F, Sayritupac J, Colin H. A comparison of
arrays. Sol Energy. 1997;61(6):381-387. bifacial PV system modelling tools. In: 6th BifiPV Workshop,
25. Deline C, Dobos A, Janzou S, Meydbray J, Donovan M. A simplified Amsterdam, NL. 2019. p. 1–21.
model of uniform shading in large photovoltaic arrays. Sol Energy. 46. Holmgren WF, Andrews RW, Lorenzo AT, Stein JS. PVLIB Python
2013;96:274-282. 2015. In: 42nd IEEE photovoltaic specialist conference (PVSC), New
26. Olalla C, Clement D, Maksimovic D, Deline C. A cell-level Orleans, LA 2015. p. 1–5.
photovoltaic model for high-granularity simulations. In: 15th 47. McIntosh KR, Abbott MD, Sudbury BA, Meydbray J. Mismatch loss in
European Conference on Power Electronics and Applications (EPE), bifacial modules due to non-uniform illumination in 1D tracking
Lille, France. 2013. p. 1–10. systems. In: 46th IEEE PVSC, Chicago. 2019. p. 1–9.
27. MacAlpine S, Deline C, Dobos A. Measured and estimated 48. Ayala Pelaez S, Deline C, Stein JS, Marion B, Anderson K, Muller M.
performance of a fleet of shaded photovoltaic systems with string Effect of torque-tube parameters on rear-irradiance and rear-shading
and module-level inverters. Prog Photovoltaics Res Appl. 2017;25(8): loss for bifacial PV performance on single-axis tracking systems. In:
714-726. 46th IEEE Photovoltaic Specialists Conference Proceedings, Chicago
28. Bailey R, Jaubert J. A simplified method to approximate bifacial IL. 2019.
system mismatch losses. In: 5th Bifi PV Workshop, Denver, Colorado. 49. Alonso-Garcia MC, Ruiz JM, Chenlo F. Experimental study of
2018. mismatch and shading effects in the I-V characteristic of a
29. Sinapis K, Tzikas C, Litjens G, et al. A comprehensive study on partial photovoltaic module. Sol Energy Mater Sol Cells. 2006;90(3):329-340.
shading response of c-Si modules and yield modeling of string
inverter and module level power electronics. Sol Energy. 2016;135:
731-741.
30. Yitzhaki S. Gini's mean difference: a superior measure of variability How to cite this article: Deline C, Ayala Pelaez S,
for non-normal distributions. Metron. 2003;61(2):285-316.
MacAlpine S, Olalla C. Estimating and parameterizing
31. Bishop JW. Computer simulation of the effects of electrical
mismatches in photovoltaic cell interconnection circuits. Sol Cells. mismatch power loss in bifacial photovoltaic systems. Prog
1988;25(1):73-89. Photovolt Res Appl. 2020;1–13. https://doi.org/10.1002/pip.
32. Quaschning V, Hanitsch R. Numerical simulation of current-voltage 3259
characteristics of photovoltaic systems with shaded solar cells. Sol
Energy. 1996;56(6):513-520.

You might also like