applied
sciences
Article
A Deep Learning Model to Forecast Solar Irradiance Using a
Sky Camera
Rial A. Rajagukguk , Raihan Kamil






Citation: Rajagukguk, R.A.; Kamil,
R.; Lee, H.-J. A Deep Learning Model
to Forecast Solar Irradiance Using a
Sky Camera. Appl. Sci. 2021, 11, 5049.
https://doi.org/10.3390/app
11115049
Academic Editors: Takahiko
Miyazaki, Joon Ahn and Koji Enoki
Received: 30 March 2021
Accepted: 25 May 2021
Published: 29 May 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Appl. Sci. 2021, 11, 5049. https://doi.org/10.3390/app11115049
and Hyun-Jin Lee *
Department of Mechanical Engineering, Kookmin University, 77 Jeongneung-ro, Seoul 02727, Korea;
arifinrial@gmail.com (R.A.R.); raihan.kamil12@gmail.com (R.K.)
* Correspondence: hyunjinlee@kookmin.ac.kr; Tel.: +82-2-910-5466
Abstract: Solar irradiance fluctuates mainly due to clouds. A sky camera offers images with high
temporal and spatial resolutions for a specific solar photovoltaic plant. The cloud cover from sky
images is suitable for forecasting local fluctuations of solar irradiance and thereby solar power.
Because no study applied deep learning for forecasting cloud cover using sky images, this study
attempted to apply the long short-term memory algorithm in deep learning. Cloud cover data were
collected by image processing of sky images and used for developing the deep learning model
to forecast cloud cover 10 min ahead. The forecasted cloud cover data were plugged into solar
radiation models as input in order to predict global horizontal irradiance. The forecasted results
were grouped into three categories based on sky conditions: clear sky, partly cloudy, and overcast
sky. By comparison with solar irradiance measurement at a ground station, the proposed model was
evaluated. The proposed model outperformed the persistence model under high variability of solar
irradiance such as partly cloudy days with relative root mean square differences for 10-min-ahead
forecasting are 25.10% and 39.95%, respectively. Eventually, this study demonstrated that deep
learning can forecast the cloud cover from sky images and thereby can be useful for forecasting solar
irradiance under high variability.
Keywords: cloud cover; sky image; solar irradiance; deep learning
1. Introduction
Clouds have a serious impact on photovoltaic (PV) power production. By limiting the
levels of solar irradiance reaching the PV, the clouds can contribute to the high variability
of PV power output. Therefore, it is necessary to monitor cloud formation and obtain
information about clouds in order to be able to prepare backup energy sources that will
cover any reductions in energy output. Cloud information is applied in weather analyses
and meteorological data; also, it is used for energy applications such as solar irradiance
and PV power estimation. Furthermore, cloud information, such as cloud cover and cloud
motion, has frequently been examined as a source of renewable energy. Chow et al. [1]
used a sky camera installed in California to develop a method for intra-hour cloud motion
and to forecast global horizontal irradiance (GHI). Kim et al. [2] used sky images to retrieve
cloud cover and validated these images through human observations to estimate solar
irradiance. Lothon et al. [3] investigated an algorithm to estimate the cloud cover from
sky images.
Cloud information is a key input variable for solar irradiance forecasting, which
is a critical issue to manage uncontrollable production of PV power. In general, many
methods are able to forecast solar irradiance, and these methods are grouped according
to the forecast horizon. For long-term forecast horizons, numerical weather prediction
that relies on the numerical solution of governing equations in meteorological sciences is
most useful. Because predictions can be made up to 15 days in advance, such long-term
forecast horizon helps operation optimization and market participation. On the other hand,
https://www.mdpi.com/journal/applsci
Appl. Sci. 2021, 11, 5049 2 of 15

sky images and satellite images are used for forecasting solar irradiance in the short-term
forecast horizon. These methods are able to forecast solar irradiance several minutes to
several hours in advance. This short-term forecasting is useful in the anticipation of ramp
events caused by variations in solar irradiance. Compared to satellite images, sky images
offer more detailed cloud information with higher spatial resolution. Therefore, the solar
irradiance forecasting based on sky images is appropriate for management of a specific PV
system, in spite of shorter time horizon.
In general, information about clouds, especially for cloud cover data, is collected
through human observations, and the World Meteorological Organization provides the
rules for the registration of cloud cover. Observers estimate the cloud cover in oktas or
tenths, in which the sky is divided into 8 or 10 regions and the regions that are covered by
clouds are evaluated. However, the accuracy of this traditional method is deemed unsat-
isfactory: it provides low temporal and spatial resolutions, and errors can be introduced
through the subjective nature of observers’ judgments. Hence, a hemispherical sky camera
is an alternative solution that can be used to address some of these problems. The camera
gathers images at high frequency during the hours of daylight.
The fast evolution of low-cost hemispherical sky cameras has been preferred due to
their application in gathering cloud information [4], and they have become very popular
in the fields of solar energy and cloud motion detection. The sky camera is able to pro-
duce cloud cover information by image processing from pixel distribution. The data are
calculated based on an algorithm that applies the red, green, and blue (RGB) channels of
an image and combines them with an adaptive threshold in order to distinguish the cloud
and sky pixels. Besides, this data can be combined with physical models to estimate solar
irradiance. For example, Kim et al. [5] used cloud cover to estimate solar irradiance in
South Korea using the cloud cover radiation model. This model is a regression-type model
that was developed by Kasten et al. [6] to estimate GHI on an hourly basis. Furthermore,
some researchers have also used sky images to forecast solar irradiance in short-term
forecast horizons. Caldas et al. [7] forecasted GHI using sky images and real-time GHI
measurements. They applied the cloud correlation model (CCM) to these images to predict
GHI up to 10 min in advance. These results indicate that the proposed model was able to
predict GHI in a very short-term forecast horizon under high variability of solar irradiance.
So far, the GHI forecasting using sky images was calculated using conventional
methods, such as CCM and cloud motion vector (CMV). They estimated the cloud motion
by calculating the motion of the pixels on images. However, these methods did not provide
good accuracy because the cloud motion is not linear and hard to predict. Meanwhile,
deep learning, which is the subpart of artificial intelligence (AI), has been increasingly
used in solar irradiance and PV power forecasting [8–11]. Various deep learning models,
such as recurrent neural network (RNN), long short-term memory (LSTM), gated recurrent
unit (GRU), and deep belief network (DBN), provide high accuracy in forecasting solar
irradiance for both short-term and long-term forecast horizons. Nevertheless, no studies
applied deep learning models to forecast the solar irradiance using cloud cover data
obtained from sky images.
In this study, we proposed a method to estimate solar irradiance in high variability
of solar irradiance such as partly cloudy days. The proposed method combines the cloud
cover obtained from sky images, the deep learning model to forecast the cloud cover ten
minutes ahead, and the physical model to estimate GHI with the forecasted cloud cover.
The LSTM was selected out of deep learning algorithms because it is recommendable for
predicting time-series data compared with other deep learning models [8]. The forecasting
was conducted for clear, cloudy, and overcast days on a minute basis and validated by the
comparison with GHI measurement at a ground station in Seoul, South Korea. Because no
study used the deep learning model to forecast the cloud cover from sky images, this study
will investigate its applicability.
 2. Data and Instrument
Two instruments have been utilized in this study: a sky camera used to capture the
Appl. Sci. 2021, 11, 5049
sky image and a pyranometer used to measure the GHI. The instruments were set up on
3 of 15
the rooftop at Kookmin University in Seoul, South Korea (latitude: 37.61200° N, longitude:
126.99770° E).

2.1.Data
2. Sky and
Image
Instrument
Sky images
Two instrumentswere have
obtained
beenusing a Total
utilized Sky
in this Camera
study: a skyJ1006
cameraas shown
used toincapture
Figure the
1a,
sky
which image and asky
captured pyranometer used
images every to measure
minute. the GHI.
A fish-eye lensThe
as theinstruments were set
angle projection wasupthen
on
the rooftop ◦ N, longitude:
attached to at
theKookmin
camera,University in Seoul,
which provides South
a 180° Korea
field (latitude:
of view 37.61200
in the horizontal direction.
126.99770 ◦
The cameraE). system is based on a digital camera with a maximum resolution of 2272 × 1704
pixels, and the images were stored locally in a computer at the Kookmin University. The
2.1.
cameraSky system
Image contains large-capacity silica gel drying cartridges to keep the inside of the
Sky images
instrument free ofwere obtained
moisture. using a Total
A ventilation unit Sky
wasCamera J1006toasprevent
also applied shownthe in deposit
Figure 1a,
of
which
dust and captured
dew onsky imagessystem
the optical every minute. A fish-eye
and to protect the skylens as the
camera angle thermal
against projection was
heating
then attached to the camera, which provides a 180◦ field of view in the horizontal di-
effects.
rection.
The The camera captured
sky images system isthrough
based on theafish-eye
digital camera
lens were with a maximum
calibrated using resolution
a checker-
of 2272 × 1704 pixels, and the images were stored locally in
board calibration pattern to correct the fish-eye image. First, pictures were a computer at the Kookmin
taken of the
University. The camera system contains large-capacity silica gel
checkerboard in different positions and orientations, as shown in Figure 1b. Then, drying cartridges to keep
the
the inside of
MATLAB the instrument
toolbox was usedfree of moisture.
to detect A ventilation
the image and world unit was on
points alsothe
applied to prevent
checkerboard as
the
the deposit of dust
calibration and dewThe
parameters. on image
the optical system and topoints
and checkerboard protect the applied
were sky camera against
as the input
thermal heating
for the image effects. in converting all the images to flat images.
calibration

(a) (b)
Figure Aninstrument
1. An
Figure 1. instrumenttotorecord
recordsky
sky image
image data.
data. (a) (a) Total
Total SkySky Camera
Camera J1006;
J1006; (b) camera
(b) sky sky camera
cali-
calibration using the checkerboard
bration using the checkerboard image. image.

The sky
2.2. Solar images captured through the fish-eye lens were calibrated using a checker-
Irradiance
board calibration pattern to correct the fish-eye image. First, pictures were taken of the
Solar irradiance
checkerboard measurement
in different positionsrevolves around three
and orientations, components,
as shown that1b.
in Figure is, global
Then,hor-
the
izontal irradiance (GHI), diffuse horizontal irradiance (DHI), and
MATLAB toolbox was used to detect the image and world points on the checkerboard as direct normal irradiance
(DNI).
the The GHI
calibration is the totalThe
parameters. solar irradiance
image that reaches points
and checkerboard the earth’s
weresurface,
applied which is the
as the input
combination
for the image of DNI andin
calibration DHI. The GHI
converting allcan
the be measured
images to flat using
images.a pyranometer that has
been designed to receive solar irradiance from all directions. In this study, the GHI was
measured
2.2. using a EKO pyranometer MS-80 in Kookmin University (see Figure 2a,b). The
Solar Irradiance
pyranometer was then
Solar irradiance installed with
measurement ventilation
revolves aroundunits
threethat cover the that
components, dome is,of the pyra-
global hori-
nometer
zontal to protect
irradiance it from
(GHI), dew and
diffuse frost during
horizontal the morning
irradiance (DHI), and anddirect
to reduce
normal theirradiance
effects of
raindrops
(DNI). Theand
GHIdust.
is theThe pyranometer
total has athat
solar irradiance response
reachestime of less surface,
the earth’s than 0.5which
sec with an
is the
approximate sensitivity of 10 µ V/Wm −2 and collects the GHI in 1-min, 10-min, 1-hr, and 1-
combination of DNI and DHI. The GHI can be measured using a pyranometer that has
day intervals.
been designedThe pyranometer
to receive was calibrated
solar irradiance from following the ISO
all directions. 9060:2018
In this study, standard.
the GHI was
measured using a EKO pyranometer MS-80 in Kookmin University (see Figure 2a,b). The
pyranometer was then installed with ventilation units that cover the dome of the pyra-
nometer to protect it from dew and frost during the morning and to reduce the effects
of raindrops and dust. The pyranometer has a response time of less than 0.5 s with an
approximate sensitivity of 10 µV/Wm−2 and collects the GHI in 1-min, 10-min, 1-hr, and
1-day intervals. The pyranometer was calibrated following the ISO 9060:2018 standard.
Appl. Sci. 2021, 11, 5049 4 of 15
Appl. Sci. 2021, 11, 5049 4 of 15

(a) (b)
Figure 2. Solar irradiance measurement equipment installed on the rooftop of Kookmin University.
Figure 2. Solar irradiance measurement equipment installed on the rooftop of Kookmin Univer-
(a) MS-80 pyranometer to measure the global horizontal irradiance; (b) measurement station location.
sity. (a) MS-80 pyranometer to measure the global horizontal irradiance; (b) measurement station
location.
2.3. Classification of Sky Conditions
In order to forecast solar irradiance using sky images during a variety of sky conditions,
2.3. Classification of Sky Conditions
the 25 days from May to August in July 2020 were selected. These days provided all-sky
In order
conditions to forecast
based solar irradiance
on the clear-sky index andusing sky images
variability. during
Thereafter, a variety
days of sky
have been condi-
grouped
tions, the 25
according daysmean
to the fromofMay
thetoclear-sky
August in Julyand
index 2020clear-sky
were selected.
index These days which
variability, provided
is
all-sky conditions based on the clear-sky
calculated using the following equation: index and variability. Thereafter, days have been
grouped according to the mean of the clear-sky index and clear-sky index variability,
1 n GH Im
which is calculated using the following equation:
µkt = n ∑ (1)
1 n GHI GH I
 kt  i=1 m c (1)
n i 1 GHI c
δkt = Std{ktt−1 − ktt } (2)
where kt is the clear-sky index, µkt ktisthe ktt 1  of
Std{mean ktt the
} clear-sky index, GH I is the GHI
m (2)
measurement, GH Ic is the clear-sky GHI, and δkt is the clear sky index variability. Statistics
where
from thekt is the clear-sky
clear-sky be usedto
index can index, is the mean of the clear-sky index, GHI m index
kt determine the sky conditions. The clear-sky
is the
estimates the atmospheric
GHI measurement, GHI c attenuation
is the clear-skydueGHI,
to the  kt isby
andclouds themeasuring the ratio
clear sky index of the
variability.
surface solar radiation to the solar radiation received under a clear sky.
Statistics from the clear-sky index can be used to determine the sky conditions. The clear- In this study, the
days have been grouped into three categories [12] as shown in Figure
sky index estimates the atmospheric attenuation due to the clouds by measuring the ratio 3: clear-sky day
(of
µktthe> surface
0.65) and (δradiation
solar kt < 0.03to ), partly
the solarcloudy (0.3 ≤ under
dayreceived
radiation µkt ≤ a0.65 ) and
clear sky.(In > study,
δktthis 0.10),
and overcast day (µkt < 0.3) and (δkt < 0.03). Figure 4 compares the GHI measurement
the days have been grouped into three categories [12] as shown in Figure 3: clear-sky day
variability in all sky conditions for three representative days. It illustrates the small
( kt  0.65) and ( kt  0.03) , partly cloudy day (0.3  kt  0.65) and ( kt  0.10) , and overcast
variation in GHI for clear and overcast days from 9:00 a.m. to 5:00 p.m. On the other
day ( 
hand, the  0.3) and ( kt  day
kt partly cloudy
. Figurethe
0.03)presents 4 compares
high GHIthe GHI measurement
variability variability
with the maximum andinthe
all
sky conditions
minimum valuesfor
of three
GHI arerepresentative −
956.0 Wm days. 2 It illustrates
and 184.0 −
Wm , the2 small variation in GHI for
respectively.
clear and overcast days from 9:00 am to 5:00 pm. On the other hand, the partly cloudy day
presents the high GHI variability with the maximum and the minimum values of GHI are
956.0 Wm−2 and 184.0 Wm−2, respectively.
Appl. Sci.
Appl. Sci. 2021,11,
11, 5049 555of
of15
15
2021, 11,5049
Appl. Sci.2021, 5049 of 15

Figure 3. Day classification based on the distribution of mean clear-sky index ( kt ) and clear-sky
Figure 3.
Figure 3. Day classification based
based on
on the
the distribution
distributionof
ofmean
meanclear-sky index( (ktµ)ktand
clear-skyindex clear-sky
) and clear-sky
variability ( kt ))..
index variability
index variability((δkt
index kt
) .

Figure 4.
4. for clear-sky day (8 June 2020), partly cloudy day (4
Figure 4. The
Figure The variability
variability of
of GHI
GHI measurement
measurement for clear-sky day (8 June 2020), partly cloudy
cloudy day (4
day (4
June
June 2020), and overcast day (24 June 2020).
June2020),
2020),and
andovercast
overcastday
day(24
(24June
June2020).
2020).
3.
3. Methods
Methods
3. Methods
The method to
The method to forecast
forecastsolarsolarirradiance,
irradiance,which which is comprised
is comprised of the
of the twotwo steps,steps,
is pre-is
The method
presented in this to forecast
section. In solar
the irradiance,
first step, the which
cloud is comprised
cover is of thefrom
calculated two steps,
sky is pre-
images,
sented in this section. In the first step, the cloud cover is calculated from sky images, and
sented in the
and thisfuture
section. In thecover
first step, the cloud cover is calculated from sky images, and
thenthenthe future cloud cloud
cover is predictis predictusingusing
LSTM. LSTM.
In the Insecond
the secondstep, step,
the GHIthe GHI
is fore- is
then the
forecasted future
by cloud
applying cover
the is predict
predicted using
cloud LSTM.
cover In
for the
the second
solar step,
radiation the GHI
model is
as fore-
the
casted by applying the predicted cloud cover for the solar radiation model as the input
casteddata.
input by applying the predicted cloud cover for the solar radiation model as the input
data.
data.
3.1. Cloud Cover Algorithm
3.1. Cloud Cover Algorithm
3.1. Cloud Coverfurther
To include Algorithmdetails of the cloud cover retrieval, we have created a simple flow
chartTo of include
the further used
algorithm
To include further details
detailstoof the cloud
detect
of
coverand
the cloud
the cloud
retrieval, we have
sky pixels
cover retrieval, we have
created aassimple
on images,
created a simpleshown flow
in
flow
chart
Figure of the
5. the algorithm
Thisalgorithm used
flow chartused to
outlines detect the
the process cloud and
of obtaining sky pixels on images, as shown in
chart of to detect the cloud and skythe cloudoncover
pixels fromas
images, sky images
shown in
Figure
by using5. the
This flow chart outlines the process
values of obtaining thefirst
cloud cover
is tofrom skytheimages
Figure 5. Thisdistribution of the pixel
flow chart outlines the process in images. The
of obtaining the cloud step
cover collect
from sky images sky
by using
images fromthethe
distribution
sky camera. of These
the pixel values
images arein images. The
converted first step is to image
collect the sky
by using the distribution of the pixel values in images. intoThe afirst
one-channel
step is to collectinthe order
sky
images
to improvefrom the
thethe sky camera.
contrast These
and toThese images
reduceimages are
the noise converted into a one-channel image in or-
images from sky camera. areof the imageinto
converted by aapplying
one-channelthe RBR imagemethod;
in or-
der
this to improve the contrast and to reduce the noise of the image by applying the RBR
der method
to improve was theproposed
contrast byandShield et al. [13]
to reduce theand is able
noise of theto successfully
image by applyingdetect the thecloud
RBR
 clouds from the sky. We calculated the value using the Otsu thresholding method that
Nobuyaski Otsu proposed back in 1979 [15]. This method was chosen on account of its
flexibility and robustness in identifying the cloud and sky pixels. Another advantage of
this method is the simple process involved in determining the threshold values: since the
Appl. Sci. 2021, 11, 5049 calculation requires one-dimensional intensity data, the other parameters, such 6asof the 15
shape or geometric components of an object, do not affect the accuracy of the threshold
value.
It is worth noting that the basic concept in thresholding is to segment an image based
and sky pixels in anbetween
on the difference image. Mathematically,
the pixel value. the RBR method
Therefore, a fixed is threshold
defined asvalue
the ratio of the
cannot be
red
applied to all-sky images because the brightness in sky images differs on account of be
and blue channels of an image with the pixel value ranging from 0 to 255. It should the
noted that the value
ever-changing of theofblue
position thechannel
sun. Toisaddress
increasedthese
by 1 problems,
if it equals the
0 to threshold
avoid dividing
valuebyis
0adapted
[14]. to all images, and this makes the value different in each image.

Figure5.5. Flow
Figure Flow chart
chart for
for determining
determiningcloud
cloudcover
coverusing
usingred, green,
red, and
green, blue
and (RGB)
blue channels
(RGB) of an
channels of
image.
an image.

Finally,
After the
that, cloud
the covervalue
threshold was calculated
is applied using
to the the ratio of cloud
one-channel imagepixels to total pixels
to distinguish the
based on the percentage of the cloud pixels in the binary image. The
clouds from the sky. We calculated the value using the Otsu thresholding method measurement of cloud
that
cover is reported
Nobuyaski in the number
Otsu proposed back of
in parts of theThis
1979 [15]. sky covered
method by wasclouds.
chosenTheonsky can be
account ofsplit
its
into 8 (oktas)
flexibility and or 10 (tenths)
robustness in parts representing
identifying theand
the cloud amount of cloud
sky pixels. in a particular
Another advantagesky.ofIt
should
this methodalsoisbethe
noted thatprocess
simple cloud involved
cover does in not describe the threshold
determining cloud thickness
values:and that
since theit
only refersrequires
calculation to the amount of the skyintensity
one-dimensional covered data,
by clouds in a parameters,
the other particular location.
such as the shape
or geometric components of an object, do not affect the accuracy of the threshold value.
3.2. Solar Radiation
It is worth Model
noting that the basic concept in thresholding is to segment an image based
on theThedifference betweenmodel
solar radiation the pixel
has value. Therefore,
been used a fixed
to estimate GHI threshold value input
using various cannotdata.
be
applied to all-sky images because the brightness in sky images differs on
One of the solar radiation models that uses cloud cover as the input data is the Kastenaccount of the
ever-changing
model [6]. Thisposition of the sun.
regression-type modelTo estimates
address these
GHI byproblems, the threshold
using a correlation value the
between is
adapted to all images, and this makes the value different in each image.
Finally, the cloud cover was calculated using the ratio of cloud pixels to total pixels
based on the percentage of the cloud pixels in the binary image. The measurement of cloud
cover is reported in the number of parts of the sky covered by clouds. The sky can be split
into 8 (oktas) or 10 (tenths) parts representing the amount of cloud in a particular sky. It
should also be noted that cloud cover does not describe the cloud thickness and that it only
refers to the amount of the sky covered by clouds in a particular location.

3.2. Solar Radiation Model

The solar radiation model has been used to estimate GHI using various input data.
One of the solar radiation models that uses cloud cover as the input data is the Kasten
model [6]. This regression-type model estimates GHI by using a correlation between
the cloud cover and clear-sky irradiance. In this model, the cloud cover was divided
into 9 classes, where 0 refers to a clear sky and 8 an overcast sky. This model has been
Appl. Sci. 2021, 11, 5049 7 of 15

widely used in different locations, but some researchers had to modify the coefficients in
order to obtain results in their location. In this specific location in Seoul, the coefficients
were obtained from research conducted by Yoo et al. [16]. The GHI is calculated using
this formula:
GH I = GH Ic × (1 − A × (0.125 × CC ) B ) (3)
GH Ic = C × sin(α) − D (4)
where CC is the cloud cover, GH Ic is the solar irradiance on clear sky condition, α is
solar elevation, A = 0.75, B = 2.6, C = 963, and D =106. However, the specific value does not
reach the GHI under overcast sky conditions, since the minimum value for an overcast sky
is a quarter of clear-sky irradiance. Therefore, we proposed a new model by modifying
the coefficient in Kasten’s model and the clear sky irradiance model. The coefficients to
calculate the GHI for each model are presented in Table 1.

Table 1. The coefficients to estimate global horizontal irradiance (GHI) for each model.

Model A B
Kasten 0.75 3.4
Yoo [16] 0.75 2.6
Proposed 0.95 3.4

Many clear-sky irradiance models have been made available to calculate the solar irra-
diance under clear sky conditions, and each of these models requires different parameters
as the input. For example, Dazhi et al. [17] calculated clear-sky irradiance in combination
with the solar zenith angle and the eccentricity of the earth. Antonanzas-Torres et al. [18]
estimated solar irradiance based on commonly measured variables, such as temperature,
rainfall, and humidity. Yang et al. [19] proposed a model to calculate solar irradiance
based on ozone absorption, water vapor absorption, permanent gas absorption, aerosol
extinction, and Rayleigh scattering. In this work, we used the Ineichen clear-sky model
after modification by Reno et al. [20], as this model affords good accuracy and fairly easy
to execute. The equations to obtain the clear-sky irradiance obtained from are expressed
as follows:
GH Ic = c g1 × I × cos(θz ) × exp(−c g2 × AM × ( f h1 + f h2 × ( TL − 1))) × exp(0.01 × AM1.8 ) (5)

1
AM = (6)
cos(θz ) + 0.50572 × (96.07995 − θz )−1.6364
f h1 = exp(−d/8000) (7)
f h2 = exp(−d/1250) (8)
−5
c g1 = 5.09 × 10 · d + 0.868 (9)

c g2 = 3.92 × 10−5 · d + 0.0387 (10)

where I is extraterrestrial normal incident irradiance, θz is solar zenith, AM is air mass,
TL is linked turbidity factor, and d is the ground elevation expressed in meters.
Here, the linked turbidity refers to the optical thickness of the atmosphere due to the
presence of gaseous water vapor and the absorption and scattering by the aerosol [21]. It
expresses the transparency of the sky or cloudless atmosphere. In a case where the sky is
perfectly blue (clean), the TL value is close to 1. However, if the sky has high water vapor
and the color is closer to white, the TL becomes larger. The air mass refers to the relative
path length of the direct solar beam through the atmosphere, and it describes the ratio
of the distance traveled by solar radiation in reaching the atmosphere to the distance of
the sun directly overhead. Note, in this solar irradiance model, the air mass is dependent
solely on the so-lar zenith.
Appl. Sci. 2021, 11, 5049 8 of 15

3.3. Deep Learning Model

In this study, we used LSTM as a deep learning model to forecast the cloud cover up
to several minutes ahead. LSTM was chosen as it provides satisfactory results in handling
time series data. Rajagukguk et al. [7] identified that this model performs better than
other deep learning models, such as RNN and GRU. Furthermore, this deep learning
model demonstrates good performance in solar irradiance forecasting in both short-term
and long-term forecast horizons. LSTM was developed to overcome the problems with
vanishing and explosion gradients that often occur in other deep learning models. For
instance, in the case of the RNN, when these problems occurred in the learning process,
the learning performance failed to increase [22].
The LSTM model was proposed by Hochreiter and Scmidhuber to adapt to the long-
term dependence on the information [23]. The unit, as has been illustrated in Figure 6,
consists of forget gate, input gate, output gate, and cell state. For simplicity’s sake, the
structure can be formulated as follows:
 
f t = σ W f × x t + U f × h t −1 + b f (11)

it = σ (Wi × xt + Ui × ht−1 + bi ) (12)

St = tanh(Wc × xt + Uc × ht−1 + bc ) (13)
Ct = it  St + f t  Ct−1 (14)
ot = σ (Wo × xt + Uo × ht−1 + Vo × Ct + bo ) (15)
ht = ot  tanh(Ct ) (16)
where f t is the forget gate, i is the input gate, ot is the output gate, x t is input data, b
is bias, W,U,V are weight matrices, ht is the value of memory cell, S t is candidate state
Appl. Sci. 2021, 11, 5049 9 of 15
of the memory cell, tanh and σ are the activation functions, and C t is the state of the
memory cell.

Figure6.
Figure 6. The
The architecture
architectureof
ofthe
thelong
longshort-term
short-termmemory
memory(LSTM).
(LSTM).

In
In Equations
Equations (11),
(11), (12)
(12),and
and(15),
(15),the
thesigmoid
sigmoidfunction
functionisisused
usedto tocalculate
calculatethe
theamount
amount
of
ofinformation
informationthatthatpasses
passesthrough
throughthethegate with
gate values
with from
values from 0 to
0 1.
to The candidate
1. The candidatestate of
state
the memory cell in Equation (13) contains the tanh function that makes the
of the memory cell in Equation (13) contains the tanh function that makes the value rang- value ranging
from −1 to
ing from −11toin1 order to calculate
in order thethe
to calculate newnewinformation.
information.The Theinput
inputandandforgot
forgotgates
gatesin
in
Equation (14) are operated with the Hadamard product () to calculate the
Equation (14) are operated with the Hadamard product ( ) to calculate the state memory state memory of
the cell. The final output of the memory cell in Equation (16) was obtained after multiplying
of the cell. The final output of the memory cell in Equation (16) was obtained after multi-
with the output gate.
plying with the output gate.
A total of 12,000 images were processed to obtain minutely cloud cover information
A total of 12,000 images were processed to obtain minutely cloud cover information
in this study. These calculated data provided the input for LSTM to forecast the future
in this study. These calculated data provided the input for LSTM to forecast the future
cloud cover. To illustrate the model design, the details of these hyperparameters are listed
in Table 2.

Table 2. The hyperparameters used in designing the LSTM model.

Appl. Sci. 2021, 11, 5049 9 of 15

cloud cover. To illustrate the model design, the details of these hyperparameters are listed
in Table 2.

Table 2. The hyperparameters used in designing the LSTM model.

Hyperparameter Searching for Optimal Value Optimal Value

Epoch [50,100,200,300,400,500] 300
Activation function ReLU ReLU
Batch size [64,128,256] 128
Optimization Adam Adam
Learning rate 0.001 0.001

The optimum hyperparameters such as epochs, batch size, learning rate, optimization
algorithm, and activation functions for deep learning models depend on the datasets. The
epochs describe the number of complete passes (forward and backward) through the neural
network. The batch size denotes the number of training examples in one pass (forward and
backward). The learning rate controls the weight in a neural network with reference to loss
gradient. The optimization algorithm is used to find the attributes, such as determining the
weights neural network to reduce the losses. Along with this structure, this LSTM model is
also equipped with an activation function known as a rectified linear unit (ReLU). ReLu is
a non-linear function that oppresses a value below 0 to become exactly 0 but still inherits
some linear property for cases above 0. Because this function has a linear characteristic, it
can easily train by a deep network of neurons and also solve a case of gradient problem
by ignoring the negative values. It should be noted that there is no fixed value in deep
learning models to explain the optimum design for each model because the networks
inside deep learning models are trained iteratively. Therefore, the best way to determine
the optimal hyperparameter was to use errors during validation and training to assess the
algorithm’s accuracy.

3.4. Evaluation Metric

In order to validate the performance of forecasting, various common evaluation
metrics have been used to calculate the accuracy of the model, including mean bias differ-
ence (MBD), root mean square difference (RMSD), relative root means square difference
(rRMSD), and relative mean bias difference (rMBD). We prefer to use differences rather
than errors as the measurement data by nature include uncertainty; hence, the true values
are unknown [24]. The evaluation metrics are given in the following equations:

1 n
n i∑
MBD = ( Pe − Pm ) (17)
=1
s
1 n
n i∑
RMSD = ( Pe − Pm )2 (18)
=1
n
1
n ∑ ( Pe − Pm )
i =1
rMBD = n (19)
1
n ∑ Pm
i =1
s
n
1 2
n ∑ ( Pe − Pm )
i =1
rRMSD = n · 100 (20)
1
n ∑ Pm
i =1

where Pe is the estimated values at each time, Pm is the measured values at each time,
and n is the number of sample data for the period.
Appl. Sci. 2021, 11, 5049 10 of 15

The RMSD explains the deviation from the measurement and it always generates
a positive value. This value measures how close the prediction is to the measurement;
thus, a smaller value is deemed better. The MBD shows the average bias of the prediction,
and it also provides the long-term performance of the model [25], where the positive and
negative values represent overprediction and underprediction, respectively. Furthermore,
in cases where the data vary with location and time scale, relative metrics such as rRMSD
and rMBD are more useful, as they provide the percentage difference.

4. Results and Discussion

This section provides the results from both the RBR model used to obtain the cloud
cover and the deep learning model used to forecast solar irradiance. The RBR method
detects the presence of clouds by considering the ratio of the red channels to blue channels
in a color image. These one-channel images combined with the threshold values were
able to distinguish cloud pixels from the background. In the sky images, the cloud and
sky pixels were detected as white and black colors and were further transformed from
the original image into the binary image to facilitate the identification of cloud and sky
pixels. However, in some cases, errors occurred when the cloud pixel was almost the same
as the sky pixel. In addition, if the clouds are extremely thin, the pixels are lower than the
threshold value, and they are not detected as clouds. Hence, it is essential to determine an
accurate threshold value to lessen the occurrence of such classification errors.
In order to understand the correlation between cloud cover and the clear-sky index,
the distribution of the daily calculated cover data is presented in Figure 7, which illustrates
the variation of mean cloud cover for each day. To investigate the results, the mean clear-
Appl. Sci. 2021, 11, 5049 sky index has been converted into an opposite value on account of the differences11in of the
15
definition of sky conditions in relation to cloud cover and the clear-sky index; for example,
in referring to cloud cover, a value of 0 is a clear-sky condition, while in the clear-sky index,
aand
value
the of 0 refersindex
clear-sky to anisovercast sky. The correlation
worth mentioning between
because it shows the cloud
similar trends,cover
whichand the
indi-
clear-sky
cates thatindex is worth
the method mentioning
provides becauseresults
satisfactory it shows similar trends,
in obtaining which
the cloud indicates
cover that
from sky
the method provides satisfactory results in obtaining the cloud cover from sky images.
images.

Figure 7. Mean cloud cover distribution and mean clear-sky index for each day data. Note: the
Figure 7. the mean
mean clear-sky
clear-sky indexchanged
index was was changed
by theby the opposite
opposite original
original valuevalue to follow
to follow the cloud
the cloud covercover defini-
definition.
tion.
In forecasting results, the forecast cloud cover using the LSTM model is converted
In forecasting
into GHI data using results, the forecast
the correlation cloudcover
cloud coverand
using the LSTM
clear-sky model isasconverted
irradiance described
into
in GHI data
Equation using
(3). Thethe correlation
results for thecloud
minutescover and clear-sky
forecasting wereirradiance
compared as with
described in
the GHI
Equation (3). The
measurement results
in terms of for the minutes
evaluation forecasting
metrics. were compared
The summary with the GHI
of solar irradiance meas-
forecasting
urement
is grouped inin
terms
threeofcategories
evaluation metrics.
based Theconditions:
on sky summary of solarcloudy,
partly irradiance
clear,forecasting is
and overcast
grouped
days. Theinproposed
three categories
model by based on skythe
modifying conditions: partly
coefficient cloudy,model
in Kasten’s clear, outperformed
and overcast
days. The proposed model by modifying the coefficient in Kasten’s model outperformed
the Yoo model and the persistence model for partly cloudy days as shown in Table 3. After
10 min forecasting, the RMSD for the proposed, the Yoo, and the persistence models are
199.75 Wm−2, 214.97 Wm−2, and 317.94 Wm−2, respectively. For reference, another forecast-
Appl. Sci. 2021, 11, 5049 11 of 15

the Yoo model and the persistence model for partly cloudy days as shown in Table 3.
After 10 min forecasting, the RMSD for the proposed, the Yoo, and the persistence models
are 199.75 Wm−2 , 214.97 Wm−2 , and 317.94 Wm−2 , respectively. For reference, another
forecasting model by Caldas et al. [7], which used the cross-correlation method (CCM) in
all-sky imaging, attained an RMSD of 251 Wm−2 for partly cloudy days in forecasting solar
irradiance up to 10 min in advance.
The results for clear days and overcast days are shown in Table 4. The results indicate
that under clear day conditions the proposed model outperforms the persistence and the
Yoo models for forecast horizon more than 5 min. The RMSD in 10 min forecasting for
the proposed, the Yoo, and the persistence models are 43.08 Wm−2 , 63.74 Wm−2 , and
52.49 Wm−2 , respectively. However, the information was different on overcast days. In this
case, the smallest error by the persistence model, the RMSD of 9.98 Wm−2 in 10 min ahead
of forecasting, demonstrated that the persistence model performed well under the overcast
condition. The proposed and the Yoo models provide the RMSD of 17.37 Wm−2 and
169.86 Wm−2 , respectively. The result in these overcast days was not surprising since the
value of solar irradiance does not change significantly. The experiment by Caldas et al. [7]
also reported that the persistence model has better performance than their proposed model
for overcast day conditions, with an RMSD of 110 Wm−2 for 10 min ahead forecast horizon.
To further analyze the variability of the solar irradiance under different sky conditions,
the metric that indicates the variability of solar irradiance is presented in Figure 8. The
metric was calculated using the universal variability index (UVI) to quantify the variation
of solar irradiance under clear, partly cloudy, and overcast days by comparing the measured
irradiance and calculated clear-sky irradiance. As expected, the small variations in clear
and overcast days resulted in the values of UVI close to one. However, the UVI for
Appl. Sci. 2021, 11, 5049 13 of 15
partly cloudy days does not remain constant anymore due to the high variability of solar
irradiance. It shows the high variability index with the value ranging from 2.68 to 61.16.

Figure8.8.Universal
Figure Universalvariability
variabilityindex
index(UVI)
(UVI)for
for10
10min
minutes ahead
ahead forecasting
forecasting in different
in different sky condi-
sky conditions:
tions: clear sky, partly cloudy, and overcast
clear sky, partly cloudy, and overcast day. day.
Appl. Sci. 2021, 11, 5049 12 of 15

Table 3. Evaluation metrics of forecast performance between the proposed, the Yoo, and the persistence models for partly cloudy days.

Forecast Proposed Yoo [16] Persistence

Horizon RMSD MBD rMSD rMBD RMSD MBD rMSD rMBD RMSD MBD rMSD rMBD
(minute) (Wm−2 ) (Wm−2 ) (%) (%) (Wm−2 ) (Wm−2 ) (%) (%) (Wm−2 ) (Wm−2 ) (%) (%)
1 137.00 7.58 18.05 1.66 144.12 −50.72 18.98 −11.14 219.70 −10.13 28.94 3.81
2 143.67 12.76 20.08 2.97 153.20 −60.91 21.41 −14.19 343.17 35.52 47.96 6.70
3 154.72 −6.57 20.92 −1.48 165.26 −78.41 22.35 −17.67 379.92 11.76 51.38 6.95
4 167.50 −30.19 21.31 −6.40 173.02 −92.77 22.01 −19.67 319.28 −59.65 40.61 5.17
5 166.49 −14.75 21.68 −3.20 172.28 −78.49 22.44 −17.04 296.78 −43.61 38.65 5.03
6 173.84 17.77 23.15 3.94 181.78 −39.79 24.21 −8.83 310.31 −35.94 41.33 5.50
7 178.33 −16.36 23.49 −3.59 185.04 −84.15 24.37 −18.47 169.01 −12.96 22.26 2.93
8 190.83 −20.86 26.03 −4.74 201.44 −100.89 27.47 −22.93 170.84 7.13 23.30 3.18
9 187.02 −31.19 25.49 −7.09 205.31 −115.21 27.99 −26.17 299.60 15.42 40.84 5.57
10 199.75 −60.65 25.10 −12.70 214.97 −131.53 27.01 −27.54 317.94 −43.28 39.95 5.02
Appl. Sci. 2021, 11, 5049 13 of 15

Table 4. Root mean square difference (RMSD) between the proposed, the Yoo, and the persistence
models for clear days and overcast days.

Forecast Clear Days Overcast Days

Horizon Proposed Yoo [16] Persistence Proposed Yoo [16] Persistence
Appl. Sci. 2021, 11, 5049 13 of 15
(minute) (Wm−2 ) (Wm−2 ) (Wm−2 ) (Wm−2 ) (Wm−2 ) (Wm−2 )
1 20.59 35.44 6.20 17.74 74.32 2.52
2 31.59 39.15 7.79 16.73 90.66 3.66
3 37.80 31.78 14.04 18.17 98.26 4.67
4 45.70 41.40 20.92 19.41 99.86 9.43
5 45.56 57.56 28.26 19.34 104.51 6.64
6 29.69 59.68 56.24 20.20 138.79 5.48
7 34.76 54.05 41.74 20.46 119.06 8.30
8 34.15 53.76 38.28 17.33 158.65 7.26
9 40.27 58.20 49.26 15.60 155.47 6.48
10 43.08 63.74 52.49 17.37 169.86 9.98

The performance of forecast models significantly differs by solar irradiance variability.

Figure 9 shows the mean RMSD and the mean UVI of 10-min-ahead forecast results
under different sky conditions. It clearly explains that the proposed model remarkably
outperformed the persistence model in partly cloudy days with the mean UVI > 20. On
the other hand, when the mean UVI < 5, the forecast performance of the proposed model
is comparable to that of the persistence model. In overcast days, the proposed model is
slightly
Figure 8.less accurate
Universal than the
variability persistence
index model.
(UVI) for 10 Probably,
minutes the reason
ahead forecasting is that the
in different solar
sky condi-
radiation
tions: clearmodel is less
sky, partly accurate
cloudy, and under overcast
overcast day. sky conditions.

Figure9.9.Comparison
Figure Comparisonbetween
between proposed
proposed and
and persistence
persistence models
models using
using the
the mean
mean values
values of
of RMSD
RMSD
and UVI for 10-min forecast horizon in partly cloudy, clear, and overcast days.
and UVI for 10-min forecast horizon in partly cloudy, clear, and overcast days.

5.5.Conclusions
Conclusions
In
Inthis
this study, model to
study, a model toforecast
forecastsolar
solarirradiance
irradiance from
from sky sky images
images waswas proposed
proposed and
and evaluated.
evaluated. The images
The images recorded
recorded usingusing
a skyacamera
sky camera
were were
designeddesigned to capture
to capture a hemi-a
hemispherical
spherical image image
eacheach minute.
minute. ImageImage processing
processing using
using the the
RBRRBR method
method waswas used
used to
to im-
improve
prove the the qualityofofthe
quality theimage
imageinindetecting
detectingthe
thecloud
cloudpixels
pixelsininthe
theimage.
image.The
TheLSTM,
LSTM,as asa
adeep
deeplearning
learningmodel,
model,was wasapplied
applied toto forecast
forecast the cloud cover for several minutes ahead.
several minutes ahead.
Then,
Then,the
thephysical
physical model
model was used
was to estimate
used the GHI
to estimate by using
the GHI the forecasted
by using cloud cover.
the forecasted cloud
cover.
When the sky is partly cloudy, i.e., the UVI is larger than 20, the proposed model is
able to forecast GHI for 10 minutes in advance with RMSD, rMSD, MBD, and rMBD values
of 199.75 Wm−2, 25.1%, −60.65 Wm−2, and −12.70%. These evaluation metrics demonstrate
that the proposed model has a better performance than the Yoo and the persistence mod-
Appl. Sci. 2021, 11, 5049 14 of 15

When the sky is partly cloudy, i.e., the UVI is larger than 20, the proposed model is
able to forecast GHI for 10 min in advance with RMSD, rMSD, MBD, and rMBD values of
199.75 Wm−2 , 25.1%, −60.65 Wm−2 , and −12.70%. These evaluation metrics demonstrate
that the proposed model has a better performance than the Yoo and the persistence models.
On the other hand, for clear and overcast sky conditions when the UVI is smaller than 5,
performances of the proposed and the persistence models are comparable.

Author Contributions: Formal Analysis R.A.R.; Conceptualization R.A.R. and R.K.; Writing—
original draft R.A.R.; Writing—review and editing H.-J.L.; Funding acquisition H.-J.L. All authors
have read and agreed to the published version of the manuscript.
Funding: This study is financially supported by grants from the National Research Foundation of
Korea (NRF), Ministry of Science and ICT (2018M1A3A3A02065823, 2019R1A2C1009501).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Data are contained within this study.
Conflicts of Interest: The authors declare no conflict of interest.

Nomenclature

AM Air mass
CC Cloud cover
Ct The state of the memory cell
d Ground elevation (m)
ft Forget gate
ht The value of memory cell
it Input gate
Io Extraterrestrial irradiance
kt Clear-sky index
ot Output gate
St The candidate state of the memory cell
TL Linke turbidity
xt Input data
Subscripts
cc Cloud cover
kt Clear-sky index
e Estimated
m Measured
c Clear sky
Greek Symbols
α Solar angle
δkt Clear-sky index variability
θz Sun zenith
σ Sigmoid function
Abbreviations
GHI Global horizontal irradiance (Wm−2 )
LSTM Long short-term memory
MBD Mean bias difference (Wm−2 )
PV Photovoltaic
RBR Red blue ratio
ReLU Rectified linear unit
rMBD Relative mean bias difference (%)
RMSD Root mean square difference (Wm−2 )
rRMSD Relative root mean square difference (%)
UVI Universal variability index
Appl. Sci. 2021, 11, 5049 15 of 15

References
1. Chow, C.W.; Urquhart, B.; Lave, M.; Dominguez, A.; Kleissl, J.; Shields, J.; Washom, B. Intra-Hour Forecasting with a Total Sky
Imager at the UC San Diego Solar Energy Testbed. Sol. Energy 2011, 85, 2881–2893. [CrossRef]
2. Kim, B.Y.; Jee, J.B.; Zo, I.S.; Lee, K.T. Cloud Cover Retrieved from Skyviewer: A Validation with Human Observations. Asia Pac. J.
Atmos. Sci. 2016, 52, 1–10. [CrossRef]
3. Lothon, M.; Barnéoud, P.; Gabella, O.; Lohou, F.; Derrien, S.; Rondi, S.; Chiriaco, M.; Bastin, S.; Dupont, J.C.; Haeffelin, M.; et al.
ELIFAN, an Algorithm for the Estimation of Cloud Cover from Sky Imagers. Atmos. Meas. Tech. 2019, 12, 5519–5534. [CrossRef]
4. Wacker, S.; Gröbner, J.; Zysset, C.; Diener, L.; Tzoumanikas, P.; Kazantzidis, A.; Vuilleumier, L.; Stöckli, R.; Nyeki, S.; Kämpfer, N.
Cloud Observations in Switzerland Using Hemispherical Sky Cameras. J. Geophys. Res. 2015, 120, 695–707. [CrossRef]
5. Kim, K.H.; Kie-Whan Oh, J.; Jeong, W.S. Study on Solar Radiation Models in South Korea for Improving Office Building Energy
Performance Analysis. Sustainable 2016, 8, 589. [CrossRef]
6. Kasten, F.; Czeplack, G. Solar and Terrestrial Radiation Dependent on the Amount and Type of Cloud. Sol. Energy 1980, 24,
177–189. [CrossRef]
7. Caldas, M.; Alonso-Suárez, R. Very Short-Term Solar Irradiance Forecast Using All-Sky Imaging and Real-Time Irradiance
Measurements. Renew. Energy 2019, 143, 1643–1658. [CrossRef]
8. Rajagukguk, R.A.; Ramadhan, R.A.A.; Lee, H.-J. A Review on Deep Learning Models for Forecasting Time Series Data of Solar
Irradiance and Photovoltaic Power. Energies 2020, 13, 6623. [CrossRef]
9. Brahma, B.; Wadhvani, R. Solar Irradiance Forecasting Based on Deep Learning Methodologies and Multi-Site Data. Symmetry
2020, 12, 1–20. [CrossRef]
10. Wang, F.; Yu, Y.; Zhang, Z.; Li, J.; Zhen, Z.; Li, K. Wavelet Decomposition and Convolutional LSTM Networks Based Improved
Deep Learning Model for Solar Irradiance Forecasting. Appl. Sci. 2018, 8, 1286. [CrossRef]
11. Husein, M.; Chung, I.Y. Day-Ahead Solar Irradiance Forecasting for Microgrids Using a Long Short-Term Memory Recurrent
Neural Network: A Deep Learning Approach. Energies 2019, 12, 1856. [CrossRef]
12. Alves, M.D.C.; Sanches, L.; de Nogueira, J.S.; Silva, V.A.M. Effects of Sky Conditions Measured by the Clearness Index on the
Estimation of Solar Radiation Using a Digital Elevation Model. Atmos. Clim. Sci. 2013, 3, 618–626. [CrossRef]
13. Shields, J.E.; Johnson, R.W.; Koehler, T.L. Automated Whole Sky Imaging Systems for Cloud Field Assessment. In Proceedings of
the Fourth Symposium on Global Change Studies of the American Meteorological Society, Boston, MA, USA, 17–22 January 1993.
14. Li, Q.; Lu, W.; Yang, J. A Hybrid Thresholding Algorithm for Cloud Detection on Ground-Based Color Images. J. Atmos. Ocean.
Technol. 2011, 28, 1286–1296. [CrossRef]
15. Nobuyuki, O. A Threshold Selection Method from Gray-Level Histograms. IEEE Trans. Syst. Man Cybern. 1979, 9, 62–66.
16. Yoo, H.-C.; Lee, K.-H.; Park, S.-H. Analysis of Data and Calculation of Global Solar Radiation Based on Cloud Data for Major
Cities in Korea. J. Korean Sol. Energy Soc. 2008, 28, 17–24.
17. Dazhi, Y.; Jirutitijaroen, P.; Walsh, W.M. The Estimation of Clear Sky Global Horizontal Irradiance at the Equator. Energy Procedia
2012, 25, 141–148. [CrossRef]
18. Antonanzas-Torres, F.; Sanz-Garcia, A.; Martínez-de-Pisón, F.J.; Perpiñán-Lamigueiro, O. Evaluation and Improvement of
Empirical Models of Global Solar Irradiation: Case Study Northern Spain. Renew. Energy 2013, 60, 604–614. [CrossRef]
19. Yang, K.; Huang, G.W.; Tamai, N. Hybrid Model for Estimating Global Solar Radiation. Sol. Energy 2001, 70, 13–22. [CrossRef]
20. Reno, M.J.; Hansen, C.W. Identification of Periods of Clear Sky Irradiance in Time Series of GHI Measurements. Renew. Energy
2016, 90, 520–531. [CrossRef]
21. Chaâbane, M.; Masmoudi, M.; Medhioub, K. Determination of Linke Turbidity Factor from Solar Radiation Measurement in
Northern Tunisia. Renew. Energy 2004, 29, 2065–2076. [CrossRef]
22. Cortez, B.; Carrera, B.; Kim, Y.J.; Jung, J.Y. An Architecture for Emergency Event Prediction Using LSTM Recurrent Neural
Networks. Expert Syst. Appl. 2018, 97, 315–324. [CrossRef]
23. Hochreiter, S.; Schmidhuber, J. Long Short-Term Memory. Neural Comput. 1997, 9, 1735–1780. [CrossRef] [PubMed]
24. Lave, M.; Hayes, W.; Pohl, A.; Hansen, C.W. Evaluation of Global Horizontal Irradiance to Plane-of-Array Irradiance Models at
Locations across the United States. IEEE J. Photovolt. 2015, 5, 597–606. [CrossRef]
25. Gairaa, K.; Bakelli, Y. A Comparative Study of Some Regression Models to Estimate the Global Solar Radiation on a Horizontal
Surface from Sunshine Duration and Meteorological Parameters for Ghardaïa Site, Algeria. ISRN Renew. Energy 2013, 2013, 1–11.
[CrossRef]