Given
the
rapid
growth
and
development
of
solar
energy
in
recent
years,
accurate
forecasting
output
has
become
one
most
critical
formidable
challenges
modern
power
system.
This
paper
introduces
an
approach
for
short-term
global
irradiance
forecasting,
combining
Complete
Ensemble
Empirical
Mode
Decomposition
with
Adaptive
Noise
(CEEMDAN)
Artificial
Neural
Network
(ANN)
Multiple
Linear
Regression
(MLR)
models.
The
CEEMDAN
decomposition
is
employed
to
decompose
original
data
series,
extracting
crucial
features
forecasting.
model's
performance
evaluated
on
two
distinct
sites
Algeria,
characterized
by
Mediterranean
desert
climates.
Statistical
tests
reveal
a
significant
enhancement
nRMSE
values,
approximately
14.09%
7.86%
improvements
ANN_CEEMDAN
model
about
17.52%
8.97%
enhancements
MLR_CEEMDAN
model,
specifically
first
hour
forecast.
Sensors,
Journal Year:
2023,
Volume and Issue:
23(5), P. 2799 - 2799
Published: March 3, 2023
With
the
rapid
development
of
solar
energy
plants
in
recent
years,
accurate
prediction
power
generation
has
become
an
important
and
challenging
problem
modern
intelligent
grid
systems.
To
improve
forecasting
accuracy
generation,
effective
robust
decomposition-integration
method
for
two-channel
irradiance
is
proposed
this
study,
which
uses
complete
ensemble
empirical
mode
decomposition
with
adaptive
noise
(CEEMDAN),
a
Wasserstein
generative
adversarial
network
(WGAN),
long
short-term
memory
(LSTM).
The
consists
three
essential
stages.
First,
output
signal
divided
into
several
relatively
simple
subsequences
using
CEEMDAN
method,
noticeable
frequency
differences.
Second,
high
low-frequency
are
predicted
WGAN
LSTM
models,
respectively.
Last,
values
each
component
integrated
to
obtain
final
results.
developed
model
data
technology,
together
advanced
machine
learning
(ML)
deep
(DL)
models
identify
appropriate
dependencies
topology.
experiments
show
that
compared
many
traditional
methods
can
produce
results
under
different
evaluation
criteria.
Compared
suboptimal
model,
MAEs,
MAPEs,
RMSEs
four
seasons
decreased
by
3.51%,
6.11%,
2.25%,
Frontiers in Energy Research,
Journal Year:
2023,
Volume and Issue:
11
Published: July 26, 2023
Introduction:
Smart
grid
(SG)
technologies
have
a
wide
range
of
applications
to
improve
the
reliability,
economics,
and
sustainability
power
systems.
Optimizing
large-scale
energy
storage
for
smart
grids
is
an
important
topic
in
optimization.
By
predicting
historical
load
electricity
price
system,
reasonable
optimization
scheme
can
be
proposed.
Methods:
Based
on
this,
this
paper
proposes
prediction
model
combining
convolutional
neural
network
(CNN)
gated
recurrent
unit
(GRU)
based
attention
mechanism
explore
grid.
The
CNN
extract
spatial
features,
GRU
effectively
solve
gradient
explosion
problem
long-term
forecasting.
Its
structure
simpler
faster
than
LSTM
models
with
similar
accuracy.
After
CNN-GRU
extracts
data,
features
are
finally
weighted
by
module
performance
further.
Then,
we
also
compared
different
forecasting
models.
Results
Discussion:
results
show
that
our
has
better
predictive
computational
power,
making
contribution
developing
schemes
grids.
Energy Reports,
Journal Year:
2024,
Volume and Issue:
11, P. 1711 - 1722
Published: Jan. 20, 2024
Power
grid
stability
depends
on
the
ability
to
forecast
solar
power
generation
at
regional
level.
Most
previous
research
probabilistic
forecasting
has
focused
use
of
machine
learning
predict
output
individual
plants
rather
than
generation,
and
few
studies
have
considered
effects
seasonal
weather
patterns
In
this
study,
climate
geographic
data
were
collected
from
83
stations
between
2019
2021
for
in
developing
a
model
by
which
generation.
The
results
pattern
analysis
based
unsupervised
ensemble
voting
used
build
quantile
regression
short-term
prediction
capacity.
efficacy
was
assessed
using
48
PV
plants,
included
four
sub-datasets
pertaining
target
regions.
Highly
accurate
consistently
obtained
across
all
regions
both
winter
summer
seasons.
proposed
outperformed
conventional
deterministic
6.55%
4.03%
terms
total
normalized
mean
absolute
error
(NMAE).
Prediction
intervals
generated
could
be
as
input
parameters
dispatch
decisions.
IEEE Access,
Journal Year:
2024,
Volume and Issue:
12, P. 17915 - 17925
Published: Jan. 1, 2024
Deep
learning
has
grown
among
the
prediction
tools
used
within
renewable
energy
options.
Solar
belongs
to
options
with
lowest
atmosphere
impact
after
considering
their
limitations.
In
last
five
years,
Brazil
seen
expansion
of
wind
and
solar
almost
all
over
country,
preserve
Amazon
rainforest,
use
helped
large
small
cities
towards
a
greener
future.
The
novelty
this
research
covers
Learning
data
from
twelve
in
state
Amazonas
forecast
irradiation
(W.h/m
2
)
30
days.
input
came
ground
stations,
as
much
possible,
NASA
satellite
models,
daily
time
aggregation.
types
neural
networks
considered
are
Long
Short-Term
Memory
(LSTM),
Multi-Layer
Perceptron
(MLP),
an
LSTM
Gated
Recurrent
Unit
(GRU).
Among
metrics
check
algorithm's
performance,
Mean
Absolute
Percentage
Error
(MAPE)
indicates
that
values
coherent
other
scenarios
energy;
boundary
conditions
were
not
same,
however.
MAPE
was
observed
city
Labrea
GRU.
