Applied Energy,
Journal Year:
2023,
Volume and Issue:
350, P. 121645 - 121645
Published: Aug. 10, 2023
Renewable
energies,
such
as
solar
power,
offer
a
clean
and
cost-effective
energy
source.
However,
their
integration
into
national
electricity
grids
poses
challenges
due
to
dependence
on
climate
geography.
While
numerous
studies
have
focused
time
series,
few
specifically
addressed
the
critical
task
of
forecasting
production
at
level.
Accurate
national-level
is
crucial
for
optimizing
management,
informing
policy
development,
promoting
environmental
sustainability.
This
study
aims
address
associated
with
significant
variability
in
renewable
its
impact
grid
stability
by
improving
accuracy
existing
approaches.
To
achieve
this
goal,
we
evaluate
effectiveness
univariate
multivariate
approaches
series
data
from
ESIOS
(the
Spanish
System
Operator).
Our
primary
focus
leveraging
external
variables,
irradiance
data.
end,
propose
methodology
integrate
forecasts
historical
power
plants
Spain
improve
performance
models.
Subsequently,
compare
classical
regression
techniques
state-of-the-art
deep
learning
algorithms,
presenting
models
three
forecast
horizons
(1
h,
24
48
h).
Finally,
assess
our
best
comparing
them
official
ESIOS.
findings
indicate
that
best-performing
are
deep-learning
approaches,
which
benefit
incorporating
forecasts,
particularly
longer
(24
h
h),
avoid
detrimental
effects
Hughes
Phenomenon,
seems
hamper
non-deep-learning
forecasters.
The
top-performing
models,
based
Convolutional
Networks
+
Recurrent
Neural
Networks,
outperform
reducing
mean
absolute
error
41%
47.58%,
respectively.
Case Studies in Thermal Engineering,
Journal Year:
2024,
Volume and Issue:
60, P. 104743 - 104743
Published: June 24, 2024
In
this
study,
eXtreme
Gradient
Boosting
(XGBoost)
and
Light
(LightGBM)
algorithms
were
used
to
model-predict
the
drying
characteristics
of
banana
slices
with
an
indirect
solar
drier.
The
relationships
between
independent
variables
(temperature,
moisture,
product
type,
water
flow
rate,
mass
product)
dependent
(energy
consumption
size
reduction)
established.
For
energy
consumption,
XGBoost
demonstrates
superior
performance
R2
0.9957
during
training
0.9971
testing,
alongside
minimal
MSE
0.0034
0.0008
testing
phase
indicating
high
predictive
accuracy
low
error
rates.
Conversely,
LGBM
shows
lower
values
(0.9061
training,
0.8809
testing)
higher
0.0747
0.0337
reflecting
poorer
performance.
Similarly,
for
shrinkage
prediction,
outperforms
LGBM,
evidenced
by
(0.9887
0.9975
(0.2527
0.4878
testing).
comparative
statistics
showed
that
regularly
outperformed
LightGBM.
game
theory-based
Shapley
functions
revealed
temperature
types
most
influential
features
model.
These
findings
illustrate
practical
applicability
LightGBM
models
in
food
operations
towards
optimizing
conditions,
improving
quality,
reducing
consumption.
The Astrophysical Journal Supplement Series,
Journal Year:
2024,
Volume and Issue:
270(1), P. 15 - 15
Published: Jan. 1, 2024
Abstract
Solar
energetic
particle
(SEP)
events
and
their
major
subclass,
solar
proton
(SPEs),
can
have
unfavorable
consequences
on
numerous
aspects
of
life
technology,
making
them
one
the
most
harmful
effects
activity.
Garnering
knowledge
preceding
such
by
studying
operational
data
flows
is
essential
for
forecasting.
Considering
only
cycle
(SC)
24
in
our
previous
study,
we
found
that
it
may
be
sufficient
to
utilize
soft
X-ray
(SXR)
parameters
SPE
forecasts.
Here,
report
a
catalog
recording
≥10
MeV
flux
unit
SPEs
with
properties,
spanning
SCs
22–24,
using
NOAA’s
Geostationary
Operational
Environmental
Satellite
data.
We
an
additional
daily
SXR
statistics
this
period,
employing
test
application
machine
learning
(ML)
prediction
support
vector
(SVM)
extreme
gradient
boosting
(XGBoost).
explore
training
models
from
two
SCs,
evaluating
how
transferable
model
might
across
different
time
periods.
XGBoost
proved
more
accurate
than
SVMs
almost
every
considered,
while
also
outperforming
SWPC
NOAA
predictions
persistence
forecast.
Interestingly,
done
SC
produces
weaker
true
skill
statistic
Heidke
scores
2
,
even
when
paired
22
or
23,
indicating
transferability
issues.
This
work
contributes
toward
validating
forecasts
long-spanning
data—an
understudied
area
SEP
research
should
considered
verify
cross
robustness
ML-driven
Transportation Engineering,
Journal Year:
2024,
Volume and Issue:
16, P. 100228 - 100228
Published: Feb. 2, 2024
This
study
investigates
the
combination
of
audio
and
image
data
to
classify
road
conditions,
particularly
focusing
on
loose
gravel
scenarios.
The
dataset
underwent
binary
categorisation,
comprising
segments
capturing
sounds
corresponding
images.
Early
feature
fusion,
utilising
a
pre-trained
Very
Deep
Convolutional
Networks
19
(VGG19)
Principal
component
analysis
(PCA),
improved
accuracy
Random
Forest
classifier,
surpassing
other
models
in
accuracy,
precision,
recall,
F1-score.
Late
involving
decision-level
processing
with
logical
disjunction
conjunction
gates
(AND
OR)
individual
classifiers
for
images
based
Densely
Connected
121
(DenseNet121),
demonstrated
notable
performance,
especially
OR
gate,
achieving
97%
accuracy.
late
fusion
method
enhances
adaptability
by
compensating
limitations
one
modality
information
from
other.
Adapting
maintenance
identified
conditions
minimises
unnecessary
environmental
impact.
can
help
identify
roads,
substantially
improving
safety
implementing
precise
strategy
through
data-driven
approach.
Advanced Theory and Simulations,
Journal Year:
2024,
Volume and Issue:
7(7)
Published: April 30, 2024
Abstract
Effective
solar
energy
utilization
demands
improvements
in
forecasting
due
to
the
unpredictable
nature
of
irradiance
(SI).
This
study
introduces
and
rigorously
tests
two
innovative
models
across
different
locations:
Sequential
Deep
Artificial
Neural
Network
(SDANN)
Hybrid
Random
Forest
Gradient
Boosting
(RFGB).
SDANN,
leveraging
deep
learning,
aims
identify
complex
patterns
weather
data,
while
RFGB,
combining
Boosting,
proves
more
effective
by
offering
a
superior
balance
efficiency
accuracy.
The
research
highlights
SDANN
model's
learning
capabilities
along
with
RFGB
unique
blend
their
comparative
success
over
existing
such
as
eXtreme
(XGBOOST),
Categorical
(CatBOOST),
Gated
Recurrent
Unit
(GRU),
K‐Nearest
Neighbors
(KNN)
XGBOOST
hybrid.
With
lowest
Mean
Squared
Error
(147.22),
Absolute
(8.77),
high
R
2
value
(0.80)
studied
region,
stands
out.
Additionally,
detailed
ablation
studies
on
meteorological
feature
impacts
model
performance
further
enhance
accuracy
adaptability.
By
integrating
cutting‐edge
AI
SI
forecasting,
this
not
only
advances
field
but
also
sets
stage
for
future
renewable
strategies
global
policy‐making.
