Energies,
Journal Year:
2024,
Volume and Issue:
17(18), P. 4739 - 4739
Published: Sept. 23, 2024
Photovoltaic
(PV)
power
generation
is
highly
stochastic
and
intermittent,
which
poses
a
challenge
to
the
planning
operation
of
existing
systems.
To
enhance
accuracy
PV
prediction
ensure
safe
system,
novel
approach
based
on
seasonal
division
periodic
attention
mechanism
(PAM)
for
proposed.
First,
dataset
divided
into
three
components
trend,
period,
residual
under
fuzzy
c-means
clustering
(FCM)
decomposition
(SD)
method
according
four
seasons.
Three
independent
bidirectional
long
short-term
memory
(BiLTSM)
networks
are
constructed
these
subsequences.
Then,
network
optimized
using
improved
Newton–Raphson
genetic
algorithm
(NRGA),
innovative
PAM
added
focus
characteristics
data.
Finally,
results
each
component
summarized
obtain
final
results.
A
case
study
Australian
DKASC
Alice
Spring
plant
demonstrates
performance
proposed
approach.
Compared
with
other
paper
models,
MAE,
RMSE,
MAPE
evaluation
indexes
show
that
has
excellent
in
predicting
output
stability.
Results in Engineering,
Journal Year:
2024,
Volume and Issue:
21, P. 101886 - 101886
Published: Feb. 8, 2024
Photovoltaic
(PV)
panels
stand
as
a
prominent
solution
to
meet
the
world's
growing
energy
demands,
due
their
resistance
hard
climate
conditions,
low-cost
maintenance,
and
long
lifetime.
Nonetheless,
integration
of
PV
power
into
electrical
grids
poses
significant
challenge
its
inherent
intermittency.
This
study
aims
forecast
future
based
on
historical
records
using
Bidirectional
Long
Short-Term
Memory
(Bi-LSTM),
One-Dimensional
Convolutional
Neural
Network
(1D-CNN),
Gated
Recurrent
Unit
(GRU).
Various
performance
metrics
have
been
used
evaluate
compare
accuracy
three
models,
including
mean
squared
error,
root
absolute
max
error
R-squared
for
evaluation.
The
prediction
photovoltaic
values
registered
in
last
hour
was
carried
out.
Two
scenarios
investigated,
with
without
nighttime
fit
models.
Results
reveal
that
forecasting
models
provide
exceptional
accuracy,
achieving
correlation
coefficient
range
96.9–97.2%
daytime
both
scenarios,
indicating
promising
potential
these
DNNs
forecasters
optimizing
production
improving
overall
system
efficiency.
GRU,
BiLSTM-based
showed
identical
results
terms
RMSE,
MSE
MAE
while
1D-CNN
forecaster
accurate
second
scenario,
However,
despite
this
improvement,
it
still
falls
behind
Bi-LSTM
or
GRU
scenarios.
Journal of Forecasting,
Journal Year:
2024,
Volume and Issue:
43(6), P. 2064 - 2087
Published: March 11, 2024
Abstract
Wind
power
has
emerged
as
a
successful
component
within
systems.
The
ability
to
reliably
and
accurately
forecast
wind
speed
is
of
great
importance
in
maintaining
the
security
stability
grid.
However,
significance
explaining
prediction
models
often
been
overlooked
by
researchers.
To
address
this
gap,
study
introduces
novel
approach
forecasting
that
incorporates
significant
decomposition
method,
attention‐based
machine
learning,
local
explanation
techniques.
proposed
model
utilizes
grid
search
variational
mode
decompose
sequence
into
different
modes
while
employing
gate
recurrent
unit
with
an
attention
mechanism
achieve
superior
performance.
Experimental
evaluations
conducted
on
eight
real‐world
datasets
demonstrate
outperforms
other
popular
across
multiple
performance
criteria.
In
two
specific
experiments,
achieved
minimal
mean
absolute
percentage
error
2.74%
1.70%,
respectively.
Furthermore,
interpretable
model‐agnostic
explanations
(LIME)
were
employed
assess
influence
factors,
highlighting
whether
they
positively
or
negatively
affected
predicted
values.
