Clean Technologies and Recycling,
Год журнала:
2024,
Номер
4(2), С. 108 - 124
Опубликована: Янв. 1, 2024
<p>The
escalating
concern
over
the
adverse
effects
of
greenhouse
gas
emissions
on
Earth's
climate
has
intensified
need
for
sustainable
and
renewable
energy
sources.
Among
alternatives,
wind
emerged
as
a
key
solution
mitigating
impacts
global
warming.
The
significance
generation
lies
in
its
abundance,
environmental
benefits,
cost-effectiveness
contribution
to
security.
Accurate
forecasting
is
crucial
managing
intermittent
nature
ensuring
effective
integration
into
electricity
grid.
We
employed
machine
learning
techniques
predict
power
by
utilizing
historical
weather
data
conjunction
with
corresponding
data.
dataset
was
sourced
from
real-time
SCADA
obtained
turbines,
allowing
comprehensive
analysis.
differentiated
this
research
evaluating
not
only
conditions
but
also
meteorological
factors
physical
measurements
turbine
components,
thus
considering
their
combined
influence
overall
production.
utilized
Decision
Tree,
Random
Forest,
K-Nearest
Neighbors
(KNN),
XGBoost
algorithms
estimate
generation.
performance
these
models
assessed
using
evaluation
criteria:
R<sup>2</sup>,
Mean
Absolute
Error
(MAE),
Squared
(MSE),
Root
(RMSE),
Percentage
(MAPE).
findings
indicated
algorithm
outperformed
other
models,
achieving
high
accuracy
while
demonstrating
computational
efficiency,
making
it
particularly
suitable
applications
forecasting.</p>
With
the
increasing
capacity
of
grid-connected
wind
power
systems,
forecasting
has
become
a
major
research
problem
in
systems
under
background
dual-carbon
policy,
and
it
is
great
practical
significance
to
develop
reliable
methods.
In
order
overcome
difficulties
data
noise
reduction,
feature
extraction
uncertainty
estimation,
new
system
proposed.
The
improved
variational
mode
decomposition
algorithm
used
denoise
data,
overcoming
subjective
parameter
selection
traditional
method.
time
convolutional
network,
Transformer
bidirectional
long
short-term
memory
network
are
extract
sequence
features
comprehensively
ensure
that
local,
long-term,
considered
simultaneously.
multi-objective
Bayesian
optimization
achieve
Pareto
optimal
solution,
quantile
regression
set
for
interval
forecasting,
so
as
systematically
enhance
model
ability.
performance
evaluated
based
on
two
different
datasets
England,
taking
Penmanshiel
farm
an
example,
at
confidence
level
0.10,
MAE
RMSE
values
low
17.23
21.25
respectively,
while
WS
value
high
74.10%.
experimental
results
show
proposed
better
point
ability
than
comparison
model.
