Water Resources Research,
Год журнала:
2023,
Номер
59(9)
Опубликована: Сен. 1, 2023
Abstract
Accurate
runoff
forecasting
plays
a
vital
role
in
issuing
timely
flood
warnings.
Whereas,
previous
research
has
primarily
focused
on
historical
and
precipitation
variability
while
disregarding
other
factors'
influence.
Additionally,
the
prediction
process
of
most
machine
learning
models
is
opaque,
resulting
low
interpretability
model
predictions.
Hence,
this
study
develops
an
ensemble
deep
to
forecast
from
three
hydrological
stations.
Initially,
time‐varying
filtered
based
empirical
mode
decomposition
employed
decompose
series
into
several
internal
functions
(IMFs).
Subsequently,
complexity
each
IMF
component
evaluated
by
multi‐scale
permutation
entropy,
IMFs
are
classified
high‐
low‐frequency
portions
entropy
values.
Considering
high‐frequency
still
exhibit
great
volatility,
robust
local
mean
adopted
perform
secondary
portions.
Then,
meteorological
variables
processed
Relief
algorithm
variance
inflation
factor
features
as
inputs,
individual
subsequences
preliminary
outputs
bidirectional
gated
recurrent
unit
extreme
models.
Random
forests
(RF)
introduced
nonlinear
predicted
sub‐models
obtain
final
results.
The
proposed
outperforms
various
evaluation
metrics.
Meanwhile,
due
opaque
nature
models,
shapley
assess
contribution
selected
variable
long‐term
trend
runoff.
could
serve
essential
reference
for
precise
warning.
Energy Reports,
Год журнала:
2022,
Номер
8, С. 8965 - 8980
Опубликована: Июль 14, 2022
As
a
clean
and
renewable
energy
source,
wind
power
is
of
great
significance
for
addressing
global
shortages
environmental
pollution.
However,
the
uncertainty
speed
hinders
direct
use
power,
resulting
in
high
proportion
abandoned
wind.
Therefore,
accurate
prediction
improving
utilization
rate
energy.
In
this
study,
hybrid
model
proposed
based
on
seasonal
autoregressive
integrated
moving
average
(SARIMA),
ensemble
empirical
mode
decomposition
(EEMD),
long
short-term
memory
(LSTM)
methods.
First,
original
data
were
resampled
to
obtain
within
time
scales
15,
30,
60
min.
The
SARIMA
was
used
extract
linear
features
nonlinear
residual
sequences
series
at
different
scales,
EEMD
decompose
sequence
intrinsic
functions
(IMFs)
sub-residual
sequences.
For
IMFs
obtained
after
decomposition,
LSTM
method
training,
predicted
IMFs,
sequence,
series,
final
speed.
To
verify
superiority
large
farm
as
case
study.
Finally,
compared
with
other
models,
verifying
that
experimental
has
higher
accuracy.
Measurement Science and Technology,
Год журнала:
2022,
Номер
34(4), С. 045005 - 045005
Опубликована: Дек. 15, 2022
Abstract
It
confronts
great
difficulty
to
apply
the
traditional
rolling
bearing
fault
diagnosis
methods
adaptively
extract
features
conducive
under
complex
operating
conditions,
and
obtaining
numerous
data
real
conditions
is
difficult
costly.
To
address
this
problem,
a
method
based
on
two-dimensional
time-frequency
images
augmentation
proposed.
begin
with,
original
one-dimensional
time
series
signal
converted
into
by
continuous
wavelet
transform
obtain
input
suitable
for
convolutional
neural
network
(CNN).
Secondly,
technique
employed
expand
labeled
data.
Finally,
generated
are
served
as
training
samples
train
model
CNNs.
Experimental
studies
conducted
standard
real-world
datasets
validate
proposed
demonstrate
its
superiority
over
in
detecting
faults.
