Journal of Geophysical Research Atmospheres,
Год журнала:
2024,
Номер
129(20)
Опубликована: Окт. 10, 2024
Abstract
Different
types
of
weather
events,
including
tropical
cyclones
(TCs),
mesoscale
convective
systems
(MCSs),
and
atmospheric
rivers
(ARs),
significantly
impact
precipitation
patterns
in
East
Asia.
This
study
pioneers
the
application
deep
learning
(DL)
methods,
convolutional
neural
network,
U‐Net,
Attention
U‐Net
models,
to
simulate
associated
with
these
events.
The
spatial
permutation
method
is
also
used
identify
key
meteorological
variables
for
accurately
generating
DL
models.
models
trained
on
all
timeslots
consistently
surpass
performance
state‐of‐the‐art
numerical
simulations,
although
their
efficacy
slightly
diminishes
during
extreme
outperformance
attributed
appropriate
emphasis
that
capture
processes,
such
as
low‐level
moisture
mid‐level
pressure
fields.
However,
new
separately
TCs,
MCSs,
ARs
using
clipped
output
does
not
exceed
previous
Among
input
features,
contribute
most
at
low
intensity,
while
importance
other
increases
more
intense
precipitation,
some
discrepancies
vary
across
event
types.
results
further
reveal
detailed
locations
are
essential
simulating
related
areas
high
specific
humidity
strong
winds.
could
acquire
useful
information
from
region
remote
events
improve
simulation.
Overall,
serve
promising
tools
enhancing
our
understanding
various
Frontiers in Forests and Global Change,
Год журнала:
2023,
Номер
6
Опубликована: Дек. 8, 2023
Introduction
Atmospheric
temperature
affects
the
growth
and
development
of
plants
has
an
important
impact
on
sustainable
forest
ecological
systems.
Predicting
atmospheric
is
crucial
for
management
planning.
Methods
Artificial
neural
network
(ANN)
deep
learning
models
such
as
gate
recurrent
unit
(GRU),
long
short-term
memory
(LSTM),
convolutional
(CNN),
CNN-GRU,
CNN-LSTM,
were
utilized
to
predict
change
monthly
average
extreme
temperatures
in
Zhengzhou
City.
Average
data
from
1951
2022
divided
into
training
sets
(1951–2000)
prediction
(2001–2022),
22
months
used
model
input
next
month.
Results
Discussion
The
number
neurons
hidden
layer
was
14.
Six
different
algorithms,
along
with
13
various
functions,
trained
compared.
ANN
evaluated
terms
correlation
coefficient
(R),
root
mean
square
error
(RMSE),
absolute
(MAE),
good
results
obtained.
Bayesian
regularization
(trainbr)
best
performing
algorithm
predicting
average,
minimum
maximum
compared
other
algorithms
R
(0.9952,
0.9899,
0.9721),
showed
lowest
values
RMSE
(0.9432,
1.4034,
2.0505),
MAE
(0.7204,
1.0787,
1.6224).
CNN-LSTM
performance.
This
method
had
generalization
ability
could
be
forecast
areas.
Future
climate
changes
projected
using
model.
temperature,
2030
predicted
17.23
°C,
−5.06
42.44
whereas
those
2040
17.36
−3.74
42.68
respectively.
These
suggest
that
continue
warming
future.
Engineering Applications of Artificial Intelligence,
Год журнала:
2024,
Номер
133, С. 108156 - 108156
Опубликована: Март 6, 2024
Ground
settlement
prediction
during
mechanized
tunneling
is
of
paramount
importance
and
remains
a
challenging
research
topic.
Typically,
two
paradigms
are
existing:
physics-driven
approach
utilizing
numerical
simulation
models
for
prediction,
data-driven
employing
machine
learning
techniques
to
learn
mappings
between
influencing
factors
the
settlement.
To
integrate
advantages
both
approaches
assimilate
data
from
different
sources,
we
propose
multi-fidelity
deep
operator
network
(DeepONet)
framework,
leveraging
recently
developed
methods.
The
presented
framework
comprises
components:
low-fidelity
subnet
that
captures
fundamental
ground
patterns
obtained
finite
element
simulations,
high-fidelity
learns
nonlinear
correlation
real
engineering
monitoring
data.
A
pre-processing
strategy
causality
adopted
consider
spatio-temporal
characteristics
tunnel
excavation.
results
show
proposed
method
can
effectively
capture
physical
information
provided
by
simulations
accurately
fit
measured
(R2
around
0.9)
as
well.
Notably,
even
when
dealing
with
very
limited
noisy
(with
50%
error),
model
robust,
achieving
satisfactory
R2>0.8.
In
comparison,
R2
score
pure
simulation-based
only
0.2.
utilization
transfer
significantly
reduces
training
time
20
min
within
30
s,
showcasing
potential
our
real-time
construction.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Май 27, 2024
Abstract
Precipitation
due
to
its
complex
nature
requires
a
comprehensive
model
for
forecasting
purposes
and
the
efficiency
of
improved
ARIMA
(IARIMA)
forecasts
has
been
proved
relative
conventional
models.
This
study
used
two
procedures
in
structure
IARIMA
obtain
accurate
monthly
precipitation
four
stations
located
northern
Iran;
Bandar
Anzali,
Rasht,
Ramsar,
Babolsar.
The
first
procedure
applied
support
vector
regression
(SVR)
modeling
statistical
characteristics
each
class,
IARIMA-SVR,
which
evaluation
metrics
so
that
decrease
Theil's
coefficient
average
variance
all
was
21.14%
17.06%,
respectively.
Two
approaches
are
defined
second
includes
forecast
combination
(C)
scheme,
IARIMA-C-particle
swarm
optimization
(PSO),
artificial
intelligence
technique.
Generally,
most
time,
IARIMA-C-PSO
other
approach,
exhibited
acceptable
results
accuracy
improvement
greater
than
zero
at
stations.
Comparing
procedures,
it
is
found
capability
higher
concerning
normalized
mean
squared
error
value
from
IARIMA-SVR
36.72%
39.92%,
respectively
residual
predictive
deviation
(RPD)
2,
indicates
high
performance
model.
With
investigation,
Anzali
station
better
By
developing
an
model,
one
can
achieve
identifying
time
series
issues
interest
importance.
