Journal of Physics Conference Series,
Год журнала:
2023,
Номер
2666(1), С. 012006 - 012006
Опубликована: Дек. 1, 2023
Abstract
Transmission
lines
are
directly
exposed
to
the
natural
environment
and
prone
failure
due
meteorological
factors.
A
novel
approach
for
diagnosing
transmission
line
faults
under
various
conditions
has
been
introduced.
This
method,
known
as
SDA-ISSA-XGBoost,
combines
power
of
Stacked
Denoising
Autoencoder
(SDA),
an
improved
Sparrow
Search
Algorithm
(ISSA)
enhanced
with
chaotic
mapping
sequences,
adaptive
weights,
iterative
local
search,
a
random
differential
mutation
strategy,
eXtreme
Gradient
Boosting
(XGBoost).
The
process
begins
SDA,
which
extracts
essential
features
from
initial
data.
Subsequently,
ISSA
is
applied
optimize
parameters
XGBoost
model.
results
in
ISSA-XGBoost
fault
diagnosis
performance
this
model
compared
PSO-XGBoost
SSA-XGBoost.
experimental
findings
demonstrate
that
achieves
impressive
accuracy
94.39%,
surpassing
both
SSA-XGBoost
by
6.54%
3.74%,
respectively.
Agriculture,
Год журнала:
2025,
Номер
15(2), С. 181 - 181
Опубликована: Янв. 15, 2025
An
intelligent
prediction
model
for
rice
yield
in
small-scale
cultivation
areas
can
provide
precise
forecasting
results
farmers,
planting
enterprises,
and
researchers,
holding
significant
importance
agricultural
industries
crop
science
research
within
small
regions.
Although
machine
learning
handle
complex
nonlinear
problems
to
enhance
accuracy,
further
improvements
models
are
still
needed
accurately
predict
yields
facing
environments,
thereby
enhancing
performance.
This
study
employs
four
phenotypic
traits,
namely,
panicle
angle,
length,
total
branch
grain
number,
along
with
seven
methods—multiple
linear
regression,
support
vector
machine,
MLP,
random
forest,
GBR,
XGBoost,
LightGBM—to
construct
a
group.
Subsequently,
the
top
three
best
performance
individual
predictions
integrated
using
voting
stacking
ensemble
methods
obtain
optimal
model.
Finally,
impact
of
different
traits
on
stacked
is
explored.
Experimental
indicate
that
forest
performs
after
modeling,
RMSE,
R2,
MAPE
values
0.2777,
0.9062,
17.04%,
respectively.
After
integration,
Stacking–3m
demonstrates
performance,
0.2483,
0.9250,
6.90%,
Compared
RMSE
decreased
by
10.58%,
R2
increased
1.88%,
0.76%,
indicating
improved
ensemble.
The
model,
which
demonstrated
comprehensive
evaluation
metrics,
was
selected
validation,
validation
were
satisfactory,
MAE,
8.3384,
0.9285,
0.2689,
above
findings
demonstrate
this
possesses
high
practical
value
fills
gap
Yunnan
Plateau
region.
Agricultural and Forest Meteorology,
Год журнала:
2024,
Номер
353, С. 110055 - 110055
Опубликована: Май 18, 2024
Rice
yield
depends
on
factors
including
variety,
weather,
field
management,
nutrient
and
water
availability.
We
analyzed
important
drivers
of
variability
at
the
scale,
developed
forecast
models
for
crops
in
temperate
irrigated
rice
growing
region
Australia.
fused
a
time-series
Sentinel-1
Sentinel-2
satellite
remote
sensing
imagery,
spatial
weather
data
management
information.
phenology
was
predicted
using
previously
reported
models.
Higher
yields
were
associated
with
early
flowering,
higher
chlorophyll
indices
temperatures
around
flowering.
Successive
cropping
same
lower
(p<0.001).
After
running
series
leave-one-year-out
cross
validation
experiments,
final
trained
2018–2022
data,
applied
to
predicting
1580
fields
(43,700
hectares)
from
an
independent
season
challenging
conditions
(2023).
Models
which
aggregated
phenological
periods
provided
more
accurate
predictions
than
that
these
predictors
calendar
periods.
The
accuracy
improved
as
progressed,
reaching
RMSE=1.6
t/ha
Lin's
concordance
correlation
coefficient
(LCCC)
0.67
30
days
after
flowering
level.
Explainability
SHAP
method,
revealing
likely
overall,
individual
fields.
Remote Sensing,
Год журнала:
2024,
Номер
16(7), С. 1259 - 1259
Опубликована: Апрель 2, 2024
The
timely
and
robust
prediction
of
wheat
yield
is
very
significant
for
grain
trade
food
security.
In
this
study,
the
model
was
developed
by
coupling
an
ensemble
with
multi-source
data,
including
vegetation
indices
(VIs)
meteorological
data.
results
showed
that
green
chlorophyll
index
(GCVI)
optimal
remote
sensing
(RS)
variable
predicting
compared
other
VIs.
accuracy
adaptive
boosting-
long
short-term
memory
(AdaBoost-LSTM)
higher
than
LSTM
model.
AdaBoost-LSTM
coupled
input
data
had
best
performance.
strong
robustness
under
different
irrigation
extreme
weather
events
in
general.
Additionally,
rainfed
counties
most
years
except
years.
characteristic
variables
window
from
February
to
April
smaller
requirements,
which
window.
Therefore,
can
be
accurately
predicted
one
two
months
lead
time
before
maturity
HHHP.
Overall,
achieve
accurate
large-scale
regions.
Scientific Reports,
Год журнала:
2024,
Номер
14(1)
Опубликована: Июль 13, 2024
Abstract
To
obtain
seasonable
and
precise
crop
yield
information
with
fine
resolution
is
very
important
for
ensuring
the
food
security.
However,
quantity
quality
of
available
images
selection
prediction
variables
often
limit
performance
prediction.
In
our
study,
synthesized
Landsat
MODIS
were
used
to
provide
remote
sensing
(RS)
variables,
which
can
fill
missing
values
well
cover
study
area
completely.
The
deep
learning
(DL)
was
combine
different
vegetation
index
(VI)
climate
data
build
wheat
model
in
Hebei
Province
(HB).
results
showed
that
kernel
NDVI
(kNDVI)
near-infrared
reflectance
(NIRv)
slightly
outperform
normalized
difference
(NDVI)
And
regression
algorithm
had
a
more
prominent
effect
on
prediction,
while
using
Long
Short-Term
Memory
(LSTM)
outperformed
Light
Gradient
Boosting
Machine
(LGBM).
combining
LSTM
NIRv
best
relatively
stable
single
year.
optimal
then
generate
30
m
maps
past
20
years,
higher
overall
accuracy.
addition,
we
define
optimum
time
at
April,
consider
simultaneously
lead
time.
general,
expect
this
understand
ensure
