Scientific Reports,
Год журнала:
2023,
Номер
13(1)
Опубликована: Окт. 16, 2023
Traditional
linear
regression
and
neural
network
models
demonstrate
suboptimal
fit
lower
predictive
accuracy
while
the
quality
of
electrolytic
copper
is
estimated.
A
more
dependable
accurate
model
essential
for
these
challenges.
Notably,
maximum
information
coefficient
was
employed
initially
to
discern
non-linear
correlation
between
nineteen
factors
influencing
five
control
indicators.
Additionally,
random
forest
algorithm
elucidated
primary
governing
quality.
hybrid
model,
integrating
particle
swarm
optimization
with
least
square
support
vector
machine,
devised
predict
based
on
factors.
Concurrently,
a
combining
relevance
machine
developed,
focusing
The
outcomes
indicate
that
identified
principal
quality,
corroborated
by
analysis
via
coefficient.
when
accounting
all
factors,
comparable
optimization-least
surpassed
both
conventional
models.
error
forest-relevance
notably
less
than
sole
index
being
under
5%.
intricate
variation
pattern
influenced
numerous
unveiled.
advanced
circumvents
deficiencies
seen
in
findings
furnish
valuable
insights
management.
Ecological Informatics,
Год журнала:
2024,
Номер
82, С. 102695 - 102695
Опубликована: Июнь 20, 2024
Accurate
and
efficient
long-term
prediction
of
marine
dissolved
oxygen
(DO)
is
crucial
for
the
sustainable
development
aquaculture.
However,
multidimensional
time
dependency
lag
effects
chemical
variables
present
significant
challenges
when
handling
multiple
inputs
in
univariate
tasks.
To
address
these
issues,
we
designed
a
multivariate
time-series
model
(LMFormer)
based
on
Transformer
architecture.
The
proposed
decomposition
strategy
effectively
leverages
feature
information
at
different
scales,
thereby
reducing
loss
critical
information.
Additionally,
dynamic
variable
selection
gating
mechanism
was
to
optimize
collinearity
problem
data
extraction
process.
Finally,
an
two-stage
attention
architecture
capture
long-range
dependencies
between
features.
This
study
conducted
high-precision
7-day
advance
DO
predictions
two
case
studies,
environmentally
stable
Shandong
Peninsula
China
San
Juan
Islands
United
States,
which
are
affected
by
extreme
conditions
such
as
ocean
currents.
experimental
results
demonstrate
superior
performance
generalizability
model.
In
case,
mean
absolute
error
(MAE),
root
square
(RMSE),
coefficient
determination
(R2),
Kling–Gupta
efficiency
(KGE)
reached
0.0159,
0.126,
0.9743,
0.9625,
respectively.
MAE
reduced
average
42.34%
compared
that
baseline
model,
RMSE
24.57%,
R2
increased
22.54%,
KGE
improved
12.04%.
Overall,
achieves
data,
providing
valuable
references
management
decision-making
Water,
Год журнала:
2024,
Номер
16(24), С. 3616 - 3616
Опубликована: Дек. 15, 2024
Surface
waterbodies
are
heavily
exposed
to
pollutants
caused
by
natural
disasters
and
human
activities.
Empowering
sensor
technologies
in
water
quality
monitoring,
sufficient
measurements
have
become
available
develop
machine
learning
(ML)
models.
Numerous
ML
models
quickly
been
adopted
predict
indicators
various
surface
waterbodies.
This
paper
reviews
78
recent
articles
from
2022
October
2024,
categorizing
utilizing
into
three
groups:
Point-to-Point
(P2P),
which
estimates
the
current
target
value
based
on
other
at
same
time
point;
Sequence-to-Point
(S2P),
utilizes
previous
series
data
one
point
ahead;
Sequence-to-Sequence
(S2S),
uses
forecast
sequential
values
future.
The
used
each
group
classified
compared
according
indicators,
availability,
model
performance.
Widely
strategies
for
improving
performance,
including
feature
engineering,
hyperparameter
tuning,
transfer
learning,
recognized
described
enhance
effectiveness.
interpretability
limitations
of
applications
discussed.
review
provides
a
perspective
emerging
Scientific Reports,
Год журнала:
2023,
Номер
13(1)
Опубликована: Окт. 16, 2023
Traditional
linear
regression
and
neural
network
models
demonstrate
suboptimal
fit
lower
predictive
accuracy
while
the
quality
of
electrolytic
copper
is
estimated.
A
more
dependable
accurate
model
essential
for
these
challenges.
Notably,
maximum
information
coefficient
was
employed
initially
to
discern
non-linear
correlation
between
nineteen
factors
influencing
five
control
indicators.
Additionally,
random
forest
algorithm
elucidated
primary
governing
quality.
hybrid
model,
integrating
particle
swarm
optimization
with
least
square
support
vector
machine,
devised
predict
based
on
factors.
Concurrently,
a
combining
relevance
machine
developed,
focusing
The
outcomes
indicate
that
identified
principal
quality,
corroborated
by
analysis
via
coefficient.
when
accounting
all
factors,
comparable
optimization-least
surpassed
both
conventional
models.
error
forest-relevance
notably
less
than
sole
index
being
under
5%.
intricate
variation
pattern
influenced
numerous
unveiled.
advanced
circumvents
deficiencies
seen
in
findings
furnish
valuable
insights
management.
