Irrigation
water
quality
is
crucial
for
sustainable
agriculture
and
environmental
health,
influencing
crop
productivity
ecosystem
balance
globally.
This
study
evaluates
the
performance
of
multiple
deep
learning
models
in
classifying
Water
Quality
Index
(IWQI),
addressing
challenge
accurate
prediction
by
examining
impact
increasing
input
complexity,
particularly
through
chemical
ions
derived
indices.
The
tested
include
convolutional
neural
networks
(CNN),
CNN-Long
Short-Term
Memory
(CNN-LSTM),
CNN-bidirectional
Long
(CNN-BiLSTM),
Gated
Recurrent
Unit
(CNN-BiGRUs).
Feature
selection
via
SHapley
Additive
exPlanations
(SHAP)
provided
insights
into
individual
feature
contributions
to
model
predictions.
objectives
were
compare
16
identify
most
effective
approach
IWQI
classification.
utilized
data
from
166
wells
Algeria’s
Naama
region,
with
70%
training
30%
testing.
Results
indicate
that
CNN-BiLSTM
outperformed
others,
achieving
an
accuracy
0.94
area
under
curve
(AUC)
0.994.
While
CNN
effectively
capture
spatial
features,
they
struggle
temporal
dependencies—a
limitation
addressed
LSTM
BiGRU
layers,
which
further
enhanced
bidirectional
processing
model.
importance
analysis
revealed
index
(qi)
qi-Na
was
significant
predictor
both
Model
15
(0.68)
(0.67).
qi-EC
showed
a
slight
decrease
importance,
0.19
0.18
between
models,
while
qi-SAR
qi-Cl
maintained
similar
levels.
Notably,
included
qi-HCO3
minor
score
0.02.
Overall,
these
findings
underscore
critical
role
sodium
levels
predictions
suggest
areas
enhancing
performance.
Despite
computational
demands
model,
results
contribute
development
robust
management,
thereby
promoting
agricultural
sustainability.
Water,
Год журнала:
2025,
Номер
17(3), С. 310 - 310
Опубликована: Янв. 23, 2025
When
confronted
with
different
influent
conditions,
WWTPs
often
lack
targeted
and
effective
operational
control
strategies.
For
the
three
typical
scenarios
of
low
C/N,
water
temperature
high
temperature,
441
carbon
source
dosage
DO
concentration
coordination
strategies
were
designed
under
premise
ensuring
effluent
quality
meets
standard.
The
purpose
was
to
provide
clear
guidance
for
efficient
operation
in
scenarios.
To
determine
optimal
strategy,
prediction
model
based
on
LSTM
GRU
constructed
testing.
results
showed
that:
(1)
LSTM-GRU
is
better
than
SVR
RF
predicting
COD
TN;
(2)
In
C/N
scenario,
should
be
controlled
between
0.23
t/h
0.26
t/h,
ranging
from
2.0
mg/L
2.6
mg/L;
(3)
0.25
0.27
2.8
(4)
0.20
2.5
mg/L.
Hydrology,
Год журнала:
2025,
Номер
12(2), С. 20 - 20
Опубликована: Янв. 21, 2025
The
forecasting
of
river
flows
and
pollutant
concentrations
is
essential
in
supporting
mitigation
measures
for
anthropogenic
climate
change
effects
on
rivers
their
environment.
This
paper
addresses
two
aspects
receiving
little
attention
the
literature:
high-resolution
(sub-daily)
data-driven
modeling
prediction
phosphorus
compounds.
It
presents
a
series
artificial
neural
networks
(ANNs)
to
forecast
soluble
reactive
(SRP)
total
(TP)
under
wide
range
conditions,
including
low
storm
events
(0.74
484
m3/s).
Results
show
correct
along
stretch
River
Swale
(UK)
with
an
anticipation
up
15
h,
at
resolutions
3
h.
concentration
improved
compared
previous
application
advection–dispersion
model.
Energies,
Год журнала:
2025,
Номер
18(4), С. 842 - 842
Опубликована: Фев. 11, 2025
Accurate
oil
and
gas
production
forecasting
is
essential
for
optimizing
field
development
operational
efficiency.
Steady-state
capacity
prediction
models
based
on
machine
learning
techniques,
such
as
Linear
Regression,
Support
Vector
Machines,
Random
Forest,
Extreme
Gradient
Boosting,
effectively
address
complex
nonlinear
relationships
through
feature
selection,
hyperparameter
tuning,
hybrid
integration,
achieving
high
accuracy
reliability.
These
maintain
relative
errors
within
acceptable
limits,
offering
robust
support
reservoir
management.
Recent
advancements
in
spatiotemporal
modeling,
Physics-Informed
Neural
Networks
(PINNs),
agent-based
modeling
have
further
enhanced
transient
forecasting.
Spatiotemporal
capture
temporal
dependencies
spatial
correlations,
while
PINN
integrates
physical
laws
into
neural
networks,
improving
interpretability
robustness,
particularly
sparse
or
noisy
data.
Agent-based
complements
these
techniques
by
combining
measured
data
with
numerical
simulations
to
deliver
real-time,
high-precision
predictions
of
dynamics.
Despite
challenges
computational
scalability,
sensitivity,
generalization
across
diverse
reservoirs,
future
developments,
including
multi-source
lightweight
architectures,
real-time
predictive
capabilities,
can
improve
forecasting,
addressing
the
complexities
supporting
sustainable
resource
management
global
energy
security.
