AQUA - Water Infrastructure Ecosystems and Society,
Год журнала:
2024,
Номер
73(8), С. 1621 - 1642
Опубликована: Июль 15, 2024
ABSTRACT
The
water
quality
of
drinking
reservoirs
directly
impacts
the
supply
safety
for
urban
residents.
This
study
focuses
on
Da
Jing
Shan
Reservoir,
a
crucial
source
Zhuhai
City
and
Macau
Special
Administrative
Region.
aim
is
to
establish
prediction
model
reservoirs,
which
can
serve
as
vital
reference
plants
when
formulating
their
plans.
In
this
research,
after
smoothing
data
using
Hodrick-Prescott
filter,
we
utilized
long
short-term
memory
(LSTM)
network
create
Reservoir.
Simulation
calculations
reveal
that
model's
fitting
degree
consistently
above
60%.
Specifically,
accuracy
pH,
dissolved
oxygen
(DO),
biochemical
demand
(BOD)
in
aligns
with
actual
results
by
more
than
70%,
effectively
simulating
reservoir's
changes.
Moreover,
parameters
such
DO,
BOD,
total
phosphorus,
relative
forecasting
error
LSTM
less
10%,
confirming
validity.
offer
an
essential
predicting
Journal of Marine Science and Engineering,
Год журнала:
2024,
Номер
12(6), С. 1013 - 1013
Опубликована: Июнь 18, 2024
The
precise
interpolation
of
oceanic
temperature
and
salinity
is
crucial
for
comprehending
the
dynamics
marine
systems
implications
global
climate
change.
Prior
neural
network-based
methods
face
constraints
related
to
their
capacity
delineate
intricate
spatio-temporal
patterns
that
are
intrinsic
ocean
data.
This
research
presents
an
innovative
approach,
known
as
Discretized
Spatial
Encoding
Neural
Network
(DSE-NN),
comprising
encoder–decoder
model
designed
on
basis
deep
supervision,
network
visualization,
hyperparameter
optimization.
Through
discretization
input
latitude
longitude
data
into
specialized
vectors,
DSE-NN
adeptly
captures
temporal
trends
augments
precision
reconstruction,
concurrently
addressing
complexity
fragmentation
characteristic
sets.
Employing
North
Atlantic
a
case
study,
this
investigation
shows
enhanced
accuracy
in
comparison
with
traditional
network.
outcomes
demonstrate
its
quicker
convergence
lower
loss
function
values,
well
ability
reflect
spatial
distribution
characteristics
physical
laws
salinity.
emphasizes
potential
providing
robust
tool
three-dimensional
reconstruction.
AQUA - Water Infrastructure Ecosystems and Society,
Год журнала:
2024,
Номер
73(8), С. 1621 - 1642
Опубликована: Июль 15, 2024
ABSTRACT
The
water
quality
of
drinking
reservoirs
directly
impacts
the
supply
safety
for
urban
residents.
This
study
focuses
on
Da
Jing
Shan
Reservoir,
a
crucial
source
Zhuhai
City
and
Macau
Special
Administrative
Region.
aim
is
to
establish
prediction
model
reservoirs,
which
can
serve
as
vital
reference
plants
when
formulating
their
plans.
In
this
research,
after
smoothing
data
using
Hodrick-Prescott
filter,
we
utilized
long
short-term
memory
(LSTM)
network
create
Reservoir.
Simulation
calculations
reveal
that
model's
fitting
degree
consistently
above
60%.
Specifically,
accuracy
pH,
dissolved
oxygen
(DO),
biochemical
demand
(BOD)
in
aligns
with
actual
results
by
more
than
70%,
effectively
simulating
reservoir's
changes.
Moreover,
parameters
such
DO,
BOD,
total
phosphorus,
relative
forecasting
error
LSTM
less
10%,
confirming
validity.
offer
an
essential
predicting