IEEE Access,
Год журнала:
2024,
Номер
12, С. 47830 - 47841
Опубликована: Янв. 1, 2024
Due
to
the
characteristics
of
strong
suddenness,
high
harmfulness,
and
frequent
occurrence
mountain
flood
disasters
in
small
watersheds,
accuracy
reliability
forecasting
are
insufficient
watersheds.
This
paper
studies
key
theories
technologies,
that
is
uncertainty
model
based
on
hydrologic
physical
mechanism.
We
design
Bayesian
Deep
Learning
(DL)
models,
it
suitable
for
transfer
spatiotemporal
factors
caused
by
floods
disaster
probability.
The
models
include
Linear
Long
Short-Term
Memory
(LSTM)
model,
we
hope
achieve
an
acceptable
balance
between
(uncertainty
confidence
coverage)
(confidence
interval
width).
Meanwhile,
extract
effective
information
from
multi-source
multi-dimensional
hazard
factors'
big
data.
experiment
shows
differences
DL
have
long-term
probability
ability
at
both,
but
LSTM
superior
terms
reliability,
computational
consumption.
Results in Engineering,
Год журнала:
2024,
Номер
22, С. 102104 - 102104
Опубликована: Апрель 10, 2024
Forecasting
streamflows,
essential
for
flood
mitigation
and
the
efficient
management
of
water
resources
drinking,
agriculture
hydroelectric
power
generation,
presents
a
formidable
challenge
in
most
real-world
scenarios.
In
this
study,
two
models,
first
based
on
Additive
Regression
Radial
Basis
Function
Neural
Networks
(AR-RBF)
second
stacking
with
Pace
Multilayer
Perceptron
Random
Forest
(MLP-RF-PR),
were
compared
prediction
short-term
(1–3
days
ahead)
medium-term
(7
daily
streamflow
rates
three
different
rivers
Germany:
Elbe
River
at
Wittenberge,
Leine
Herrenhausen,
Saale
Hof
The
lagged
values
rate,
precipitation
temperature
considered
modeling.
Moreover,
Bayesian
Optimization
(BO)
algorithm
was
used
to
assess
optimal
number
hyperparameters.
Both
models
showed
accurate
predictions
forecasting,
R2
1-day
ahead
ranging
from
0.939
0.998
AR-RBF
0.930
0.996
MLP-RF-PR,
while
MAPE
ranged
2.02
%
8.99
2.14
9.68
when
exogeneous
variables
included.
As
forecast
horizon
increased,
reduction
forecasting
accuracy
observed.
However,
both
could
still
predict
overall
flow
pattern,
even
7-day-ahead
predictions,
0.772
0.871
0.703
0.840
10.60
20.45
10.44
19.65
MLP-RF-PR.
Overall,
outcomes
study
suggest
that
MLP-RF-PR
can
be
reliable
tools
short-
rate
prediction,
requiring
short
parameters
optimized,
making
them
easy
implement
reducing
calculation
time
required.
Water Practice & Technology,
Год журнала:
2024,
Номер
19(6), С. 2442 - 2459
Опубликована: Июнь 1, 2024
ABSTRACT
Measurement
inaccuracies
and
the
absence
of
precise
parameters
value
in
conceptual
analytical
models
pose
challenges
simulating
rainfall–runoff
modeling
(RRM).
Accurate
prediction
water
resources,
especially
scarcity
conditions,
plays
a
distinctive
pivotal
role
decision-making
within
resource
management.
The
significance
machine
learning
(MLMs)
has
become
pronounced
addressing
these
issues.
In
this
context,
forthcoming
research
endeavors
to
model
RRM
utilizing
four
MLMs:
Support
Vector
Machine,
Gene
Expression
Programming
(GEP),
Multilayer
Perceptron,
Multivariate
Adaptive
Regression
Splines
(MARS).
simulation
was
conducted
Malwathu
Oya
watershed,
employing
dataset
comprising
4,765
daily
observations
spanning
from
July
18,
2005,
September
30,
2018,
gathered
rainfall
stations,
Kappachichiya
hydrometric
station.
Of
all
input
combinations,
incorporating
Qt−1,
Qt−2,
R̄t
identified
as
optimal
configuration
among
considered
alternatives.
models'
performance
assessed
through
root
mean
square
error
(RMSE),
average
(MAE),
coefficient
determination
(R2),
developed
discrepancy
ratio
(DDR).
GEP
emerged
superior
choice,
with
corresponding
index
values
(RMSE,
MAE,
R2,
DDRmax)
(43.028,
9.991,
0.909,
0.736)
during
training
process
(40.561,
10.565,
0.832,
1.038)
testing
process.
