Soil Science Society of America Journal,
Год журнала:
2023,
Номер
87(6), С. 1275 - 1284
Опубликована: Авг. 8, 2023
Abstract
Soil
thermal
conductivity
(λ)
has
broad
applications
in
soil
science,
hydrology,
and
engineering.
In
this
study,
we
applied
the
percolation‐based
effective‐medium
approximation
(P‐EMA)
to
estimate
saturation
dependence
of
()
using
data
from
38
undisturbed
samples
collected
across
state
Kansas.
The
P‐EMA
model
four
parameters
including
a
scaling
exponent
(
t
s
),
critical
water
content
(θ
c
conductivities
at
oven‐dry
(λ
dry
)
full
sat
conditions.
To
curve,
values
λ
were
measured
properties
analyzer
θ
estimated
as
function
clay
content.
Thermal
was
also
Johansen
model.
By
comparison
with
observations,
resulted
root
mean
square
error
(RMSE)
ranging
0.029
0.158
W
m
−1
K
,
whereas
had
an
RMSE
0.021
0.173
.
Our
results
demonstrate
that
comparable
accuracy
widely
used
saturation‐dependent
soils
minimal
input
parameters.
Future
studies
should
focus
on
better
understanding
physical
meaning
improve
our
ability
percolation
principles.
International Journal of Pavement Engineering,
Год журнала:
2022,
Номер
24(2)
Опубликована: Июль 11, 2022
Resilient
modulus
(MR)
plays
the
most
critical
role
in
evaluation
and
design
of
flexible
pavement
foundations.
MR
is
utilised
as
principal
parameter
for
representing
stiffness
behaviour
foundation
experimental
semi-empirical
approaches.
To
determine
MR,
cyclic
triaxial
compressive
experiments
under
different
confining
pressures
deviatoric
stresses
are
needed.
However,
such
costly
time-consuming.
In
present
study,
an
extreme
gradient
boosting-based
(XGB)
model
presented
predicting
resilient
The
optimised
using
four
optimisation
methods
(particle
swarm
(PSO),
social
spider
(SSO),
sine
cosine
algorithm
(SCA),
multi-verse
(MVO))
a
database
collected
from
previously
published
technical
literature.
outcomes
that
all
developed
designs
have
good
workability
estimating
foundation,
but
PSO−XGB
models
best
prediction
accuracy
considering
both
training
testing
datasets.
Canadian Geotechnical Journal,
Год журнала:
2023,
Номер
61(2), С. 258 - 274
Опубликована: Июнь 6, 2023
This
study
proposes
a
generalised
framework
for
developing
hybrid
machine
learning
(ML)
model
that
combines
support
vector
regression
(SVR)
with
hyperparameter
optimisation
to
predict
thermal
conductivity
(
k)
uncertainty.
The
contains
four
phases:
data
pre-processing,
determining
the
best-performing
model,
selecting
optimal
input
combination,
and
uncertainty
implementation.
A
database
containing
2197
points
is
first
compiled
train
ML
model.
Three
algorithms
are
adopted
tune
hyperparameters,
their
performance
evaluated
by
evaluation
metrics.
Results
show
SVR
Bayesian
(SVR-BO)
since
it
produces
more
accurate
predictions
k
than
models
employ
grid
random
searches.
Given
sample
insufficiency
issue
encountered
in
practice,
SVR-BO
144
combinations
analysed.
compassion
among
under
various
indicates
incorporating
temperature
as
an
additional
can
provide
moderate
improvement
accuracy
generalisability
of
Based
on
comparison,
five-input
selected
best
candidate
implement
k.
demonstrate
predicted
possesses
higher
reliability
denser
datasets
shows
promising
potential
applications
assessments.
