Industrial & Engineering Chemistry Research,
Год журнала:
2024,
Номер
63(44), С. 18940 - 18956
Опубликована: Окт. 25, 2024
Carbon
materials
possess
active
sites
and
functionalities
on
the
surface
that
can
attract
prominent
interest
as
solid
adsorbents
for
diverse
gas
adsorption.
This
study
aimed
to
predict
optimized
methane
uptake,
adsorption
energy
(Ead),
adsorbent
rediscovery
through
multitechniques
of
neural,
regression,
classifier
ML-DFT,
Uniform
Manifold
Approximation
Projection
(UMAP).
Nitrogen
oxygen
(N/O)
graphene,
graphene
oxide
(GO),
N-doped
GO
were
applied
storage
medium.
Multi-ML
algorithms
employed
CH4
uptake
(i)
N/O
such
pyridinic
(N-py),
carboxyl
(O–II),
oxidized
(N-x),
hydroxyl
(O-h),
Nitroso
(N-ni),
Amine
(primary,
secondary,
tertiary).
(ii)
The
surfaces
are
decorated
with
heteroatoms
construct
(GO)
GO.
DFT
calculations
by
PW91
Dmol3
package.
N/O-functionalities
in
distance
∼2.0
3.1
Å
groups
obtained
Ead
approximately
−2.0
−4
eV.
Further,
ML
models
accomplished
forthcoming
physisorption
using
multiadsorptive
features
an
R2
0.99.
ML-derived
sensitivity
analysis
approach
was
specifications
deformation
energy,
functionality
type,
structure.
indicate
levels
−0.03
0.02
synergetic
DFT/ML
approaches
distinguished
modeled
rediscovered
phases
functional
structures.
UMAP
is
a
new
screening
play
complementary
role
modeling
process.
International Journal of Hydrogen Energy,
Год журнала:
2024,
Номер
58, С. 485 - 494
Опубликована: Янв. 25, 2024
Underground
hydrogen
storage
(UHS)
offers
a
promising
approach
for
the
of
significant
volumes
gas
(H2)
within
deep
geological
formations,
which
can
later
be
utilized
energy
generation
when
necessary.
Interfacial
tension
(IFT)
between
H2
and
formation
brine
plays
vital
role
in
influencing
distribution
at
pore
scale
and,
ultimately,
capacity.
In
this
research,
we
developed
four
intelligent
models:
Decision
Trees
(DT),
Random
Forests
(RF),
Support
Vector
Machines
(SVM),
Multi-Layer
Perceptron
(MLP).
These
models
were
designed
to
predict
IFT
utilizing
pressure,
temperature,
molality.
Additionally,
fine-tuned
three
explicit
correlations
previously
our
research.
To
assess
influence
each
parameter
on
IFT,
conducted
comprehensive
analysis
raw
data
exclude
doubtful
samples.
This
was
followed
by
rigorous
model
development,
including
hyperparameter
tuning,
finally,
an
examination
using
testing
data.
The
results
clearly
demonstrate
superiority
RF
model,
achieving
high
accuracy
reliability
with
coefficients
determination
(R2),
root
mean
square
error
(RMSE),
average
absolute
relative
deviation
(AARD)
values
0.96,
1.50,
1.84
%,
respectively.
exemplary
performance
attributed
its
inherent
characteristics.
ensemble
excels
capturing
complex
relationships,
thereby
enhancing
predictive
solidifying
over
other
study.
Furthermore,
feature
importance
revealed
that
temperature
has
most
influence,
molality
pressure.
Moreover,
assessed
these
through
external
not
used
initial
training
stages.
Our
study
highlights
exceptional
power
emphasizing
practical
selecting
enhanced
reliability.
proposed
method
shows
potential
industrial
applications,
especially
optimizing
underground
storage.
International Petroleum Technology Conference,
Год журнала:
2025,
Номер
unknown
Опубликована: Фев. 17, 2025
Abstract
A
practical
solution
to
energy
transition
and
the
increasing
demand
for
is
underground
hydrogen
storage
(UHS).
The
contribution
of
(H2)
as
a
clean
source
has
proven
be
an
effective
substitute
future
use
meet
net-zero
target
reduce
anthropogenic
greenhouse
gas
emissions.
One
most
important
factors
affecting
H2
displacement
capacity
under
geological
circumstances
column
height.
objective
this
study
underscore
importance
large-scale
reliable
machine
learning
algorithms
evaluate
predict
height
varied
thermophysical
salinity
conditions.
In
study,
dataset
540
datapoints
evaluation
prediction
generated,
which
involves
three
main
parameters:
density
difference
(Δρ),
interfacial
tension
(IFT)
contact
angle
(θ).
correlation
angles
against
various
reservoir
depths
used
evaluated.
Thermophysical
conditions
include
pressures
(0.1-20
MPa),
temperatures
(25-70°C),
salinities
including
deionized
water,
seawater
brines
1
3
molar
concentrations
salts
(NaCl,
KCl,
MgCl2,
CaCl2,
Na2SO4)
from
our
experimental
data.
(h)
predicted
using
(ML)
models,
viz.,
random
forest
(RF),
decision
tree
(DT)
gradient
boosting
(GB).
Statistical
data
analysis
performed
generate
distribution
coefficient
calculated
while
feature
determined
identify
relationship
each
input
parameter
with
output
Pearson,
Spearman,
Kendall
models.
RF
GB,
demonstrated
in
have
shown
promising
results
providing
accurate
predictions
maintaining
generalizability.
Various
error
assessment
metrics
MSE,
RMSE,
MAPE
R2
are
utilized
evaluation.
Prediction
resulted
values
0.995
training
0.999
testing
model.
Whereas
GB
model
also
superior
performance
0.997
during
phase
phase.
However,
DT
0.994
phases
respectively.
While
MSE
value
0
obtained
indicated
overfitting.
findings
suggest
that
data-driven
ML
models
can
powerful
tool
accurately
predicting
effectively
determine
capacity,
reducing
time
cost
associated
determination
traditional
methods.
addition,
advanced
explored
overcome
challenges
pertinent
