International Journal of Electrochemical Science,
Journal Year:
2024,
Volume and Issue:
unknown, P. 100686 - 100686
Published: June 1, 2024
The
transportation
of
oil
and
gas
relies
heavily
on
pipelines,
pipeline
corrosion
is
a
major
factor
affecting
reliability.
It
can
lead
to
failure
other
damage.
Pipeline
prediction
great
importance
for
integrity
management
prevention.
A
physical
law
intervening
RF(Random
Forest)-PSO(Particle
Swarm
Optimization)-BP(Back
Propagation
Neural
Network)
algorithm
proposed
predict
rate.
DeWaard
model
first
fitted
the
data,
predicts
form
new
feature,
which
then
combined
with
features
extracted
by
RF
feature
that
used
as
an
input
metric
data-driven
model.
Secondly,
already
constructed
are
divided
into
training
set
testing
set.
train
PSO-BP
model,
test
accuracy
evaluated
using
metrics
such
MAE,
MBE,
MAPE,
R2.
To
show
superiority
compared
models.
results
has
some
advantages
in
both
analysis
prediction,
it
theoretical
guidance
protection.
npj Materials Degradation,
Journal Year:
2024,
Volume and Issue:
8(1)
Published: Aug. 26, 2024
Abstract
Accurately
assessing
the
residual
strength
of
corroded
oil
and
gas
pipelines
is
crucial
for
ensuring
their
safe
stable
operation.
Machine
learning
techniques
have
shown
promise
in
addressing
this
challenge
due
to
ability
handle
complex,
non-linear
relationships
data.
Unlike
previous
studies
that
primarily
focused
on
enhancing
prediction
accuracy
through
optimization
single
models,
work
shifts
emphasis
a
different
approach:
stacking
ensemble
learning.
This
study
applies
model
composed
seven
base
learners
three
meta-learners
predict
using
dataset
453
instances.
Automated
hyperparameter
tuning
libraries
are
utilized
search
optimal
hyperparameters.
By
evaluating
various
combinations
meta-learners,
configuration
was
determined.
The
results
demonstrate
model,
k-nearest
neighbors
as
meta-learner
alongside
learners,
delivers
best
predictive
performance,
with
coefficient
determination
0.959.
Compared
individual
also
significantly
improves
generalization
performance.
However,
model’s
effectiveness
low-strength
limited
small
sample
size.
Furthermore,
incorporating
original
features
into
second-layer
did
not
enhance
likely
because
first-layer
had
already
extracted
most
critical
features.
Given
marginal
contribution
accuracy,
offers
novel
perspective
improving
findings
important
practical
implications
integrity
assessment
pipelines.
Applied Sciences,
Journal Year:
2025,
Volume and Issue:
15(7), P. 4031 - 4031
Published: April 6, 2025
The
accurate
prediction
of
the
residual
strength
defective
pipelines
is
a
critical
prerequisite
for
ensuring
safe
operation
oil
and
gas
pipelines,
it
holds
significant
implications
pipeline’s
remaining
service
life
preventive
maintenance.
Traditional
machine
learning
algorithms
often
fail
to
comprehensively
account
correlative
factors
influencing
exhibit
limited
capability
in
extracting
nonlinear
features
from
data,
suffer
insufficient
predictive
accuracy.
Furthermore,
models
typically
lack
interpretability.
To
address
these
issues,
this
study
proposes
hybrid
model
based
on
Bayesian
optimization
(BO)
eXtreme
Gradient
Boosting
(XGBoost).
This
approach
resolves
issues
excessive
iterations
high
computational
costs
associated
with
conventional
hyperparameter
methods,
significantly
enhancing
model’s
performance.
performance
evaluated
using
mainstream
metrics
such
as
Mean
Absolute
Percentage
Error
(MAPE),
Coefficient
Determination
(R2),
Root
Square
(RMSE),
robustness
analysis,
overfitting
grey
relational
analysis.
enhance
interpretability
predictions,
reveal
significance
features,
confirm
prior
domain
knowledge,
Shapley
additive
explanations
(SHAP)
are
employed
conduct
relevant
research.
results
indicate
that,
compared
Random
Forest,
LightGBM,
Support
Vector
Machine,
gradient
boosting
regression
tree,
Multi-Layer
Perceptron,
BO-XGBoost
exhibits
best
performance,
MAPE,
R2,
RMSE
values
5.5%,
0.971,
1.263,
respectively.
Meanwhile,
proposed
demonstrates
highest
robustness,
least
tendency
overfitting,
most
relation
degree
value.
SHAP
analysis
reveals
that
ranked
descending
order
importance,
defect
depth
(d),
wall
thickness
(t),
yield
(σy),
external
diameter
(D),
length
(L),
tensile
(σu),
width
(w).
development
contributes
improving
integrity
management
provides
decision
support
intelligent
fields.
