Information Sciences,
Journal Year:
2023,
Volume and Issue:
657, P. 119932 - 119932
Published: Nov. 24, 2023
Sequential
decision-making
requires
balancing
multiple
conflicting
objectives
through
multi-objective
reinforcement
learning
(MORL).
Moreover,
decision-makers
desire
dense
solutions
that
satisfy
their
requirements
and
consider
the
trade-offs
between
different
(Pareto
optimal
solutions).
Most
deep
methods
focus
on
single-objective
problems
or
solve
using
simple
linear
combinations,
which
may
oversimplify
underlying
problem
lead
to
suboptimal
results.
This
study
proposes
a
neuroevolutionary
diversity
policy
search
approach
address
MORL
problems.
It
employs
neural
networks,
each
equipped
with
buffer
for
storing
recent
experiences,
representing
individuals
in
population.
The
non-dominated
sorting
method
distance
metric
are
employed
evolutionary
process
select
high-quality
as
teachers.
teachers
use
gradient-based
genetic
operators
guide
population
produce
offspring,
thereby
achieving
Pareto
solutions.
Furthermore,
we
introduce
three
benchmarks
distinct
characteristics:
(1)
continuous
sea
treasure
convex
nonconvex
fronts;
(2)
mountain
car
sparse
rewards
discontinuous
front;
(3)
HalfCheetah
high-dimensional
action-state
spaces.
experimental
results
demonstrate
superiority
of
proposed
algorithm
obtaining
Scientific Reports,
Journal Year:
2025,
Volume and Issue:
15(1)
Published: May 8, 2025
The
increasing
global
interest
in
utilizing
wind
turbines
for
power
generation
emphasizes
the
importance
of
accurate
forecasting
managing
power.
This
paper
proposed
a
framework
that
integrates
data
transformation
mechanism
with
multi-objective
none-dominated
sorting
genetic
algorithm
III
(NSGA-III),
coupled
hybrid
deep
Recurrent
Network
(DRN)
and
Long
Short-Term
Memory
(LSTM)
architecture
modeling
feature
selection
algorithm,
NSGA-III,
identifies
optimal
subset
features
from
energy
datasets.
These
selected
undergo
process
before
being
input
into
DRN-LSTM
forecasting.
A
comparative
study
demonstrates
proposal's
superior
effectiveness
robustness
compared
to
existing
frameworks
proposal
achieving
2.6593e-10
1.630e-05
terms
MSE
RMSE
respectively
whereas
classical
recorded
8.8814e-07
9.424e-04.
study's
contributions
lie
its
approach
integration
notable
enhancements
accuracy.
Furthermore,
offers
valuable
insights
guide
research
efforts
future.
Applied Soft Computing,
Journal Year:
2023,
Volume and Issue:
151, P. 111141 - 111141
Published: Dec. 13, 2023
Computer
systems
store
massive
amounts
of
data
with
numerous
features,
leading
to
the
need
extract
most
important
features
for
better
classification
in
a
wide
variety
applications.
Poor
performance
various
machine
learning
algorithms
may
be
caused
by
unimportant
that
increase
time
and
memory
required
build
classifier.
Feature
selection
(FS)
is
one
efficient
approaches
reducing
features.
This
paper,
therefore,
presents
new
FS,
named
BDE-BSS-DR,
utilizes
Binary
Differential
Evolution
(BDE),
Stochastic
Search
(BSS)
algorithm,
Dimensionality
Reduction
(DR)
mechanism.
The
BSS
algorithm
increases
search
capability
BDE
escaping
from
local
optimal
points
exploring
space.
DR
mechanism
then
reduces
dimensions
space
gradually.
As
result
using
DR,
optima
problem
wrong
removal
before
starting
process
are
reduced.
algorithm's
efficiency
evaluated
on
20
different
medical
datasets.
obtained
outcomes
indicate
BDE-BSS-DR
outperforms
BDE-BSS
significantly.
Furthermore,
effectiveness
proposed
selecting
heart
disease
data,
several
cancer
diseases,
COVID-19
also
compared
other
state-of-the-art
methods.
Our
results
show
SVM
classifier
has
significant
advantage
over
methods
an
average
accuracy
95.05%
99.40%
disease.
In
addition,
comparisons
made
KNN
prove
generating
subset
informative
