SSRN Electronic Journal,
Journal Year:
2022,
Volume and Issue:
unknown
Published: Jan. 1, 2022
Multi-Start
metaheuristics
(MSM)
are
commonly
used
to
solve
vehicle
routing
problems
(VRPs).
These
methods
create
different
initial
solutions
and
improve
them
through
local-search.
The
goal
of
these
is
deliver
the
best
solution
found.
We
introduce
initial-solution
classification
(ISC)
predict
if
a
local-search
algorithm
should
be
applied
in
MSM.
This
leads
faster
convergence
MSM
higher-quality
when
amount
computation
time
limited.
In
this
work,
we
extract
known
features
capacitated
VRP
(CVRP)
additional
features.
With
machine-learning
classifier
(random
forest),
show
how
ISC
--significantly--
improves
performance
greedy
randomized
adaptive
search
procedure
(GRASP),
over
benchmark
instances
from
CVRP
literature.
objective
evaluating
ISC's
with
algorithms,
implemented
composed
classical
neighborhoods
literature
another
only
variation
Ruin-and-Recreate.
both
cases,
significantly
quality
found
almost
all
evaluated
instances.
Networks,
Journal Year:
2024,
Volume and Issue:
84(4), P. 465 - 480
Published: July 24, 2024
Abstract
We
present
machine
learning
(ML)
methods
for
automatically
selecting
a
“best”
performing
fast
algorithm
the
capacitated
vehicle
routing
problem
(CVRP)
with
unit
demands.
Algorithm
selection
is
to
choose
among
portfolio
of
algorithms
one
that
predicted
work
best
given
instance,
and
configuration
select
algorithm's
parameters
are
instance.
framework
incorporating
both
includes
configured
“Sweep
Algorithm,”
first
generated
feasible
solution
hybrid
genetic
search
algorithm,
Clarke
Wright
algorithm.
The
selected
shown
here
deliver
high‐quality
solutions
within
very
small
running
times
making
it
highly
suitable
real‐time
applications
generating
initial
global
optimization
CVRP.
These
results
bode
well
effectiveness
utilizing
ML
improving
combinatorial
methods.
Applied Sciences,
Journal Year:
2022,
Volume and Issue:
12(11), P. 5454 - 5454
Published: May 27, 2022
In
a
multi-agent
system,
multi-job
assignment
is
an
optimization
problem
that
seeks
to
minimize
total
cost.
This
can
be
generalized
as
complex
in
which
several
variations
of
vehicle
routing
problems
are
combined,
and
NP-hard
problem.
The
parameters
considered
include
the
number
agents
jobs,
loading
capacity,
speed
agents,
sequence
consecutive
positions
jobs.
this
study,
deep
neural
network
(DNN)
model
was
developed
solve
job
constant
time
regardless
state
parameters.
To
generate
large
training
dataset
for
DNN,
planning
domain
definition
language
(PDDL)
used
describe
problem,
optimal
solution
obtained
using
PDDL
solver
preprocessed
into
sample
dataset.
A
DNN
constructed
by
concatenating
fully-connected
layers.
via
inference
increased
average
traveling
up
13%
compared
with
ground
As
cost,
required
hundreds
seconds,
execution
at
approximately
20
ms
SSRN Electronic Journal,
Journal Year:
2022,
Volume and Issue:
unknown
Published: Jan. 1, 2022
Multi-Start
metaheuristics
(MSM)
are
commonly
used
to
solve
vehicle
routing
problems
(VRPs).
These
methods
create
different
initial
solutions
and
improve
them
through
local-search.
The
goal
of
these
is
deliver
the
best
solution
found.
We
introduce
initial-solution
classification
(ISC)
predict
if
a
local-search
algorithm
should
be
applied
in
MSM.
This
leads
faster
convergence
MSM
higher-quality
when
amount
computation
time
limited.
In
this
work,
we
extract
known
features
capacitated
VRP
(CVRP)
additional
features.
With
machine-learning
classifier
(random
forest),
show
how
ISC
--significantly--
improves
performance
greedy
randomized
adaptive
search
procedure
(GRASP),
over
benchmark
instances
from
CVRP
literature.
objective
evaluating
ISC's
with
algorithms,
implemented
composed
classical
neighborhoods
literature
another
only
variation
Ruin-and-Recreate.
both
cases,
significantly
quality
found
almost
all
evaluated
instances.
