The
effectiveness
of
Fuzzy
Inference
Systems
(FISs)
in
manipulating
uncertainty
and
nonlinearity
makes
them
a
subject
significant
interest
for
decision-making
embedded
systems.
Accordingly,
optimizing
FIS
hardware
improves
its
performance,
efficiency,
capabilities,
leading
to
better
user
experience,
increased
productivity,
cost
savings.
To
be
compatible
with
the
limited
power
budget
most
systems,
this
paper
presents
framework
realize
ultra-low
hardware.
It
supports
optimizations
both
conventional
arithmetic
as
well
MSDF-computing
highly
consistent
MSDF-based
sensors.
In
all
processes
fuzzification,
inference,
defuzzification
are
done
on
serially
coming
data
bits.
demonstrate
efficiency
proposed
framework,
we
utilized
Matlab,
Chisel3,
Vivado
implement
it
from
high-level
descriptions
synthesis.
We
also
developed
Scala
library
Chisel3
establish
connection
between
these
tools,
bridging
gap,
facilitating
design
space
exploration
at
level.
Furthermore,
realized
an
navigation
autonomous
mobile
robots
unknown
environments.
Synthesis
results
show
superiority
output
our
suggested
terms
resource
usage
energy
consumption
compared
Matlab
HDL
code
generator
output.
Electronics,
Год журнала:
2024,
Номер
13(4), С. 690 - 690
Опубликована: Фев. 8, 2024
Fuzzy
inference
systems
(FISs)
are
a
key
focus
for
decision-making
in
embedded
due
to
their
effectiveness
managing
uncertainty
and
non-linearity.
This
study
demonstrates
that
optimizing
FIS
hardware
enhances
performance,
efficiency,
capabilities,
improving
user
experience,
heightened
productivity,
cost
savings.
We
propose
an
ultra-low
power
framework
address
constraints
systems.
supports
optimizations
conventional
arithmetic
Most
Significant
Digit
First
(MSDF)
computing,
ensuring
compatibility
with
MSDF-based
sensors.
Within
the
MSDF-computing
FIS,
fuzzification,
inference,
defuzzification
processes
occur
on
serially
incoming
data
bits.
To
illustrate
framework’s
we
implemented
it
using
MATLAB,
Chisel3,
Vivado,
starting
from
high-level
descriptions
progressing
synthesis.
A
Scala
library
Chisel3
was
developed
connect
these
tools
seamlessly,
facilitating
design
space
exploration
at
level.
applied
by
realizing
autonomous
mobile
robot
navigation
unknown
environments.
The
synthesis
results
highlight
superiority
of
our
designs
over
MATLAB
HDL
code
generator,
achieving
43%
higher
clock
frequency,
46%
67%
lower
resource
consumption,
respectively.
International Journal of Simulation Modelling,
Год журнала:
2023,
Номер
22(4), С. 712 - 722
Опубликована: Дек. 1, 2023
This
study
introduces
a
hybrid
Autoregressive
Integrated
Moving
Average
Model-Back
Propagation
(ARIMA-BP)
neural
network
model
to
improve
the
accuracy
of
production
material
demand
forecasting
amid
growing
market
competition
and
diverse
customer
requirements.By
integrating
both
linear
nonlinear
elements,
enhances
efficiency
in
planning,
inventory
optimization,
operational
cost
reduction.It
explores
novel
methods
align
supply
demand,
optimizing
interplay
procurement,
product
output,
management.The
study's
key
contribution
is
approach
that
informs
balanced
strategies,
with
significant
implications
for
effectiveness
competitive
advantage
manufacturing.
Journal of Advances in Applied & Computational Mathematics,
Год журнала:
2024,
Номер
11, С. 100 - 118
Опубликована: Окт. 9, 2024
Non-stationary
time
series
prediction
is
challenging
due
to
its
dynamic
and
complex
nature.
Fuzzy
models
offer
a
promising
solution
for
forecasting
such
data,
but
key
challenge
lies
in
partitioning
the
universe
of
discourse,
which
significantly
impacts
accuracy.
Traditional
fuzzy
often
use
equal-length
interval
partitioning,
more
suited
stationary
data
limits
their
adaptability
non-stationary
series.
This
paper
introduces
novel
variable-length
method
designed
specifically
The
developed
combines
Long
Short-Term
Memory
(LSTM)
Autoencoder
with
K-means
clustering,
enabling
dynamic,
data-driven
that
adapts
changing
characteristics
data.
LSTM
encodes
series,
clustered
using
K-means,
intervals
are
defined
based
on
cluster
centers.
Furthermore,
Variable
Length
Interval
Partitioning-based
Time
Series
model
(VLIFTS)
by
incorporating
this
concepts
Markov
chain
transition
probability
matrix.
In
model,
sets
viewed
as
states
chain,
probabilities
used
phase.
validated
stock
market
indices
Nifty
50,
NASDAQ,
S&P
500,
Dow
Jones.
Stationarity
heteroscedasticity
tested
Augmented
Dickey-Fuller
(ADF)
Levene's
tests
respectively.
Statistical
forecast
accuracy
metrics
Root
Mean
Squared
Error
(RMSE)
Absolute
Percent
(MAPE)
show
VLIFTS
improves
over
traditional
models.
hybrid
approach
enhances
modelling
can
be
applied
various
problems.
Research Square (Research Square),
Год журнала:
2023,
Номер
unknown
Опубликована: Апрель 14, 2023
Abstract
Computational
methods
for
time
series
forecasting
have
always
an
edge
over
conventional
of
due
to
their
easy
implementation
and
prominent
characteristics
coping
with
large
amount
data.
Many
computational
fuzzy
(FTS)
been
developed
in
past
using
set,
intuitionistic
set
(IFS),
hesitant
(HFS)
incorporating
uncertainty,
non-determinism,
hesitation
forecasting.
Since
probabilistic
(PFS)
incorporates
both
non-probabilistic
uncertainties
simultaneously,
we
proposed
PFS
particle
swarm
optimization
(PSO)
based
method
FTS
First,
a
then
it
is
integrated
PSO
enhance
the
accuracy
forecasted
outputs.
Unlike
other
method,
used
optimize
number
partitions
length
intervals.
Three
diversified
data
enrolments
University
Alabama,
market
price
State
Bank
India
(SBI)
share
at
Bombay
stock
exchange
(BSE)
India,
death
cases
COVID-19
are
compare
performance
before
after
its
integration
terms
root
mean
square
error
(RMSE).
After
PSO,
outputs
increased
significantly
found
better
than
many
existing
methods.
Goodness
also
tested
tracking
signal
Willmott
index.
Research Square (Research Square),
Год журнала:
2023,
Номер
unknown
Опубликована: Авг. 8, 2023
Abstract
Intuitionistic
fuzzy
time
series
methods
provide
a
good
alternative
to
the
forecasting
problem.
It
is
possible
use
historical
values
of
as
well
membership
and
non-membership
obtained
for
effective
factors
in
improving
performance.
In
this
study,
high
order
single
variable
intuitionistic
reduced
model
first
introduced.
A
new
method
proposed
solution
problem
which
functional
structure
between
information
forecast
by
bagging
decision
trees
based
on
model.
method,
c-means
clustering
used
create
series.
To
simpler
with
Bagging
trees,
input
data
from
lagged
variables,
memberships,
are
subjected
dimension
reduction
principal
component
analysis.
The
performance
compared
popular
literature
ten
different
randomly
S&P500
stock
market.
According
results
analyses,
better
than
both
classical
some
shallow
deep
neural
networks.
