Energies,
Journal Year:
2024,
Volume and Issue:
17(20), P. 5181 - 5181
Published: Oct. 17, 2024
In
order
to
solve
the
problem
of
complex
coupling
characteristics
between
multivariate
load
sequences
and
difficulty
in
accurate
multiple
forecasting
for
integrated
renewable
energy
systems
(IRESs),
which
include
low-carbon
emission
sources,
this
paper,
TCN-FECAM-Informer
model
is
proposed.
First,
maximum
information
coefficient
(MIC)
used
correlate
loads
with
weather
factors
filter
appropriate
features.
Then,
effective
screened
features
extracted
frequency
sequence
constructed
using
frequency-enhanced
channel
attention
mechanism
(FECAM)-improved
temporal
convolutional
network
(TCN).
Finally,
processed
feature
are
sent
Informer
forecasting.
Experiments
conducted
measured
data
from
IRES
Arizona
State
University,
experimental
results
show
that
TCN
FECAM
can
greatly
improve
prediction
accuracy
and,
at
same
time,
demonstrate
superiority
network,
dominated
by
attentional
mechanism,
compared
recurrent
neural
networks
prediction.
IEEE Transactions on Smart Grid,
Journal Year:
2023,
Volume and Issue:
15(2), P. 2044 - 2055
Published: July 5, 2023
Smart
meter
measurements,
though
critical
for
accurate
demand
forecasting,
face
several
drawbacks
including
consumers'
privacy,
data
breach
issues,
to
name
a
few.
Recent
literature
has
explored
Federated
Learning
(FL)
as
promising
privacy-preserving
machine
learning
alternative
which
enables
collaborative
of
model
without
exposing
private
raw
short
term
load
forecasting.
Despite
its
virtue,
standard
FL
is
still
vulnerable
an
intractable
cyber
threat
known
Byzantine
attack
carried
out
by
faulty
and/or
malicious
clients.
Therefore,
improve
the
robustness
federated
short-term
forecasting
against
threats,
we
develop
state-of-the-art
differentially
secured
FL-based
framework
that
ensures
privacy
individual
smart
meter's
while
protect
security
models
and
architecture.
Our
proposed
leverages
idea
gradient
quantization
through
Sign
Stochastic
Gradient
Descent
(SignSGD)
algorithm,
where
clients
only
transmit
'sign'
control
centre
after
local
training.
As
highlight
our
experiments
involving
benchmark
neural
networks
with
set
models,
approach
mitigates
such
threats
quite
effectively
thus
outperforms
conventional
FedSGD
models.
Frontiers in Energy Research,
Journal Year:
2023,
Volume and Issue:
10
Published: Jan. 6, 2023
Aiming
at
the
strong
non-linear
and
non-stationary
characteristics
of
power
load,
a
short-term
load
forecasting
method
based
on
bald
eagle
search
(BES)
optimization
variational
mode
decomposition
(VMD),
convolutional
bi-directional
long
memory
(CNN-Bi-LSTM)
network
considering
error
correction
is
studied
to
improve
accuracy
forecasting.
Firstly,
loss
evaluation
criterion
established,
VMD
optimal
parameters
under
are
determined
BES
quality
signal.
Then,
original
sequence
decomposed
into
different
modal
components,
corresponding
CNN-Bi-LSTM
prediction
models
established
for
each
component.
In
addition,
influence
various
holiday
meteorological
factors
error,
an
model
mine
hidden
information
contained
in
reduce
inherent
model.
Finally,
proposed
applied
public
dataset
provided
by
utility
United
States.
The
results
show
that
this
can
better
track
changes
effectively
International Journal of Machine Learning and Cybernetics,
Journal Year:
2024,
Volume and Issue:
15(12), P. 6061 - 6076
Published: Aug. 6, 2024
Abstract
Accurate
short-term
load
forecasting
(STLF)
is
crucial
for
the
power
system.
Traditional
methods
generally
used
signal
decomposition
techniques
feature
extraction.
However,
these
are
limited
in
extrapolation
performance,
and
parameter
of
modes
needs
to
be
preset.
To
end
this,
this
paper
develops
a
novel
STLF
algorithm
based
on
multi-scale
perspective
decomposition.
The
proposed
adopts
deep
neural
network
(MscaleDNN)
decompose
series
into
low-
high-frequency
components.
Considering
outliers
series,
introduces
adaptive
rescaled
lncosh
(ARlncosh)
loss
fit
distribution
data
improve
robustness.
Furthermore,
attention
mechanism
(ATTN)
extracts
correlations
between
different
moments.
In
two
sets
from
Portugal
Australia,
model
generates
competitive
results.
Frontiers in Earth Science,
Journal Year:
2024,
Volume and Issue:
12
Published: April 24, 2024
Landslides,
prevalent
in
mountainous
areas,
are
typically
triggered
by
tectonic
movements,
climatic
changes,
and
human
activities.
They
pose
catastrophic
risks,
especially
when
occurring
near
settlements
infrastructure.
Therefore,
detecting,
monitoring,
predicting
landslide
deformations
is
essential
for
geo-risk
mitigation.
The
mainstream
of
the
previous
studies
have
often
focused
on
deterministic
models
immediate
prediction.
However,
most
them,
aspect
prediction
uncertainties
not
sufficiently
addressed.
This
paper
introduces
an
innovative
probabilistic
method
using
a
Variational
Autoencoder
(VAE)
combined
with
Gated
Recurrent
Unit
(GRU)
to
forecast
from
generative
standpoint.
Our
approach
consists
two
main
elements:
firstly,
training
VAE-GRU
model
maximize
variational
lower
bound
likelihood
historical
precipitation
data;
secondly,
learned
approximated
posterior
distribution
predict
imminent
angle.
To
assess
quality,
we
use
four
widely-used
metrics:
Prediction
Interval
Coverage
Probability
(PICP),
Normalized
Average
Width
(PINAW),
Width-Based
Criterion
(CWC),
Root
Mean
Square
(PINRW).
results
demonstrate
that
our
proposed
framework
surpasses
traditional
state-of-the-art
(SOTA)
deformation
algorithms
terms
accuracy
reliability.