Energies,
Год журнала:
2024,
Номер
18(1), С. 2 - 2
Опубликована: Дек. 24, 2024
The
increasing
share
of
renewable
energies
within
energy
systems
leads
to
an
increase
in
complexity.
growing
complexity
is
due
the
diversity
technologies,
ongoing
technological
innovations,
and
fluctuating
electricity
production.
To
continue
ensure
a
secure,
economical,
needs-based
supply,
additional
information
needed
efficiently
control
these
systems.
This
impacts
public
industrial
supply
systems,
such
as
vehicle
factories.
paper
examines
influencing
factors
applicability
Temporal
Fusion
Transformer
(TFT)
model
for
weekly
demand
forecast
at
automobile
production
site.
Seven
different
TFT
models
were
trained
demand.
Six
predicted
electricity,
heat,
natural
gas.
Three
used
rolling
day-ahead
forecast,
three
entire
week
one
step.
In
seventh
model,
was
again,
with
target
values
being
same
model.
analysis
shows
that
forecasting
method
MAPE
13%
already
delivers
good
results
predicting
electrical
prediction
accuracy
achieved
sufficient
use
outcomes
basis
operational
planning
reporting.
However,
further
improvements
are
still
required
automated
system
reduce
procurement
costs.
heat
gas
demands
show
too
high
deviations,
62%
39%
accurately
predict
demands,
must
be
identified
explain
Increasingly,
homeowners
opt
for
photovoltaic
(PV)
systems
and/or
battery
storage
to
minimize
their
energy
bills
and
maximize
renewable
usage.
This
has
spurred
the
development
of
advanced
control
algorithms
that
maximally
achieve
those
goals.
However,
a
common
challenge
faced
while
developing
such
controllers
is
unavailability
accurate
forecasts
household
power
consumption,
especially
shorter
time
resolutions
(15
minutes)
in
data-efficient
manner.
In
this
paper,
we
analyze
how
transfer
learning
can
help
by
exploiting
data
from
multiple
households
improve
single
house's
load
forecasting.
Specifically,
train
an
forecasting
model
(a
temporal
fusion
transformer)
using
different
households,
then
finetune
global
on
new
with
limited
(i.e.,
only
few
days).
The
obtained
models
are
used
consumption
next
24
hours
(day-ahead)
at
resolution
15
minutes,
intention
these
as
Model
Predictive
Control.
We
show
benefit
setup
versus
solely
individual
household's
data,
both
terms
real-world
data.
Forecasting
electricity
demand
plays
a
critical
role
in
ensuring
reliable
and
cost-efficient
operation
of
the
supply.
With
global
transition
to
distributed
renewable
energy
sources
electrification
heating
transportation,
accurate
load
forecasts
become
even
more
important.
While
numerous
empirical
studies
handful
review
articles
exist,
there
is
surprisingly
little
quantitative
analysis
literature,
most
notably
none
that
identifies
impact
factors
on
forecasting
performance
across
entirety
studies.
In
this
article,
we
therefore
present
Meta-Regression
Analysis
(MRA)
examines
influence
accuracy
short-term
forecasts.
We
use
data
from
421
forecast
models
published
59
grid
level
(esp.
individual
vs.
aggregated
system),
granularity,
algorithms
used
seem
have
significant
MAPE,
bibliometric
data,
dataset
sizes,
prediction
horizon
show
no
effect.
found
LSTM
approach
combination
neural
networks
with
other
approaches
be
best
methods.
The
results
help
practitioners
researchers
make
meaningful
model
choices.
Yet,
paper
calls
for
further
MRA
field
close
blind
spots
research
practice
forecasting.
Journal of Renewable and Sustainable Energy,
Год журнала:
2023,
Номер
15(5)
Опубликована: Сен. 1, 2023
Accurate
prediction
of
solar
irradiance
is
essential
for
the
successful
integration
power
plants
into
electrical
systems.
Despite
recent
advancements
in
deep
learning
technology
yielding
impressive
results
forecasting,
their
lack
interpretability
has
hindered
widespread
adoption.
In
this
paper,
we
propose
a
novel
approach
that
integrates
Temporal
Fusion
Transformer
(TFT)
with
McClear
model
to
achieve
accurate
and
interpretable
forecasting
performance.
The
TFT
provides
transparency
its
predictions
through
use
self-attention
layers
long-term
dependencies,
recurrent
local
processing,
specialized
components
feature
selection,
gating
suppress
extraneous
components.
capable
temporal
associations
between
continuous
time-series
variables,
namely,
historical
global
horizontal
(GHI)
clear
sky
GHI,
accounting
cloud
cover
variability
conditions
are
often
ignored
by
most
machine
forecasters.
Additionally,
it
minimizes
quantile
loss
during
training
produce
probabilistic
forecasts.
study,
evaluate
performance
hourly
GHI
forecasts
on
eight
diverse
datasets
varying
climates:
temperate,
cold,
arid,
equatorial,
multiple
horizons
2,
3,
6,
12,
24
h.
benchmarked
against
both
climatological
persistence
deterministic
Complete
History
Persistence
Ensemble
forecasting.
To
prove
our
not
location
locked,
been
blind
tested
four
completely
different
datasets.
demonstrate
proposed
outperforms
counterparts
across
all
forecast
horizons.
International Journal of Energy Research,
Год журнала:
2024,
Номер
2024(1)
Опубликована: Янв. 1, 2024
Accurately
forecasting
electricity
demand
is
crucial
for
maintaining
the
balance
between
supply
and
of
electric
energy
in
real‐time,
ensuring
reliability
cost‐efficiency
power
system
operations.
The
integration
numerous
active
loads
distributed
renewable
sources
into
grid
has
led
to
increased
load
variability,
rendering
traditional
point
approach
inadequate
meeting
evolving
needs
system.
Probabilistic
forecasting,
which
predicts
complete
probability
distribution
provides
more
extensive
information
on
uncertainty,
emerged
as
a
key
solution
address
these
challenges.
long
short‐term
memory
(LSTM)
model,
known
its
strong
performance
modeling
series,
commonly
utilized
forecasting.
Therefore,
this
study
focuses
users
specific
park
Yantai.
We
propose
model
based
integrated
feature
selection
(IFS),
genetic
algorithm
(GA)
optimization
LSTM,
quantile
regression
(QR),
referred
IFS‐GA‐QRLSTM
model.
Initially,
method
employed
identify
most
influential
factors
affecting
load,
optimizing
model’s
input
features
reducing
data
redundancy.
To
subjective
nature
parameter
LSTM
we
use
GA
optimize
parameters.
combination
optimized
with
QR
enables
direct
generation
predictions,
are
further
used
kernel
density
estimation
construct
distribution.
compare
proposed
five
basic
models,
QRLSTM,
IFS‐QRCNN,
IFS‐QRRNN,
IFS‐QRLSTM,
IFS‐QRGRU,
prediction,
interval
prediction.
Experimental
results
demonstrate
that
paper
exhibits
better
prediction
performance,
smaller
errors,
greater
effectiveness
compared
aforementioned
models.
Energies,
Год журнала:
2024,
Номер
18(1), С. 2 - 2
Опубликована: Дек. 24, 2024
The
increasing
share
of
renewable
energies
within
energy
systems
leads
to
an
increase
in
complexity.
growing
complexity
is
due
the
diversity
technologies,
ongoing
technological
innovations,
and
fluctuating
electricity
production.
To
continue
ensure
a
secure,
economical,
needs-based
supply,
additional
information
needed
efficiently
control
these
systems.
This
impacts
public
industrial
supply
systems,
such
as
vehicle
factories.
paper
examines
influencing
factors
applicability
Temporal
Fusion
Transformer
(TFT)
model
for
weekly
demand
forecast
at
automobile
production
site.
Seven
different
TFT
models
were
trained
demand.
Six
predicted
electricity,
heat,
natural
gas.
Three
used
rolling
day-ahead
forecast,
three
entire
week
one
step.
In
seventh
model,
was
again,
with
target
values
being
same
model.
analysis
shows
that
forecasting
method
MAPE
13%
already
delivers
good
results
predicting
electrical
prediction
accuracy
achieved
sufficient
use
outcomes
basis
operational
planning
reporting.
However,
further
improvements
are
still
required
automated
system
reduce
procurement
costs.
heat
gas
demands
show
too
high
deviations,
62%
39%
accurately
predict
demands,
must
be
identified
explain
