Applied Energy,
Journal Year:
2024,
Volume and Issue:
359, P. 122668 - 122668
Published: Jan. 22, 2024
The
use
of
machine
learning
in
building
technology
has
become
increasingly
important
recent
years.
One
the
applications
is
heating
load
prediction,
which
enables
demand-side
flexibility.
Most
studies
consider
prediction
without
sufficient
context
with
existing
characteristics.
For
an
accurate
suitable
features
have
to
be
selected
according
their
importance,
feature
importance
(FI).
scope
this
paper
investigate
whether
there
a
relationship
between
characteristics
and
FI
if
so,
how
strong
is.
Additionally,
analysis
been
conducted
determine
characteristic
most
significant
impact
on
FI.
purpose,
full
factorial
design
room
six
different
carried
out.
In
total,
calculated
for
15
552
variants.
thermal
balance,
correlation,
random
forest
FI,
permutation
SHapley
Additive
exPlanations
(SHAP)
values
are
these
rooms.
local
SHAP
were
used
explain
model.
These
also
provide
insight
into
interaction
individual
load.
variants,
outdoor
temperature
had
highest
It
investigated
greatest
influence
values.
A
was
found
proportion
correlation
label
as
well
association
balance
study
shows
systematic
Therefore,
should
always
considered
Advances in Applied Energy,
Journal Year:
2023,
Volume and Issue:
9, P. 100123 - 100123
Published: Jan. 13, 2023
Machine
learning
has
been
widely
adopted
for
improving
building
energy
efficiency
and
flexibility
in
the
past
decade
owing
to
ever-increasing
availability
of
massive
operational
data.
However,
it
is
challenging
end-users
understand
trust
machine
models
because
their
black-box
nature.
To
this
end,
interpretability
attracted
increasing
attention
recent
studies
helps
users
decisions
made
by
these
models.
This
article
reviews
previous
that
interpretable
techniques
management
analyze
how
model
improved.
First,
are
categorized
according
application
stages
techniques:
ante-hoc
post-hoc
approaches.
Then,
analyzed
detail
specific
with
critical
comparisons.
Through
review,
we
find
broad
faces
following
significant
challenges:
(1)
different
terminologies
used
describe
which
could
cause
confusion,
(2)
performance
ML
tasks
difficult
compare,
(3)
current
prevalent
such
as
SHAP
LIME
can
only
provide
limited
interpretability.
Finally,
discuss
future
R&D
needs
be
accelerate
management.
Advances in Applied Energy,
Journal Year:
2023,
Volume and Issue:
10, P. 100135 - 100135
Published: April 6, 2023
As
one
of
the
most
important
and
advanced
technology
for
carbon-mitigation
in
building
sector,
performance
simulation
(BPS)
has
played
an
increasingly
role
with
powerful
support
energy
modelling
(BEM)
energy-efficient
designs,
operations,
retrofitting
buildings.
Owing
to
its
deep
integration
multi-disciplinary
approaches,
researchers,
as
well
tool
developers
practitioners,
are
facing
opportunities
challenges
during
application
BEM
at
multiple
scales
stages,
e.g.,
building/system/community
levels
planning/design/operation
stages.
By
reviewing
recent
studies,
this
paper
aims
provide
a
clear
picture
how
performs
solving
different
research
questions
on
varied
phase
spatial
resolution,
focus
objectives
frameworks,
methods
tools,
applicability
transferability.
To
guide
future
applications
performance-driven
management,
we
classified
current
trends
into
five
topics
that
span
through
stages
levels:
(1)
Simulation
design
new
retrofit
design,
(2)
Model-based
operational
optimization,
(3)
Integrated
using
data
measurements
digital
twin,
(4)
Building
supporting
urban
planning,
(5)
Modelling
building-to-grid
interaction
demand
response.
Additionally,
recommendations
discussed,
covering
potential
occupancy
behaviour
modelling,
machine
learning,
quantification
model
uncertainties,
linking
monitoring
systems.
Energy and Buildings,
Journal Year:
2023,
Volume and Issue:
303, P. 113768 - 113768
Published: Nov. 22, 2023
Stakeholders
such
as
urban
planners
and
energy
policymakers
use
building
performance
modeling
analysis
to
develop
strategic
sustainable
plans
with
the
aim
of
reducing
consumption
emissions
from
built
environment.
However,
inconsistent
data
lack
scalable
models
create
a
gap
between
traditional
planning
practices.
An
alternative
approach
is
conduct
large-scale
usage
survey,
which
time-consuming.
Similarly,
existing
studies
rely
on
machine
learning
or
statistical
approaches
for
calculating
performance.
This
paper
proposes
solution
that
employs
data-driven
predict
residential
buildings,
using
both
ensemble-based
end-use
demand
segregation
methods.
The
proposed
methodology
consists
five
steps:
collection,
archetype
development,
physics-based
parametric
modeling,
analysis.
devised
tested
Irish
stock
generates
synthetic
dataset
one
million
buildings
through
19
identified
vital
variables
four
archetypes.
As
part
process,
study
implemented
an
method,
including
heating,
lighting,
equipment,
photovoltaic,
hot
water,
at
scale.
Furthermore,
model's
enhanced
by
employing
approach,
achieving
91%
accuracy
compared
approach's
76%.
Accurate
prediction
enables
stakeholders,
planners,
make
informed
decisions
when
retrofit
measures.
