The
evaluation
of
solar
energy
utilization
potential
urban
building
surfaces
currently
faces
the
dilemma
high
complexity
large-scale-high-precision-multidimensional
coupled
computation.
This
study
introduces
a
more
comprehensive
method
for
clusters
splitting
and
type
identification,
uses
geometric
morphology
to
extract
multi-dimensional
feature
indicators
clusters.
A
sky
module
technology
coupling
temporal
dimension
radiation
type,
dynamic
identification
surface
orientation,
high-performance
computational
framework
metrics
parsing
have
been
developed.
Further,
variety
machine
learning
algorithms
were
examined,
finally
XGB
model,
which
balances
predictive
performance
(R2>0.95
MSE<0.10)
prevents
overfitting,
was
selected
predict
multidimensional
existing
buildings
in
non-enriched
areas.
found
that:
(a)
geographic
location
clusters,
types
can
better
characterize
variability
be
used
build
high-precision
prediction
models.
(b)
shading
typical
varies
across
orientations,
with
roofs
distributed
between
3.45%
6.98%,
façades
34.70
50.71%.
(c)The
is
significant
both
different
directions
time
dimensions,
e.g.,
winter
accounts
about
38%
summer
Chengdu
only
30%
Chongqing.
In
this
study,
we
further
captured
nonlinear
relationship
thresholds
effective
potentials
under
orientations
constructed
models
bi-directional
gains
explaining
science
advancing
applications.
Buildings,
Journal Year:
2024,
Volume and Issue:
14(9), P. 2978 - 2978
Published: Sept. 20, 2024
Although
it
is
well
established
that
thermal
environments
significantly
influence
travel
behavior,
the
synergistic
effects
of
points
interest
(POI)
and
on
behavior
remain
unclear.
This
study
developed
a
vision-based
outdoor
evaluation
model
aimed
at
uncovering
driving
factors
behind
human
in
spaces.
First,
Yolo
v5
questionnaires
were
employed
to
obtain
crowd
activity
intensity
preference
levels.
Subsequently,
target
detection
clustering
algorithms
used
derive
variables
such
as
POI
attractiveness
distance,
while
validated
environmental
simulator
was
utilized
simulate
comfort
distributions
across
different
times.
Finally,
multiple
classification
models
compared
establish
mapping
relationships
between
POI,
environment
variables,
preferences,
with
SHAP
analysis
examine
contribution
each
variable.
The
results
indicate
XGBoost
achieved
best
predictive
performance
(accuracy
=
0.95),
shadow
proportion
(|SHAP|
0.24)
distance
0.12)
identified
most
significant
influencing
preferences.
By
extrapolation,
this
can
provide
valuable
insights
for
optimizing
community
enhancing
vitality
areas
similar
climatic
cultural
contexts.
The
evaluation
of
solar
energy
utilization
potential
urban
building
surfaces
currently
faces
the
dilemma
high
complexity
large-scale-high-precision-multidimensional
coupled
computation.
This
study
introduces
a
more
comprehensive
method
for
clusters
splitting
and
type
identification,
uses
geometric
morphology
to
extract
multi-dimensional
feature
indicators
clusters.
A
sky
module
technology
coupling
temporal
dimension
radiation
type,
dynamic
identification
surface
orientation,
high-performance
computational
framework
metrics
parsing
have
been
developed.
Further,
variety
machine
learning
algorithms
were
examined,
finally
XGB
model,
which
balances
predictive
performance
(R2>0.95
MSE<0.10)
prevents
overfitting,
was
selected
predict
multidimensional
existing
buildings
in
non-enriched
areas.
found
that:
(a)
geographic
location
clusters,
types
can
better
characterize
variability
be
used
build
high-precision
prediction
models.
(b)
shading
typical
varies
across
orientations,
with
roofs
distributed
between
3.45%
6.98%,
façades
34.70
50.71%.
(c)The
is
significant
both
different
directions
time
dimensions,
e.g.,
winter
accounts
about
38%
summer
Chengdu
only
30%
Chongqing.
In
this
study,
we
further
captured
nonlinear
relationship
thresholds
effective
potentials
under
orientations
constructed
models
bi-directional
gains
explaining
science
advancing
applications.
