IEEE Internet of Things Journal,
Год журнала:
2023,
Номер
10(15), С. 13876 - 13894
Опубликована: Март 30, 2023
With
increasing
electricity
prices,
cost
savings
through
load
shifting
are
becoming
increasingly
important
for
energy
end
users.
While
dynamic
pricing
encourages
customers
to
shift
demand
low
price
periods,
the
nonstationary
and
highly
volatile
nature
of
prices
poses
a
significant
challenge
management
systems.
In
this
article,
we
investigate
flexibility
potential
data
centers
by
optimizing
heating,
ventilation,
air
conditioning
systems
with
general
model-free
reinforcement
learning
(RL)
approach.
Since
soft
actor-critic
algorithm
feedforward
networks
did
not
work
satisfactorily
in
scenario,
propose
instead
parameterization
recurrent
neural
network
architecture
successfully
handle
spot-market
data.
The
past
is
encoded
into
hidden
state,
which
provides
way
learn
temporal
dependencies
observations
rewards.
proposed
method
then
evaluated
experiments
on
simulated
center.
Considering
real
temperature
signals
over
multiple
years,
results
show
reduction
compared
proportional,
integral
derivative
controller
while
maintaining
center
within
desired
operating
ranges.
context,
demonstrates
an
innovative
applicable
RL
approach
that
incorporates
complex
economic
objectives
agent
decision-making.
control
can
be
integrated
various
Internet
Things-based
smart
building
solutions
management.
Energy Reports,
Год журнала:
2024,
Номер
11, С. 2255 - 2265
Опубликована: Фев. 6, 2024
District
energy
management
offers
possibilities
for
optimal
usage
on
a
local
scale.
However,
it
presents
challenges
due
to
multiple
stakeholders,
heterogeneous
assets,
and
varying
needs.
In
this
article,
an
IT
reference
architecture
cross-sectoral
district
system
is
presented.
This
addresses
the
aforementioned
aims
optimize
within
district.
To
define
architecture,
existing
roles
in
are
analyzed
mapped
onto
structure
of
17
key
identified,
with
manager
being
introduced
as
new
central
role.
A
requirements
analysis
identifies
main
tasks
efficient
management,
including
forecasting,
optimization
flexibilities.
During
design
four
primary
software
modules
data
preprocessing,
balancing
defined.
These
accompanied
by
five
secondary,
optional
modules.
The
modularity
prioritized,
enabling
customization
suit
specific
needs
each
comprehensively
covers
both
technical
organizational
aspects,
taking
into
consideration
relevant
roles.
By
acting
unit
district,
facilitates
holistic
management.
Journal of Renewable and Sustainable Energy,
Год журнала:
2025,
Номер
17(1)
Опубликована: Янв. 1, 2025
The
diverse
load
profile
formation
and
utility
preferences
of
multitype
electricity
users
challenge
real-time
pricing
(RTP)
welfare
equilibrium.
This
paper
designs
an
RTP
strategy
for
smart
grids.
On
the
demand
side,
it
constructs
functions
reflecting
user
characteristics
uses
multi-agents
different
interests.
Considering
industrial
users,
small-scale
microgrids,
distributed
generation,
battery
energy
storage
systems
are
included.
Based
on
supply
interest,
a
online
multi-agent
reinforcement
learning
(RL)
algorithm
is
proposed.
A
bi-level
stochastic
model
in
Markov
decision
process
framework
optimizes
strategy.
Through
information
exchange,
adaptive
scheme
balances
interest
achieves
optimal
strategies.
Simulation
results
confirm
effectiveness
proposed
method
peak
shaving
valley
filling.
Three
fluctuation
scenarios
compared,
showing
algorithm's
adaptability.
findings
reveal
potential
RL-based
resource
allocation
benefits
Innovations
modeling,
construction,
application
have
theoretical
practical
significance
market
research.
