Water,
Год журнала:
2021,
Номер
13(5), С. 644 - 644
Опубликована: Фев. 28, 2021
Every
morning,
water
suppliers
need
to
define
their
pump
schedules
for
the
next
24
h
drinking
production.
Plans
must
be
designed
in
such
a
way
that
is
always
available
and
amount
of
unused
pumped
into
network
reduced.
Therefore,
operators
accurately
estimate
day’s
consumption
profile.
In
real-life
applications
with
standard
profiles,
some
expert
system
or
vector
autoregressive
models
are
used.
Still,
recent
years,
significant
improvements
time
series
prediction
have
been
achieved
through
special
deep
learning
algorithms
called
long
short-term
memory
(LSTM)
networks.
This
paper
investigates
applicability
LSTM
demand
optimal
control
compares
LSTMs
against
other
methods
currently
used
by
suppliers.
It
shown
outperform
since
they
can
easily
integrate
additional
information
like
day
week
national
holidays.
Furthermore,
online-
transfer-learning
capabilities
investigated.
only
couple
days
training
data
achieve
reasonable
results.
As
focus
on
real-world
application
LSTMs,
from
two
different
distribution
plants
benchmarking.
Finally,
it
significantly
operation.
Journal of Hydroinformatics,
Год журнала:
2018,
Номер
20(6), С. 1343 - 1366
Опубликована: Авг. 21, 2018
Abstract
Nowadays,
a
large
number
of
water
utilities
still
manage
their
operation
on
the
instant
demand
network,
meaning
that
use
equipment
is
conditioned
by
immediate
necessity.
The
reservoirs
networks
are
filled
using
pumps
start
working
when
level
reaches
specified
minimum,
stopping
it
maximum
level.
Shifting
focus
to
management
based
future
allows
energy
cheaper,
taking
advantage
electricity
tariff
in
action,
thus
bringing
significant
financial
savings
over
time.
Short-term
forecasting
crucial
step
support
decision
making
regarding
management.
For
this
purpose,
methodologies
analyzed
and
implemented.
Several
machine
learning
methods,
such
as
neural
networks,
random
forests,
vector
machines
k-nearest
neighbors,
evaluated
real
data
from
two
Portuguese
utilities.
Moreover,
influence
factors
weather,
seasonality,
amount
used
training
forecast
window
also
analysed.
A
weighted
parallel
strategy
gathers
advantages
different
techniques
suggested.
results
validated
compared
with
those
achieved
autoregressive
integrated
moving
average
(ARIMA)
benchmarks.
IEEE Transactions on Control of Network Systems,
Год журнала:
2019,
Номер
7(2), С. 711 - 723
Опубликована: Сен. 5, 2019
With
dynamic
electricity
pricing,
the
operation
of
water
distribution
systems
(WDS)
is
expected
to
become
more
variable.
The
pumps
moving
from
reservoirs
tanks
and
consumers
can
serve
as
energy
storage
alternatives
if
properly
operated.
Nevertheless,
optimal
WDS
scheduling
challenged
by
hydraulic
law,
according
which
pressure
along
a
pipe
drops
proportionally
its
squared
flow
(WF).
(OWF)
task
formulated
here
mixed-integer
nonconvex
problem
incorporating
constraints,
critical
for
fixed-speed
pumps,
tanks,
reservoirs,
pipes.
constraints
OWF
are
subsequently
relaxed
second-order
cone
constraints.
To
restore
feasibility
original
penalty
term
appended
objective
OWF.
modified
be
solved
program,
analytically
shown
yield
WDS-feasible
minimizers
under
certain
sufficient
conditions.
Under
these
conditions,
suitably
weighting
term,
attain
arbitrarily
small
optimality
gaps,
thus
providing
solutions.
Numerical
tests
using
real-world
demands
prices
on
benchmark
demonstrate
relaxation
exact
even
setups
where
conditions
not
met.
Water,
Год журнала:
2021,
Номер
13(5), С. 644 - 644
Опубликована: Фев. 28, 2021
Every
morning,
water
suppliers
need
to
define
their
pump
schedules
for
the
next
24
h
drinking
production.
Plans
must
be
designed
in
such
a
way
that
is
always
available
and
amount
of
unused
pumped
into
network
reduced.
Therefore,
operators
accurately
estimate
day’s
consumption
profile.
In
real-life
applications
with
standard
profiles,
some
expert
system
or
vector
autoregressive
models
are
used.
Still,
recent
years,
significant
improvements
time
series
prediction
have
been
achieved
through
special
deep
learning
algorithms
called
long
short-term
memory
(LSTM)
networks.
This
paper
investigates
applicability
LSTM
demand
optimal
control
compares
LSTMs
against
other
methods
currently
used
by
suppliers.
It
shown
outperform
since
they
can
easily
integrate
additional
information
like
day
week
national
holidays.
Furthermore,
online-
transfer-learning
capabilities
investigated.
only
couple
days
training
data
achieve
reasonable
results.
As
focus
on
real-world
application
LSTMs,
from
two
different
distribution
plants
benchmarking.
Finally,
it
significantly
operation.
