TerraSwarm
Distributed Control of a Swarm of Buildings
Connected to a Smart Grid
http://terraswarm.org/
Baris Aksanli, Alper S. Akyurek, Madhur Behl, Meghan Clark, Alexandre Donze, Prabal Dutta, Patrick Lazik, Mehdi Maasoumy, Rahul Mangharam, Richard Murray, Truong
X. Nghiem, George Pappas, Vasumathi Raman, Anthony Rowe, Alberto Sangiovanni-Vincetelli, Sanjit Seshia, Tajana Rosing, Jagannathan Venkatesh
October 29-30, 2014
Goal: Provide an interface to test large scale distributed sense and control systems with application to smart buildings and grid
MPC-based HVAC Controller
Smart Grid Swarm Simulator (S2Sim)
•  Enables evaluation of the quality of distributed
control of smart buildings
•  Provides a price signal to each client to adjust
the overall power consumption to ensure grid
stability
•  Can emulate
neighborhoods
•  Replicated and
connected to S2Sim
STL-based HVAC Controller
Node: Central Ba:ery -­‐ AC/DC -­‐ Max Capacity … -­‐ Dependent Nodes Node: U1lity Source -­‐ AC/DC -­‐ Power Profile -­‐ Conversion ra'o Scheduler -­‐ Fallback source -­‐ Associated nodes Node: Solar-­‐Electric Source -­‐ AC/DC -­‐ Power Profile -­‐ Conversion ra'o Node: Washer -­‐ AC/DC -­‐ Interval -­‐ Power Profile … Node -­‐ iden'fier Node: Washer Ba:ery -­‐ AC/DC -­‐ Max Capacity … •  Pricing feedback from S2Sim based on consumption, which affects appliance
rescheduling, battery charge/discharge periods, matching solar energy with
demand
UCSD
Real-time Building Actuation
Lutron Quantum Lighting
HVAC Control with MLE+
Controller Inscope Enfuse
Breaker Monitor (10)
Router Wistat Thermostats (60)
40,000 sq ft, 5 story, 140 room,
built in 1962 with classrooms,
auditorium, offices and labs
•  Sensing and control from 6
building automation systems
802.15.4 ->
ModbusRTU
Automatrix PLC (5)
•  Live data streaming into
S2Sim
Electric
Metering
Gateway
HVAC
Gateway
Firefly
Gateway
Voltage Deviation (pu)
Lighting
Gateway
Restful
Interface
Fan Control /
Misc Plug Load (30)
802.15.4
Environmental Sensors
•  Load shedding based on
S2Sim pricing signals
Automatrix Thermostats (30)
Chilled Water and Steam Monitoring
Smart Distributed Coordination
Voltage Deviation (pu)
UCSD
•  Residential energy
simulation platform
Price
updated parameters such as new measurements and disturUPS Statuses
Controlsand
real-world
2Sim
BBB
bance predictions,
sets up theloads
new MPC algorithm
for
S
Energy
of1,S. .2.Sim
Controller
the timeas
stepakfunction
= H c, H c +
, H c + H m − 1, and solves
Demand
c
signal
changes
the new price
MPC for
this time
frame and uses only the first H
•  Flexibility of commercial buildings
values of baseline power consumption and flexibility, i.e. for
HVAC system is a significant
k = H c , . . . , 2H c − 1, and this process repeats.
regulation resource
Fig. 2 shows the schematic of the entire system archiControl
UPS
Bank
•  Defined and quantified flexibility
tecture. The solid line power
flow arrows correspond to the
Signals
(Loads not shown)
baseline system and contract. The ancillary power flow via
of building HVAC systems
the flexibility contract is shown with dashed line arrow.
Real-time state of the system such as occupancy, internal
ACme++
heat, outside weather condition,
building temperature, state
Controllable
Outlets
and input constraints are passed
as input to the algorithm.
•  Designed an MPC framework to guarantee building climate control and grid
Outlet
The utility also communicates information such as per-unit
Gateway
Fig. 2. Schematic of the proposed architecture for contractual framework.
flexibility requirements
energy and upward and downward flexibility prices to the
UMich
•  Implemented contractual framework for costs/benefits to building and grid UCB BEMS (or effectively to the algorithm). The output of the
algorithm includes baseline power consumption, downward
zones, are defined by:
flexibility and upward flexibility values, and cost. Within the
Xk := {x | T k ≤ x ≤ T k }
(19) flexibility contract framework, this information is communicated to the utility. Consequently, a flexibility signal st is
(20)
Uk := {u | U k ≤ u ≤ U k }
sent from the utility to the building to be tracked by the
CALTECH2 CALTECH1
CMU2
CMU1
where T k , and T k are the upper and lower temperature limits HVAC fan. Essentially, the utility has control of the building
•  Resistor-capacitor network to
and U k , and U k are the upper and lower feasible air mass consumption for the next Hc time slots. The control strategy
model heat transfer
flow at time t. Therefore, the feasible set of (18) is closed is actuated by sending flexibility signals (similar to frequency
Current circuit can support up to
AC
UPENN2 UPENN1
UMICH2
UMICH1
and convex. The objective function is also convex on w, as regulation signals). These signals can be sent as frequently
•  MPC to satisfy formal
12MW, corresponding to a
as
every
few
seconds.
In
fact,
we
show
through
experiments
the max is over variable w. In fact, w does not appear in
specifications in Signal Temporal the cost function. The objective function is a linear function on a real building in Section V, that buildings with HVAC
typical small US town with
Logic
on w and hence it is concave. We also consider a linearized systems equipped with variable frequency drive (VFD) fans
approx. 10000 residents
UCB1
UCSD2
UCSD1
UCB2
are capable of tracking such signals
very fast (e.g. within a
state update equation as follows:
few seconds).
•  Maintains a minimum comfortable
xt+1 = Axt + But + Edt
(21)
Time of Use Pricing − Greedy Distributed Control: Unstable
temperature when the room is
0.14
IV. C OMPUTATIONAL R ESULTS
occupied while minimizing the
the nonlinear system dynamics has been linearized with the
UNSTABLE!!!
0.12
The
algorithm
presented
in
Section
III-C
was implemented
forward
Euler
integration
formula
with
time-step
τ
=
1
hr.
cost of heating based on
on a building
model developed and validated against histor0.1
Greedy individual control with
Hence,
the
min-max
problem
(18)
is
equivalent
to:
observed prices
in [4], [6]. We used Yalmip [8], an interface to
Caltech ical data
time-of-use pricing may lead to
0.08
Minimization Problem
optimization solvers available as a MATLAB toolbox, for
instability
0.06
m
rapid prototyping
of optimal control problems. The nonH!
−1
linear optimization
problem solver Ipopt [9], was used to
min
Chvac (ut+k , πt+k ) − R(Φt+k+1 , Bt+k+1 )
0.04
⃗ t+1
u t ,Φ
⃗
k=0
solve the resulting nonlinear optimization problem arising
(22a)
Use price signals for demand
20:00
2:00 We use sampling
8:00
from the0.02
MPC approach.
time of 14:00
τ = 1 hr. 20:00
Time
of
Day
response
, dt+k )
(22b)
s. t.:
xt+k+1 = f(xt+k , ut+k + ϕt+k
E+
We assess the performance of the algorithm for the following
∀k = 0, ..., H m − 1
scenarios. Smart Price Feedback − Greedy Distributed Control: Stable
Matlab toolbox for integrated
Scenario I: Different reward rates have
been considered
xt+k+1 = f(xt+k , ut+k + ϕt+k , dt+k )
(22c)
Smart Pricing guides to stability
modeling and controls for energym
0.1
for upward
and downward flexibility at each time step.
Smart distributed coordination
∀k
=
0,
...,
H
−1
efficient buildings.
In particular, downward flexibility is rewarded more than
ϕt+k ≥ 0, ∀ k = 1, ...H m − 1
(22d)
restores
stability
0.08
upward flexibility (β > β), for most of the time. Since
Co-simulate with realistic
ϕt+k ≤ 0, ∀ k = 1, ...H m − 1
(22e)
•  Pricing feedback
MPC
maintains
the
comfort
level
using
the
least
possible
EnergyPlus buildings
0.06
xt+k ∈ Xt+k ∀ k = 1, ..., H m
(22f)
•  Stability feedback
amount
of
energy,
buildings
that
are
operated
under
nominal
xt+k ∈ Xt+k ∀ k = 1, ..., H m
(22g)
0.04
MPC such
as (1) have no down flexibility (for extended
m
ut+k + ϕt+k ∈ Ut+k ∀ k = 0, ..., H − 1 (22h)
period of time, i.e. 1 hr or more). However we show that via
20:00
2:00solving (22),
8:00
ut+k + ϕt+k ∈ Ut+k ∀ k = 0, ..., H m − 1 (22i)
proper 0.02
incentives
and by
it is possible 14:00
to provide 20:00
Time fo Day
downward flexibility when most needed. The results of this
B. Aksanli, A.S. Akyurek, M. Behl, M. Clark, A. Donze, P. Dutta, Patrick Lazik, M. Maasoumy, R. Mangharam, T.X. Nghiem,
are shown
in A.
Fig.
3 for the case when
ancillary
The argmin of the optimization problem (22) is the nom- scenario
V.Raman,
A. Rowe,
Sangiovanni-Vincentelli,
S. A.
Seshia, T. S. Rosing, J. Venkatesh. Distributed Control of a Swarm of
∗
are received
from the
utility
system Conference on Embedded Systems For Energy-Efficient
inal power consumption, ut+k , and the maximum available signalsBuildings
Connected
to a
Smartevery
Grid,minute:
1st ACMno
International
Buildings
2014
(e.g.(BuildSys),
temperature
being within the comfort zone)
flexibility, Φ∗t+k , ∀ k = 0, ..., H m − 1. The building declares
UPenn constraint
u∗t+k , and Φ∗t+1+k for the time slots: ∀ k = 0, ..., H c − 1 will be violated when the fan speed enforcement is performed
to the utility. After H c time slots, the BEMS collects the for arbitrary values of fan speed as long as fan power
…
•  Currently six clients, from UCSD, UCB,
Caltech, UMich, UPenn, CMU
•  How independent building control tools
affect each other as well as the grid
HomeSim
Beyster Battery Bank Control
CMU
Sponsored by the TerraSwarm Research Center, one of six centers administered by the STARnet phase of the Focus Center
Research Program (FCRP) a Semiconductor Research Corporation program sponsored by MARCO and DARPA.