This action might not be possible to undo. Are you sure you want to continue?
Florian D¨orﬂer and Francesco Bullo
Center for Control,
Dynamical Systems & Computation
University of California at Santa Barbara
http://motion.me.ucsb.edu
Institute for Energy Eﬃciency
UC Santa Barbara
Santa Barbara, California, October 19, 2011
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 1 / 23
Motivation: the current power grid is . . .
“. . . the greatest engineering achievement of the 20th century.”
[National Academy of Engineering ’10]
1
largescale, complex, & rich nonlinear dynamics
2
100 years old and operating at its capacity limits
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 2 / 23
Motivation: the current power grid is . . .
“. . . the greatest engineering achievement of the 20th century.”
[National Academy of Engineering ’10]
1
largescale, complex, & rich nonlinear dynamics
2
100 years old and operating at its capacity limits ⇒ BLACKOUTS
The Blackout of 2003: 8/15/2003
Failure Reveals Creaky System, Experts Believe
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 2 / 23
Motivation: the envisioned power grid
Energy is one of the top three national priorities
Expected developments in “smart grid”:
1
large number of distributed power sources
2
increasing adoption of renewables
3
sophisticated cybercoordination layer
challenges: increasingly complex networks & stochastic disturbances
opportunity: some smart grid keywords:
control/sensing/optimization ⊕ distributed/coordinated/decentralized
Central theme: “understanding and taming complexity”
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 3 / 23
Motivation: the envisioned power grid
Energy is one of the top three national priorities
Expected developments in “smart grid”:
1
large number of distributed power sources
2
increasing adoption of renewables
3
sophisticated cybercoordination layer
challenges: increasingly complex networks & stochastic disturbances
opportunity: some smart grid keywords:
control/sensing/optimization ⊕ distributed/coordinated/decentralized
Central theme: “understanding and taming complexity”
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 3 / 23
Motivation: the envisioned power grid – our viewpoint
Projects at UCSB: “power systems engineering” ⊕ “networked control”
1
detection and identiﬁcation of faults & cyberphysical attacks
(together with F. Pasqualetti)
t
1
t
2
t
3
ω
1
ω
2
ω
3
g
1
g
2
g
3
b
4
b
1
b
5
b
2
b
6
b
3
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Sensors
WECC 3/9 power system system dynamics & measurement
Objectives:
Is the attack or fault detectable/identiﬁable by measurements?
How to design (distributed) ﬁlters for detection/identiﬁcation?
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 4 / 23
Motivation: the envisioned power grid – our viewpoint
Projects at UCSB: “power systems engineering” ⊕ “networked control”
1
detection and identiﬁcation of faults & cyberphysical attacks
(together with F. Pasqualetti)
t
1
t
2
t
3
ω
1
ω
2
ω
3
g
1
g
2
g
3
b
4
b
1
b
5
b
2
b
6
b
3
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Sensors
WECC 3/9 power system system dynamics & measurement
Objectives:
Is the attack or fault detectable/identiﬁable by measurements?
How to design (distributed) ﬁlters for detection/identiﬁcation?
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 4 / 23
Motivation: the envisioned power grid – our viewpoint
Projects at UCSB: “power systems engineering” ⊕ “networked control”
2
Synchronization & transient stability
Generators have to swing synchronously
despite severe ﬂuctuations in generation/load
or faults in network/system components
Objectives [D. Hill & G. Chen ’06]: power network dynamics
?
graph
Observations from distinct ﬁelds:
◦ power networks are coupled oscillators
◦ coupled oscillators sync for large coupling
◦ graph theory quantiﬁes coupling, e.g., λ
2
⇒ plausible(?): power networks sync for large λ
2
x
x
x
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 5 / 23
Motivation: the envisioned power grid – our viewpoint
Projects at UCSB: “power systems engineering” ⊕ “networked control”
2
Synchronization & transient stability
Generators have to swing synchronously
despite severe ﬂuctuations in generation/load
or faults in network/system components
Objectives [D. Hill & G. Chen ’06]: power network dynamics
?
graph
Observations from distinct ﬁelds:
◦ power networks are coupled oscillators
◦ coupled oscillators sync for large coupling
◦ graph theory quantiﬁes coupling, e.g., λ
2
⇒ plausible(?): power networks sync for large λ
2
x
x
x
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 5 / 23
Motivation: the envisioned power grid – our viewpoint
Projects at UCSB: “power systems engineering” ⊕ “networked control”
3
Kron reduction – model reduction using algebraic graph theory
2
10
30
25
8
37
29
9
38
23
7
3
6
22
6
3
5
19
4
33
20
5
34
10
3
32
6
2
3
1
1
8
7
5
4
3
18
17
26
27
28
24
21
16
15 14
13
12
11
1
3
9
9
2
30 25
37
29
38
23
3
6
22
3
5
19
33
20
34
10
32
6 3
1
1
8
7
5
4
3
18
17
26
27
28
24
21
16
15 14
13
12
11
3
9
9
10
9
7
6
4
5
3
2
1
8
15
5
12
11
10
7
8
9
4
3
1
2
17
18
14
16
19
20
21
24
26
27
28
31
32
34 33
36
38
39
22
35
6
13
30
37
25
29
23
1
10
8
2
3
6
9
4
7
5
F
Fig. 9. The New England test system [10], [11]. The system includes
10 synchronous generators and 39 buses. Most of the buses have constant
active and reactive power loads. Coupled swing dynamics of 10 generators
are studied in the case that a linetoground fault occurs at point F near bus
16.
test system can be represented by
˙
δ
i
= ω
i
,
H
i
πf
s
˙ ω
i
= −D
i
ω
i
+ P
mi
−G
ii
E
2
i
−
10
j=1,j=i
E
i
E
j
·
· {G
ij
cos(δ
i
−δ
j
) + B
ij
sin(δ
i
−δ
j
)},
(11)
where i = 2, . . . , 10. δ
i
is the rotor angle of generator i with
respect to bus 1, and ω
i
the rotor speed deviation of generator
i relative to system angular frequency (2πf
s
= 2π ×60 Hz).
δ
1
is constant for the above assumption. The parameters
f
s
, H
i
, P
mi
, D
i
, E
i
, G
ii
, G
ij
, and B
ij
are in per unit
system except for H
i
and D
i
in second, and for f
s
in Helz.
The mechanical input power P
mi
to generator i and the
magnitude E
i
of internal voltage in generator i are assumed
to be constant for transient stability studies [1], [2]. H
i
is
the inertia constant of generator i, D
i
its damping coefﬁcient,
and they are constant. G
ii
is the internal conductance, and
G
ij
+ jB
ij
the transfer impedance between generators i
and j; They are the parameters which change with network
topology changes. Note that electrical loads in the test system
are modeled as passive impedance [11].
B. Numerical Experiment
Coupled swing dynamics of 10 generators in the
test system are simulated. E
i
and the initial condition
(δ
i
(0), ω
i
(0) = 0) for generator i are ﬁxed through power
ﬂow calculation. H
i
is ﬁxed at the original values in [11].
P
mi
and constant power loads are assumed to be 50% at their
ratings [22]. The damping D
i
is 0.005 s for all generators.
G
ii
, G
ij
, and B
ij
are also based on the original line data
in [11] and the power ﬂow calculation. It is assumed that
the test system is in a steady operating condition at t = 0 s,
that a linetoground fault occurs at point F near bus 16 at
t = 1 s−20/(60 Hz), and that line 16–17 trips at t = 1 s. The
fault duration is 20 cycles of a 60Hz sine wave. The fault
is simulated by adding a small impedance (10
−7
j) between
bus 16 and ground. Fig. 10 shows coupled swings of rotor
angle δ
i
in the test system. The ﬁgure indicates that all rotor
angles start to grow coherently at about 8 s. The coherent
growing is global instability.
C. Remarks
It was conﬁrmed that the system (11) in the New Eng
land test system shows global instability. A few comments
0 2 4 6 8 10
5
0
5
10
15
δ
i
/
r
a
d
10
02
03
04
05
0 2 4 6 8 10
5
0
5
10
15
δ
i
/
r
a
d
TIME / s
06
07
08
09
Fig. 10. Coupled swing of phase angle δi in New England test system.
The fault duration is 20 cycles of a 60Hz sine wave. The result is obtained
by numerical integration of eqs. (11).
are provided to discuss whether the instability in Fig. 10
occurs in the corresponding real power system. First, the
classical model with constant voltage behind impedance is
used for ﬁrst swing criterion of transient stability [1]. This is
because second and multi swings may be affected by voltage
ﬂuctuations, damping effects, controllers such as AVR, PSS,
and governor. Second, the fault durations, which we ﬁxed at
20 cycles, are normally less than 10 cycles. Last, the load
condition used above is different from the original one in
[11]. We cannot hence argue that global instability occurs in
the real system. Analysis, however, does show a possibility
of global instability in real power systems.
IV. TOWARDS A CONTROL FOR GLOBAL SWING
INSTABILITY
Global instability is related to the undesirable phenomenon
that should be avoided by control. We introduce a key
mechanism for the control problem and discuss control
strategies for preventing or avoiding the instability.
A. Internal Resonance as Another Mechanism
Inspired by [12], we here describe the global instability
with dynamical systems theory close to internal resonance
[23], [24]. Consider collective dynamics in the system (5).
For the system (5) with small parameters p
m
and b, the set
{(δ, ω) ∈ S
1
× R  ω = 0} of states in the phase plane is
called resonant surface [23], and its neighborhood resonant
band. The phase plane is decomposed into the two parts:
resonant band and highenergy zone outside of it. Here the
initial conditions of local and mode disturbances in Sec. II
indeed exist inside the resonant band. The collective motion
before the onset of coherent growing is trapped near the
resonant band. On the other hand, after the coherent growing,
it escapes from the resonant band as shown in Figs. 3(b),
4(b), 5, and 8(b) and (c). The trapped motion is almost
integrable and is regarded as a captured state in resonance
[23]. At a moment, the integrable motion may be interrupted
by small kicks that happen during the resonant band. That is,
the socalled release from resonance [23] happens, and the
collective motion crosses the homoclinic orbit in Figs. 3(b),
4(b), 5, and 8(b) and (c), and hence it goes away from
the resonant band. It is therefore said that global instability
47th ÌEEE CDC, Cancun, Mexico, Dec. 911, 2008 WeA18.4
2491
Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on June 10, 2009 at 14:48 from IEEE Xplore. Restrictions apply.
New England power grid graph equivalent reduced grid
Objectives: How are the two electricallyequivalent networks related in
terms of graph topology, spectrum, resistance, . . . ?
Applications of Kron reduction in smart grid problems?
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 6 / 23
Summary of activities of last 2 years
Research Areas: “power systems engineering” ⊕ “networked control”
1
synchronization and transient stability
2
detection and identiﬁcation of faults and cyberphysical attacks
3
model reduction and scalability
Academics
Education: PhD students Florian D¨orﬂer and Fabio Pasqualetti
Publications: 3 journal articles (SIAM & IFAC), 9 conference articles
Awards: two plenaries, two best paper awards
201114 NSF project “CyberPhysical Challenges of Transient
Stability and Security in Power Grids.”
NSF CyperPhysical Systems and Trustworthy Computing Programs
Collaboration with Ian Dobson (Wisconsin) and Bruno Sinopoli (CMU)
Collaboration with Los Alamos National Lab, DOE project
“Optimization and Control Theory for Smart Grids”, Misha Chertkov
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 7 / 23
Summary of activities of last 2 years
Research Areas: “power systems engineering” ⊕ “networked control”
1
synchronization and transient stability
2
detection and identiﬁcation of faults and cyberphysical attacks
3
model reduction and scalability
Academics
Education: PhD students Florian D¨orﬂer and Fabio Pasqualetti
Publications: 3 journal articles (SIAM & IFAC), 9 conference articles
Awards: two plenaries, two best paper awards
201114 NSF project “CyberPhysical Challenges of Transient
Stability and Security in Power Grids.”
NSF CyperPhysical Systems and Trustworthy Computing Programs
Collaboration with Ian Dobson (Wisconsin) and Bruno Sinopoli (CMU)
Collaboration with Los Alamos National Lab, DOE project
“Optimization and Control Theory for Smart Grids”, Misha Chertkov
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 7 / 23
Outline
1
Introduction and Motivation
2
Mathematical Modeling & Synchronization Problem
3
Synchronization in the Kuramoto Model
4
From the Kuramoto Model to the Power Network Model
5
Conclusions
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 7 / 23
Structurepreserving power network model
“New England Power Grid”
15
5
12
11
10
7
8
9
4
3
1
2
17
18
14
16
19
20
21
24
26
27
28
31
32
34 33
36
38
39
22
35
6
13
30
37
25
29
23
1
10
8
2
3
6
9
4
7
5
F
Fig. 9. The New England test system [10], [11]. The system includes
10 synchronous generators and 39 buses. Most of the buses have constant
active and reactive power loads. Coupled swing dynamics of 10 generators
are studied in the case that a linetoground fault occurs at point F near bus
16.
test system can be represented by
˙
δ
i
= ω
i
,
H
i
πf
s
˙ ω
i
= −D
i
ω
i
+ P
mi
−G
ii
E
2
i
−
10
j=1,j=i
E
i
E
j
·
· {G
ij
cos(δ
i
−δ
j
) + B
ij
sin(δ
i
−δ
j
)},
(11)
where i = 2, . . . , 10. δ
i
is the rotor angle of generator i with
respect to bus 1, and ω
i
the rotor speed deviation of generator
i relative to system angular frequency (2πf
s
= 2π ×60 Hz).
δ
1
is constant for the above assumption. The parameters
f
s
, H
i
, P
mi
, D
i
, E
i
, G
ii
, G
ij
, and B
ij
are in per unit
system except for H
i
and D
i
in second, and for f
s
in Helz.
The mechanical input power P
mi
to generator i and the
magnitude E
i
of internal voltage in generator i are assumed
to be constant for transient stability studies [1], [2]. H
i
is
the inertia constant of generator i, D
i
its damping coefﬁcient,
and they are constant. G
ii
is the internal conductance, and
G
ij
+ jB
ij
the transfer impedance between generators i
and j; They are the parameters which change with network
topology changes. Note that electrical loads in the test system
are modeled as passive impedance [11].
B. Numerical Experiment
Coupled swing dynamics of 10 generators in the
test system are simulated. E
i
and the initial condition
(δ
i
(0), ω
i
(0) = 0) for generator i are ﬁxed through power
ﬂow calculation. H
i
is ﬁxed at the original values in [11].
P
mi
and constant power loads are assumed to be 50% at their
ratings [22]. The damping D
i
is 0.005 s for all generators.
G
ii
, G
ij
, and B
ij
are also based on the original line data
in [11] and the power ﬂow calculation. It is assumed that
the test system is in a steady operating condition at t = 0 s,
that a linetoground fault occurs at point F near bus 16 at
t = 1 s−20/(60 Hz), and that line 16–17 trips at t = 1 s. The
fault duration is 20 cycles of a 60Hz sine wave. The fault
is simulated by adding a small impedance (10
−7
j) between
bus 16 and ground. Fig. 10 shows coupled swings of rotor
angle δ
i
in the test system. The ﬁgure indicates that all rotor
angles start to grow coherently at about 8 s. The coherent
growing is global instability.
C. Remarks
It was conﬁrmed that the system (11) in the New Eng
land test system shows global instability. A few comments
0 2 4 6 8 10
5
0
5
10
15
δ
i
/
r
a
d
10
02
03
04
05
0 2 4 6 8 10
5
0
5
10
15
δ
i
/
r
a
d
TIME / s
06
07
08
09
Fig. 10. Coupled swing of phase angle δ
i
in New England test system.
The fault duration is 20 cycles of a 60Hz sine wave. The result is obtained
by numerical integration of eqs. (11).
are provided to discuss whether the instability in Fig. 10
occurs in the corresponding real power system. First, the
classical model with constant voltage behind impedance is
used for ﬁrst swing criterion of transient stability [1]. This is
because second and multi swings may be affected by voltage
ﬂuctuations, damping effects, controllers such as AVR, PSS,
and governor. Second, the fault durations, which we ﬁxed at
20 cycles, are normally less than 10 cycles. Last, the load
condition used above is different from the original one in
[11]. We cannot hence argue that global instability occurs in
the real system. Analysis, however, does show a possibility
of global instability in real power systems.
IV. TOWARDS A CONTROL FOR GLOBAL SWING
INSTABILITY
Global instability is related to the undesirable phenomenon
that should be avoided by control. We introduce a key
mechanism for the control problem and discuss control
strategies for preventing or avoiding the instability.
A. Internal Resonance as Another Mechanism
Inspired by [12], we here describe the global instability
with dynamical systems theory close to internal resonance
[23], [24]. Consider collective dynamics in the system (5).
For the system (5) with small parameters p
m
and b, the set
{(δ, ω) ∈ S
1
× R  ω = 0} of states in the phase plane is
called resonant surface [23], and its neighborhood resonant
band. The phase plane is decomposed into the two parts:
resonant band and highenergy zone outside of it. Here the
initial conditions of local and mode disturbances in Sec. II
indeed exist inside the resonant band. The collective motion
before the onset of coherent growing is trapped near the
resonant band. On the other hand, after the coherent growing,
it escapes from the resonant band as shown in Figs. 3(b),
4(b), 5, and 8(b) and (c). The trapped motion is almost
integrable and is regarded as a captured state in resonance
[23]. At a moment, the integrable motion may be interrupted
by small kicks that happen during the resonant band. That is,
the socalled release from resonance [23] happens, and the
collective motion crosses the homoclinic orbit in Figs. 3(b),
4(b), 5, and 8(b) and (c), and hence it goes away from
the resonant band. It is therefore said that global instability
47th ÌEEE CDC, Cancun, Mexico, Dec. 911, 2008 WeA18.4
2491
Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on June 10, 2009 at 14:48 from IEEE Xplore. Restrictions apply.
2
1
0
3
0
25
8
37
29
9
3
8
23
7
3
6
22
6
3
5
19
4
3
3
20
5
34
10
3
3
2
6
2
3
1
1
8
7
5
4
3
18
17
26
27
28
24
21
16
15 14
13
12
11
1
3
9
9
Power network topology:
1
n generators , each connected to a bus •◦
2
m buses •◦ form a connected graph
3
admittance matrix Y
network
∈ C
(n+m)×(n+m)
characterizes the network
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 8 / 23
Structurepreserving power network model
1
n generators :
• modeled by dynamic swing equations:
M
i
¨
θ
i
+ D
i
˙
θ
i
= P
mech.in,i
−P
electr.out,i
2
10
30
25
8
37
29
9
38
23
7
3
6
22
6
3
5
19
4
33
20
5
34
10
3
32
6
2
3
1
1
8
7
5
4
3
18
17
26
27
28
24
21
16
15 14
13
12
11
1
3
9
9
2
m buses •◦:
• load model: constant real power demand
& linear frequency dependence
D
i
˙
θ
i
+ P
load,i
= −
n+m
j =1
V
i
 · V
j
 · Y
ij
 · sin
θ
i
−θ
j
Y
jk
Y
ik
D
k
k
P
load,k
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 9 / 23
Structurepreserving power network model
Classic structurepreserving model [A.R. Bergen & D. Hill ’81]:
M
i
¨
θ
i
+ D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {1, . . . , n}
D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {n + 1, . . . m}
P
ij
= V
i
 · V
j
 · Y
ij
 ≥ 0 max. power transferred i ↔j
P
i
=
P
mech.in,i
for i ∈ {1, . . . , n}
−P
load,i
for i ∈ {n + 1, . . . m}
real power injection at i
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 10 / 23
Synchronization and transient stability
M
i
¨
θ
i
+ D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {1, . . . , n}
D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {n + 1, . . . m}
1
General synchronization problem: θ
i
−θ
j
 bounded &
˙
θ
i
=
˙
θ
j
2
Classic analysis methods: Assumptions & Hamiltonian arguments
highly developed ﬁeld based on global system perspective
Open Problem [D. Hill & G. Chen ’06]:
synchronization in power networks
?
underlying graph properties
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 11 / 23
Detour: Synchronization of coupled oscillators
It all began with two pendulum clocks and the
observation of “an odd kind of sympathy ”.
[Huygens, C. Horologium Oscillatorium, 1673]
Today’s canonical coupled oscillator model
[A. Winfree ’67, Y. Kuramoto ’75]
Kuramoto model of coupled oscillators:
˙
θ
i
= ω
i
−
K
n
n
j =1
sin(θ
i
−θ
j
)
n oscillators with phase θ
i
∈ S
1
nonidentical natural frequencies ω
i
∈ R
1
elastic coupling with strength K/n
x
x
x
ω
1
ω
3
ω
2
S
1
K/n
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 12 / 23
Detour: Synchronization of coupled oscillators
Kuramoto model of coupled oscillators:
˙
θ
i
= ω
i
−
K
n
n
j =1
sin(θ
i
−θ
j
)
Just a few of various direct & related applications:
Sync of coupled pendulum clocks [C. Huygens XVII, M. Bennet et. al ’02]
Sync in Josephson junctions [S. Watanabe et. al ’97, K. Wiesenfeld et al. ’98]
Sync in a population of ﬁreﬂies [G.B. Ermentrout ’90, Y. Zhou et al. ’06]
Deepbrain stimulation and neuroscience
[P.A. Tass ’03, E. Brown et al. ’04]
Coordination of particle models
[R. Sepulchre et al. ’07, D. Klein et al. ’09]
Countless other technological, biological, & social sync
phenomena [A. Winfree ’67, S.H. Strogatz ’00, J. Acebr´ on ’01]
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 13 / 23
Detour: Synchronization of coupled oscillators
A wellunderstood linear synchronization
model [M. DeGroot ’74, J. Tsitsiklis ’84,. . . ]
Consensus protocol:
˙ x
i
= −
n
j =1
a
ij
(x
i
−x
j
)
n identical agents with state variable x
i
∈ R
interaction described by a connected graph with weights a
ij
a few applications: robotic coordination, sensor networks, distributed
computation & optimization, continuous Markov chains, . . .
R
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 14 / 23
“The big picture”
Consensus protocols: Kuramoto oscillators:
?
Open problem: sync in power networks
state, parameters, and topology of graph
˙
θ
i
= ω
i
−
n
j=1
K
n
sin(θ
i
−θ
j
)
M
i
¨
θ
i
+D
i
˙
θ
i
= P
i
−
n
j=1
P
ij
sin(θ
i
−θ
j
)
˙ x
i
= −
n
j=1
a
ij
(x
i
−x
j
)
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 15 / 23
“The big picture”
Consensus protocols: Kuramoto oscillators:
?
Open problem: sync in power networks
state, parameters, and topology of graph
˙
θ
i
= ω
i
−
n
j=1
K
n
sin(θ
i
−θ
j
)
M
i
¨
θ
i
+D
i
˙
θ
i
= P
i
−
n
j=1
P
ij
sin(θ
i
−θ
j
)
˙ x
i
= −
n
j=1
a
ij
(x
i
−x
j
)
Previous “observations” about this connection:
Power systems: [D. Subbarao et al., ’01, G. Filatrella et al., ’08, V. Fioriti et al., ’09]
Networked control: [D. Hill et al., ’06, M. Arcak, ’07]
Dynamical systems: [H. Tanaka et al., ’97, A. Arenas ’08]
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 15 / 23
Outline
1
Introduction and Motivation
2
Mathematical Modeling & Synchronization Problem
3
Synchronization in the Kuramoto Model
4
From the Kuramoto Model to the Power Network Model
5
Conclusions
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 15 / 23
Synchronization in the Kuramoto Model
Classic Homogeneous Kuramoto model
˙
θ
i
= ω
i
−
K
n
n
j =1
sin(θ
i
−θ
j
)
Notions of synchronization:
1
phase cohesiveness: θ
i
(t) −θ
j
(t) bounded
2
frequency synchronization:
˙
θ
i
(t) =
˙
θ
j
(t)
Classic intuition:
K small & ω
i
−ω
j
 large
⇒ incoherence & no sync
K large & ω
i
−ω
j
 small
⇒ cohesiveness & frequency sync
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 16 / 23
Synchronization in the Kuramoto Model
Classic homogeneous Kuramoto model:
˙
θ
i
= ω
i
−
K
n
n
j =1
sin(θ
i
−θ
j
)
Necessary and suﬃcient condition [F. D¨orﬂer & F. Bullo ’10]
The following two statements are equivalent:
1
Coupling dominates nonuniformity, i.e., K > K
critical
ω
max
−ω
min
.
2
The Kuramoto model achieves phase cohesiveness & exponential
frequency synchronization for all ω
i
∈ [ω
min
, ω
max
].
Some consequences:
1) phase cohesiveness depends on ratio K/K
critical
2) frequency synchronization to average ω
sync
=
n
i =1
ω
i
/n
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 17 / 23
Synchronization in the Kuramoto Model
Topological Kuramoto model
˙
θ
i
= ω
i
−
n
j =1
a
ij
sin(θ
i
−θ
j
)
Assume: coupling graph is
undirected and connected
x
x
x
ω
1
ω
3
ω
2
a
12
a
13
a
23
S
1
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 18 / 23
Synchronization in the Kuramoto Model
Topological Kuramoto model
˙
θ
i
= ω
i
−
n
j =1
a
ij
sin(θ
i
−θ
j
)
Assume: coupling graph is
undirected and connected
1
Necessary conditions: no sync if
n
j =1
a
ij
≤
ω
i
−
n
k=1
ω
k
n
2
Various suﬃcient conditions in the literature [F. D¨orﬂer & F. Bullo, ’09]:
Based on algebraic connectivity: λ
2
>
ω
1
−ω
2
, . . .
2
3
Under development are exact conditions for synchronization
⇒ implication: “coupling dominates nonuniformity” ⇒ sync
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 18 / 23
Synchronization in the Kuramoto Model
Topological Kuramoto model
˙
θ
i
= ω
i
−
n
j =1
a
ij
sin(θ
i
−θ
j
)
Assume: coupling graph is
undirected and connected
1
Necessary conditions: no sync if
n
j =1
a
ij
≤
ω
i
−
n
k=1
ω
k
n
2
Various suﬃcient conditions in the literature [F. D¨orﬂer & F. Bullo, ’09]:
Based on algebraic connectivity: λ
2
>
ω
1
−ω
2
, . . .
2
3
Under development are exact conditions for synchronization
⇒ implication: “coupling dominates nonuniformity” ⇒ sync
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 18 / 23
Synchronization in the Kuramoto Model
Topological Kuramoto model
˙
θ
i
= ω
i
−
n
j =1
a
ij
sin(θ
i
−θ
j
)
Assume: coupling graph is
undirected and connected
1
Necessary conditions: no sync if
n
j =1
a
ij
≤
ω
i
−
n
k=1
ω
k
n
2
Various suﬃcient conditions in the literature [F. D¨orﬂer & F. Bullo, ’09]:
Based on algebraic connectivity: λ
2
>
ω
1
−ω
2
, . . .
2
3
Under development are exact conditions for synchronization
⇒ implication: “coupling dominates nonuniformity” ⇒ sync
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 18 / 23
Outline
1
Introduction and Motivation
2
Mathematical Modeling & Synchronization Problem
3
Synchronization in the Kuramoto Model
4
From the Kuramoto Model to the Power Network Model
5
Conclusions
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 18 / 23
From the Kuramoto Model to the Power Network Model
1) Structurepreserving power network model on T
n+m
×R
n
:
M
i
¨
θ
i
+ D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {1, . . . , n}
D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {n + 1, . . . m}
2) Nonuniform variation of Kuramoto model on T
n+m
:
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {1, . . . , n + m}
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 19 / 23
From the Kuramoto Model to the Power Network Model
1) Structurepreserving power network model on T
n+m
×R
n
:
M
i
¨
θ
i
+ D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {1, . . . , n}
D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {n + 1, . . . m}
2) Nonuniform variation of Kuramoto model on T
n+m
:
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {1, . . . , n + m}
Synchronization conditions can be related via
1
extension of 1
st
order analysis [D. Koditschek ’88, Y.P. Choi et al. ’10]
2
time scale separation analysis [F. D¨ orﬂer & F. Bullo ’09]
3
today: local topological equivalence [F. D¨ orﬂer & F. Bullo ’11]
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 19 / 23
From the Kuramoto Model to the Power Network Model
⇒ near the equilibrium manifolds 1) synchronizes ⇔ 2) synchronizes
⇒ near the equilibrium manifolds 1) & 2) are topologically conjugate
22 F. D¨orﬂer and F. Bullo
0.5 1 1.5
−0.5
0
0.5
θ(t)
˙
θ
(
t
)
0.5 1 1.5
−0.5
0
0.5
θ(t)
˙
θ
(
t
)
θ(t) [rad] θ(t) [rad]
˙
θ
(
t
)
[
r
a
d
/
s
]
˙
θ
(
t
)
[
r
a
d
/
s
]
Fig. 5.1. Phase space plot of a network of n = 4 secondorder Kuramoto oscillators (1.3) with
n = m (left plot) and the corresponding ﬁrstorder scaled Kuramoto oscillators (5.8) together with
the scaled frequency dynamics (5.9) (right plot). The natural frequencies ω
i
, damping terms D
i
,
and coupling strength K are such that ω
sync
= 0 and K/K
critical
= 1.1. From the same initial
conﬁguration θ(0) (denoted by ) both ﬁrst and secondorder oscillators converge exponentially to
the same nearby phaselocked equilibria (denoted by •) as predicted by Theorems 5.1 and 5.3.
(Φ
γ,0
(t), 0
m×1
). Hence, the phasesynchronized orbit (Φ
γ,0
(t), 0
m×1
), understood as
a geometric object in T
n
×R
m
, constitutes a onedimensional equilibrium manifold of
the multirate Kuramoto model (5.11). After factoring out the translational invari
ance of the angular variable θ, the exponentiallysynchronized orbit (Φ
γ,0
(t), 0
m×1
)
corresponds to an isolated equilibrium of (5.11) in the quotient space T
n
\ S
1
×R
m
.
Since an isolated equilibrium of a smooth nonlinear system with bounded and Lips
chitz Jacobian is exponentially stable if and only if the Jacobian is a Hurwitz matrix
[30, Theorem 4.15], the locally exponentially stable orbit (Φ
γ,0
(t), 0
m×1
) must be
hyperbolic in the quotient space T
n
\ S
1
× R
m
. Therefore, the equilibrium trajec
tory (Φ
γ,0
(t), 0
m×1
) is exponentially stable in T
n
×R
m
if and only if the Jacobian of
(5.11) evaluated along (Φ
γ,0
(t), 0
m×1
), has n +m−1 stable eigenvalues and one zero
eigenvalue corresponding to the translational invariance in S
1
.
By an analogous reasoning we reach the same conclusion for the ﬁrstorder multi
rate Kuramoto model (5.6) (formulated in a rotating frame with frequency ω
sync
)
and for the scaled Kuramoto model (5.8): the exponentiallysynchronized trajectory
Φ
γ,0
(t) ∈ T
n
is exponentially stable if and only if the Jacobian of (5.8) evaluated
along Φ
γ,0
(t) has n − 1 stable eigenvalues and one zero eigenvalue. Finally, recall
that the multirate Kuramoto model (5.11), its ﬁrstorder variant (5.6) together with
frequency dynamics (5.7) (in a rotating frame), and the scaled Kuramoto model (5.8)
together with scaled frequency dynamics (5.9) are all instances of the parameterized
system (5.1). Therefore, by Theorem 5.1, the corresponding Jacobians have the same
inertia and local exponential stability of one system implies local exponential stability
of the other system. This concludes the proof of the equivalences (i) ⇔ (ii) ⇔ (iii).
We now prove the ﬁnal conjugacy statement. By the generalized Hartman
Grobman theorem [17, Theorem 6], the trajectories of the three vector ﬁelds (5.11),
(5.6)(5.7) (formulated in a rotating frame), and (5.8)(5.9) are locally topologically
conjugate to the ﬂow generated by their respective linearized vector ﬁelds (locally
near (Φ
γ,0
(t), 0
m×1
). Since the three vector ﬁelds (5.11), (5.6)(5.7), and (5.8)(5.9)
are hyperbolic with respect to (Φ
γ,0
(t), 0
m×1
) and their respective Jacobians have the
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 20 / 23
Main Synchronization Results
Structurepreserving power network model:
M
i
¨
θ
i
+ D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {1, . . . , n}
D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {n + 1, . . . m}
1
necessary condition: Let ω
sync
=
k
P
k
/
k
D
k
, then
n+m
j =1
P
ij
≤ P
i
−D
i
· ω
sync
 ⇒ no sync
2
suﬃcient condition: Let
˜
P
i
= P
i
−D
i
· ω
sync
, then
λ
2
>
˜
P
1
−
˜
P
2
, . . .
2
⇒ sync
3
conjectured exact conditions are under development
bottom line: “coupling dominates imbalance in active power”
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 21 / 23
Main Synchronization Results
Structurepreserving power network model:
M
i
¨
θ
i
+ D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {1, . . . , n}
D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {n + 1, . . . m}
1
necessary condition: Let ω
sync
=
k
P
k
/
k
D
k
, then
n+m
j =1
P
ij
≤ P
i
−D
i
· ω
sync
 ⇒ no sync
2
suﬃcient condition: Let
˜
P
i
= P
i
−D
i
· ω
sync
, then
λ
2
>
˜
P
1
−
˜
P
2
, . . .
2
⇒ sync
3
conjectured exact conditions are under development
bottom line: “coupling dominates imbalance in active power”
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 21 / 23
Main Synchronization Results
Structurepreserving power network model:
M
i
¨
θ
i
+ D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {1, . . . , n}
D
i
˙
θ
i
= P
i
−
n+m
j =1
P
ij
sin(θ
i
−θ
j
) , i ∈ {n + 1, . . . m}
1
necessary condition: Let ω
sync
=
k
P
k
/
k
D
k
, then
n+m
j =1
P
ij
≤ P
i
−D
i
· ω
sync
 ⇒ no sync
2
suﬃcient condition: Let
˜
P
i
= P
i
−D
i
· ω
sync
, then
λ
2
>
˜
P
1
−
˜
P
2
, . . .
2
⇒ sync
3
conjectured exact conditions are under development
bottom line: “coupling dominates imbalance in active power”
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 21 / 23
Main Synchronization Results
Illustration with the IEEE Reliability Test System ’96 under severe loading
220
309
310
120
103
209
102 102
118
307
302
216
202
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 22 / 23
Main Synchronization Results
Illustration with the IEEE Reliability Test System ’96 under severe loading
t [s]

θ
i
(
t
)
−
θ
j
(
t
)

[
r
a
d
]
γ
min
t [s] t [s]
˙
θ
i
(
t
)
[
r
a
d
/
s
]
θ
i
(
t
)
[
r
a
d
]
θ
i
θ(t)
˙
θ(t)
ω
i
−D
i
ω
sync
Conjectured Kuramoto sync condition is marginally satisﬁed
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 22 / 23
Main Synchronization Results
Illustration with the IEEE Reliability Test System ’96 under severe loading

θ
i
(
t
)
−
θ
j
(
t
)

[
r
a
d
]
π/2
t [s] t [s]
˙
θ
i
(
t
)
[
r
a
d
/
s
]
θ
i
(
t
)
[
r
a
d
]
50 s
7 rad
50 s
3 rad/s
−3 rad/s
15 rad
−10 rad 50 s
t [s]
Conjectured Kuramoto sync condition is marginally not satisﬁed
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 22 / 23
Outline
1
Introduction and Motivation
2
Mathematical Modeling & Synchronization Problem
3
Synchronization in the Kuramoto Model
4
From the Kuramoto Model to the Power Network Model
5
Conclusions
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 22 / 23
Conclusions
Structurepreserving
power system model
2
30 25
37
29
38
23
36
22
35
19
33
20
34
10
32
6 31
1
8
7
5
4
3
18
17
26
27
28
24
21
16
15 14
13
12
11
39
9
10
9
7
6
4
5
3
2
1
8
Networkreduced
power system model
Kron
reduction
Consensus
protocols
& robotic
coordination
Coupled
Kuramoto
oscillators
Open problem: sync in power networks
state, parameters, and topology of graph
ﬁrst principle
modeling
mathematical
analysis
graph
theory
X
2
10
30 25
8
37
29
9
38
23
7
36
22
6
35
19
4
33 20
5
34
10
3
32
6
2
31
1
8
7
5
4
3
18
17
26
27
28
24
21
16
15 14
13
12
11
1
39
9
Ambitious workplan: sharpest conditions for most realistic models
from synchronization analysis to control design
F. D¨orﬂer & F. Bullo (UCSB) Synchronization in Power Networks Inst. for Energy Eﬃciency 23 / 23