State-Space Model for a Multi-Machine System

These notes parallel section 3.4 in the text. Note that we are
dealing with classically modeled machines (IEEE Type 0.0),
constant impedance loads, and a network reduced to its internal
machine terminals.

We have found that the linearized swing equation is given by:

2 H i d 2 Δδ i n
+ ∑ PSij Δδ ij = 0 , i=1,…,n (eq. 3.26)
ωRe dt 2 i =1
i≠ j
where it is important to note that we have assumed no damping.
Let’s consider writing the linearized swing equations for a test
system (see example 3.1 in text), as shown below.

The three swing equations are:

2 H1 d 2 Δδ1
+ PS 12 Δδ12 + PS 13Δδ13 = 0
ωRe dt 2
2 H 2 d 2 Δδ 2
+ PS 21Δδ 21 + PS 23Δδ 23 = 0
ωRe dt 2

2 H 3 d 2 Δδ 3
+ PS 31Δδ 31 + PS 32 Δδ 32 = 0
ωRe dt 2

1
One important fact: The stability of a power system depends on
relative rotor angles Δδij NOT absolute rotor angles Δδi. This is
because synchronism is a relative phenomenon. That is, it makes
no sense to say “generator 1 is in synchronism.” Rather, we must
say with what it is in synchronism, i.e., “Generator 1 is in
synchronism with generators 2 and 3,” or “Generator 1 is in
synchronism with the rest of the system.”

So we need to define our states in terms of relative rotor angles. In

the above equations, the states (derivatives) are in terms of
absolute rotor angle. We can deal with this in the following way.

First, multiply through each equation by ωRe/2Hi, resulting in:

d 2 Δδ1 ωRe
2
+ [PS12Δδ12 + PS13Δδ13 ] = 0
dt 2 H1
d 2 Δδ 2 ωRe
+ [PS 21Δδ 21 + PS 23Δδ 23 ] = 0
dt 2 2H 2
d 2 Δδ 3 ωRe
2
+ [PS 31Δδ 31 + PS 32Δδ 32 ] = 0
dt 2H 3
Now subtract the last equation from each of the other two. When
we do this, the derivative terms will be affected in the following
way (and this is the main motivation for making this subtraction):
d 2 Δδ1 d 2 Δδ 3 d 2 (Δδ1 − Δδ 3 ) d 2 Δδ13
− = =
dt 2 dt 2 dt 2 dt 2
which is dω13/dt. Note, however, that this very convenient
substitution of variable will not help if damping is modeled (Diωi)
in the swing equation AND the damping is nonuniform
(nonuniform damping means that Di≠Dk for i≠k) because then it is
not possible to combine the corresponding speed variables. Let’s
look at this issue by re-writing the above equations with
nonuniform damping.

2
 d 2 Δδ1 D1 dΔδ1 ωRe
+ + [PS12Δδ12 + PS13Δδ13 ] = 0
dt 2 2 H1 dt 2 H1
d 2 Δδ 2 D2 dΔδ 2 ωRe
+ + [PS 21Δδ 21 + PS 23Δδ 23 ] = 0
dt 2 2 H 2 dt 2H 2
d 2 Δδ 3 D3 dΔδ 3 ωRe
+ + [PS 31Δδ 31 + PS 32 Δδ 32 ] = 0
dt 2 2 H 3 dt 2H 3
Subtracting the last equation from the other two (in the case of the
first equation) affects the derivative terms in the following way:
d 2 Δδ1 d 2 Δδ 3 d 2 (Δδ1 − Δδ 3 ) d 2 Δδ13
2
− 2
= 2
= 2
(as before)
dt dt dt dt

D1 dΔδ1 D3 dΔδ 3
−
2 H1 dt 2 H 3 dt Î Cannot do more with this!
It will work if the damping is uniform, however, implying that the
ratio Di/Hi is the same for all i. In this case, we get, for example,
D1 dΔδ1 D3 dΔδ 3 D ⎛ dΔδ1 dΔδ 3 ⎞ D1 dΔδ13 D
− = 1 ⎜ − ⎟= = 1 Δω13
2 H1 dt 2H 3 dt 2 H1 ⎝ dt dt ⎠ 2 H1 dt 2 H1

The implication is that we can ALWAYS reduce the number of

states by 1 due to the ability to use relative angles. But an
additional reduction of states by 1 due to the ability to use relative
speeds only occurs in the cases of no damping or of uniform
damping. In general, we do need to model nonuniform damping
and therefore we are usually only able to get the state reduction for
relative angles, but not for relative speeds.

In this particular special case of no damping, we get:

3
d 2 Δδ13 ωRe ωRe
+ [PS 12 Δδ 12 + PS 13 Δ δ 13 ] − [PS 31Δδ 31 + PS 32Δδ 32 ] = 0
dt 2 2 H1 2H 3
d 2 Δδ 23 ωRe ωRe
2
+ [ PS 21Δ δ 21 + P S 23 Δ δ 23 ] − [PS 31Δδ 31 + PS 32Δδ 32 ] = 0
dt 2H 2 2H3

Recognizing that Δδ31=-Δδ13 and that Δδ32=-Δδ23, we may change

the sign of the second term in each equation if we also make this
change of variables. This results in:
d 2 Δδ13 ωRe ωRe
+ [PS 12 Δδ 12 + PS 13 Δ δ 13 ] + [PS 31Δδ13 + PS 32Δδ 23 ] = 0
dt 2 2 H1 2H3
d 2 Δδ 23 ωRe ωRe
2
+ [ PS 21Δ δ 21 + P S 23 Δ δ 23 ] + [PS 31Δδ13 + PS 32Δδ 23 ] = 0
dt 2H 2 2H3

eq. (#)
So we have states are Δδ13 and Δδ23. But note carefully that there
are some other variables, namely Δδ12, Δδ21.

This means we have 4 variables and only 2 equations.

Can we express Δδ12 and Δδ21 in terms of Δδ13 and Δδ23. Clearly,
since Δδ12=-Δδ21, if we can do it for one, we can do it for the other.

This is done by noting first that

Δδ12 + Δδ 23 + Δδ 31 = 0 (eq. *)
We can prove this as follows:
Δδ12 + Δδ 23 + Δδ 31
= Δδ1 − Δδ 2 + Δδ 2 − Δδ 3 + Δδ 3 − Δδ1 = 0
Therefore, from eq. (*), we can write that
Δδ12 = −Δδ 23 − Δδ 31

4
Reversing the subscript order of each term on the right-hand-side,
and changing signs, we get:
Δδ12 = −Δδ 23 + Δδ13 (eq. **)
Then, since Δδ12=-Δδ21, we get
Δδ 21 = Δδ 23 − Δδ13 (eq. ***)
Substituting eq. (**) and (***) into eq (#), we obtain
d 2 Δδ13 ωRe ωRe
+ [P S 12 Δδ 13 − PS 12 Δδ 23 + PS 13 Δδ 13 ] + [PS 31Δδ13 + PS 32Δδ 23 ] = 0
dt 2 2 H1 2H 3
d 2 Δδ 23 ωRe ωRe
+ [ PS 21Δ δ 23 − P S 21Δ δ 13 + PS 23 Δ δ 23 ] + [PS 31Δδ13 + PS 32 Δδ 23 ] = 0
dt 2 2H 2 2H3

Gathering terms in each variable, we get two differential equations:

d 2 Δδ 13 ⎡ω ω ω ⎤ ⎡ω ω ⎤
+ ⎢ Re PS12 + Re PS13 + Re PS 31 ⎥ Δδ 13 + ⎢ Re PS 32 − Re PS12 ⎥ Δδ 23 = 0
dt 2 ⎣1
2H 1
444444
2H 1 2H
4244443444 3⎦ ⎣ 24
1
H3
444244
2H 1
443⎦
α 11 α12

d Δδ 23 ⎡ ωRe
2
ωRe ⎤ ⎡ ωRe ωRe ωRe ⎤
+ ⎢ 2H PS 31 − PS 21 ⎥ Δδ 13 + ⎢ 2H PS 21 + PS 23 + PS 32 ⎥ Δδ 23 = 0
dt 2 ⎣1434424 ⎦ ⎣ 3⎦
2H 2 2H 2 2H 3
443 144444
2
42444444
α 21 α 22

Denote the coefficients of the above differential equations as α11,

α12, α21, and α22, where (assuming the last equation, for bus n, is
the one that gets subtracted off in the above steps):
⎧ ⎫
⎪ ωn ⎪ ω ω Re ω Re
α ii = ⎨∑ Re PSij ⎬ + Re PSni α ij = PSnj − PSij
⎪ j =1 2 H i ⎪ 2H n 2H n 2H i
⎩ j ≠i ⎭
Note these expressions differ from those given in the text’s
addendum, pg. 650, because my alpha’s here are defined on the
left-hand-side of the equation, whereas the text’s equations are
defined on the right-hand-side of the equation.

Using these alphas, I rewrite the differential equations as

5
 d 2 Δδ13
+ α11Δδ13 + α12 Δδ 23 = 0
dt 2
d 2 Δδ 23
+ α 21Δδ13 + α 22 Δδ 23 = 0
dt 2
We can now convert these second order linear differential
equations into first order linear differential equations, in order to
develop a state-space form.

We do this by recognizing that

dδ13 dδ 23
Δω13 = , Δω23 =
dt dt
Then, the above two second order differential equations become
four first order differential equations, as follows:
Δω&13 = −α11Δδ13 − α12 Δδ 23 = 0
Δω& 23 = −α 21Δδ13 − α 22 Δδ 23 = 0
Δδ&13 = Δω13
Δδ&23 = Δω23
So let’s define the state vector as
⎡ Δδ13 ⎤
⎢ Δδ ⎥
x = ⎢ 23 ⎥
⎢ Δω13 ⎥
⎢ ⎥
Δω
⎣ 23 ⎦
Note for our 3 machine system, we have only 4 states due to the
state reduction for relative angles and relative speeds.

Then
x& = A x
More explicitly,

6
 ⎡ Δδ&13 ⎤ ⎡ 0 0 1 0⎤ ⎡ Δδ13 ⎤
⎢ & ⎥ ⎢ 0 0 1⎥ ⎢ Δδ 23 ⎥
⎢ Δδ 23 ⎥ = ⎢ 0
⎥⎢ ⎥
⎢ Δω&13 ⎥ ⎢ − α11 − α12 0 0⎥ ⎢ Δω13 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣ 23 ⎦ ⎣− α 21 − α 22
Δω& 0 0⎦ ⎣Δω23 ⎦
Your text on page 63 shows the computation of the alpha-
coefficients for the 9-bus, 3-generator system of Fig. 2.19. It is
shown that
α11=104.096
α12=59.524
α21=33.841
α22=153.460
Then, the state-space equation is:

⎡ Δδ&13 ⎤ ⎡ 0 0 1 0⎤ ⎡ Δδ13 ⎤
⎢ & ⎥ ⎢ 0 1⎥ ⎢ Δδ 23 ⎥
⎢ Δδ 23 ⎥ = ⎢ 0 0
⎥⎢ ⎥
⎢ Δω&13 ⎥ ⎢− 104.096 − 59.524 0 0⎥ ⎢ Δω13 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
ω
⎣ 23 ⎦ ⎣ − 33.841 − 153.460
Δ & 0 0⎦ ⎣Δω23 ⎦
Question is, now, what to do with the above in order to obtain
useful information about the small-disturbance behavior of our
system. We will investigate this next….