Академический Документы
Профессиональный Документы
Культура Документы
tor. L AQ lQ
I. INTRODUCTION
ld l F1D
rF
power engineering applications. There are a number of practi-
rD
cal models available for synchronous generators depending on lF
Section IV. rF
ia
The theory of observers and observability is a well re- vF iF LF
II. SYNCHRONOUS GENERATOR MODELING where the current vectors are defined as,
A synchronous generator can be modeled by a matrix dif- [ ]
i0 dq = i0 id iq T and iabc = [ia ib ic ] .
T (5)
ferential equation relating voltages, currents, flux linkages, Similarly, the transformed voltages and flux linkages are,
and operational parameters of the machine. The matrix equa-
v 0dq = Pv abc and λ0 dq = Pλabc . (6)
tion describing the synchronous generator depends on the ro-
tor model that is selected. The model developed in this re- Equation (1) in its expanded form becomes,
search work consists of three stator windings, one field wind- 0 i a λ&a v n
va ra 0 0 0 0 0
ing and three damper windings as shown in Fig. 3. Magnetic
coupling is a function of the rotor position and therefore, the
v 0 r 0 0 ib λ&b v n
0 0 0
b b
flux linking each winding is also a function of the rotor posi- vc 0 0 rc 0 ic λ&c v n
0 0 0
− v = − 0 0 0
tion [10], [11]. The instantaneous terminal voltage of any 0 i F − λ&F + 0 . (7)
rF 0 0
winding takes the form, F
− v D 0 0 0 0 i D λ&D 0
0 rD 0
v = − ri − λ& , (1)
− v 0 0 0 0 iG λ& 0
0 0 rG
where r is the winding resistance, i is the current and λ is the G G
− vQ 0 0 0 rQ iQ λ& 0
0 0 0
flux linkage. It should be noted that in this notation it is as- Q
sumed that the direction of positive stator currents is out of the Equation (7) is transformed into a 0dq frame of reference
terminals, since the synchronous machine under consideration through a Park’s transformation according to (4) and (6). The
is a generator. system of equations is converted into its per unit form and the
In (1) the voltage is expressed in terms of both currents and following relations are used to simplify the model [10].
flux linkages. This is not desirable and therefore one of the L d = L AD + l d Lq = L AD + l q
two variables has to be replaced. The flux linkage variables L F = L AD + l F LG = L AD + l G
can be substituted by appropriate expressions that relate them L D = L AD + l D LQ = L AD + l Q
to generator currents via machine inductances [10], [11].
kM F = kM D = M X = L AD
kM G = kM Q = M Y = L AQ (8)
3
The resulting synchronous generator model that is used for ration curve for a synchronous machine. This saturation curve
the parameter estimation is thus given by, will aid in analyzing the saturation of synchronous generator
v0 r + 3rn 0 0 0 0 0 0 i0 inductances.
v 0 r ω( LAQ + l q ) 0 0 ωLAQ ωLAQ id
d
vq 0 − ω( LAD + l d ) r − ωLAD − ωLAD 0 0 iq
− v = − 0 0 0 rF 0 0 0 iF
F
−
Dv 0 0 0 0 rD 0 0 iD
− v 0 0 0 0 0 rG 0 iG
G
− v
Q 0 0 0 0 0 0 r Q iQ
L0 + 3Ln 0 0 0 0 0 0 i&0
0 LAD + l d 0 LAD LAD 0 0 i&
d
1
0 0 L AQ + l q 0 0 L AQ L AQ i&q (9)
− 0 LAD 0 LAD + l F LAD 0 0 i& ,
ωB F
0 L AD 0 L AD L AD + l D 0 0 i&D
0 0 LAQ 0 0 LAQ + l G LAQ i&G
0 0 LAQ 0 0 LAQ LAQ + l Q i&Q
where Ksd and Ksq represent the saturation factors in each axis tion factors for the direct and quadrature axes can then be cal-
and the subscript u refers to the unsaturated value of the induc- culated as,
tance. The two saturation factors represent the degree of satu- 1
ration in the direct and quadrature axes. The saturation factors K sd =
1 + S GD
are in general not the same for both axes. 1
In the case of salient pole machines, Ksq is assumed to be K sq = , (14)
1 + S GQ
unity for all loading conditions and thus LAQ is considered a
constant inductance [10], [11]. This is because the path for the where S GD = S GQ = S G in the case of a round rotor synchro-
quadrature axis flux is mainly in air and therefore the satura- nous generator.
tion experienced by LAQ is insignificant. In the case of round
rotor machines, saturation is significant in both the direct and IV. DAMPER WINDING THEORY
quadrature axes. Both saturation factors need to be calculated
The synchronous generator model that was developed in
from the d axis and q axis saturation curves. However, q axis
Section II essentially formulates a multiple input-multiple
saturation curves are typically not available and it is assumed
output (MIMO) system. The system is formed in a state equa-
that the q axis saturation factor is equal to the d axis saturation
tion of the form,
factor [10]. The assumption of equal saturation factors effec- (15)
tively assumes that the reluctance of the magnetic path is ho- V = − RI − LI& ,
mogeneous around the periphery of the rotor [10]. The as- where the matrices R and L contain operational parameters of
sumption of equal saturation effects also implies that the mag- the generator, and the vectors V, I, and I& contain the voltages
netic excitation in the direct and quadrature axes is equal. and currents that are assumed to be measured directly from the
Other methods to calculate the q axis saturation factors are generator terminals.
proposed in the literature [1], [10], [13]. Equation (9) is the full matrix equation of the model of the
The saturation factors Ksd and Ksq may be calculated from synchronous generator. An examination of the voltage and
the saturation curve as supplied by the manufacturer. There- current vectors shows that the last three equations of the
model involve the voltage across and the current through each
fore, a saturation function needs to be defined to simulate the
one of the three damper windings D, G, and Q. Damper (am-
actual saturation curve. A number of saturation functions
ortisseur) windings form one of the most important factors for
have been proposed in the literature. One such function that
damping in power systems. The action of damper windings is
correctly duplicates the saturation curve provided by the very crucial to the operation of electrical generators and to the
manufacturer for this research work is given by, stability of the power system as a whole.
S = A e BG (Vt −V0 ) . (11) In the case of salient pole generators, damper windings
G G
Vt is the open circuit terminal voltage at the operating condi- consist of metal bars placed in slots in the pole faces and con-
nected together at each end. These bars can be connected to-
tion and V0 is the assumed saturation threshold, which can be
gether via a closed ring on both sides of the pole. In this case,
obtained from the saturation curve [11]. V0 is the per unit ter-
they are called complete or connected damper windings. It is
minal voltage at the point where the OCC deviates from the
also possible for these bars not to be connected in between the
air-gap line. AG and BG are constants and they can be deter- poles, but each pole to have its own independent set of metal
mined by the knowledge of two points on the saturation curve. bars. In this case, the damper windings are known as incom-
These points are usually the 1.0 p.u. and 1.2 p.u. terminal volt- plete, non-connected, or open windings [19].
ages as shown in Fig. 4. In the case of round rotor generators (steam and gas tur-
To calculate AG and BG the value of the function at the two bines), their rotors are made up of solid steel forgings [10].
selected terminal voltages needs to be calculated first. From These generators do not usually have damper windings but the
Fig. 4 at the two terminal voltages selected the values of SG solid steel rotor core provides a path for eddy currents. The
are given by, eddy currents in the rotor path produce the same effect as
I − IA damper windings. In some cases, certain manufacturers pro-
S G1.0 = B vide for additional damping effects and negative sequence
IA
braking by using interconnected metal wedges in the field
I − 1.2 I A
S G1.2 = C , (12) winding slots or by providing separate copper rods underneath
1.2 ⋅ I A the wedges [10].
Using (12) to solve for the constants AG and BG, it can be There are several reasons for providing damper windings
shown that [11], for synchronous machines. Damper windings provide starting
S2 torque for synchronous motors, condensers, and converters,
AG = G1.0 while they are used to suppress hunting. Hunting is the
1.2S G1.2 damped mechanical oscillation of the rotor about its new
1.2 S G1.2 (13) steady state angle after the mechanical speed of the rotor has
BG = 5 ln .
S changed. The suppression of hunting has been the first appli-
G1 . 0
cation of the damper windings.
Using the values for AG, BG and the saturation threshold V0, Further, damper windings are used to damp oscillations that
the saturation function SG can be calculated for any operating are started by switching or faults. The existence of damper
condition by calculating the terminal voltage Vt. The satura-
5
windings in this case causes the oscillations to damp out be seen in Fig. 6. An error signal between the plant and the
faster. In the case of asymmetrical faults, the damper currents observer is generated, and control over the state error can be
provide a braking torque, and therefore the accelerating torque achieved. Appropriate selection of the eigenvalues of the (A-
is reduced during the fault. Another important application of LC) matrix can cause the error to go to zero at a relatively
damper windings is the balancing of the terminal voltage of short time.
each phase during unbalanced loading. Damper windings
decrease the negative sequence reactance and therefore they u(t)
Plant y(t)
decrease the negative sequence voltage [19].
Additionally, during current surges in the armature circuit
(in case of internal faults), the damper windings reduce the
x 1 (t)
stress on the insulation of the field winding by the induced
x 2 (t)
flux through the windings. Finally, the damper windings pro- Model
vide additional torque for synchronizing generators. They x n (t)
help to pull the generator back into step after synchronism is
lost because of a fault [19]. Fig. 5. Open loop observer
In the case of the voltages in (9), all damper winding volt- Plant
ages vD, vG, and vQ are zero since there is no voltage source in
u(t) x& ( t ) = Ax ( t ) + Bu ( t ) y(t)
the damper windings. In the case of a doubly excited syn-
y ( t ) = Cx (t ) + Du ( t )
chronous generator the model of (9) can still describe the syn-
chronous machine but vG will not be zero. As indicated previ-
ously, damper currents are zero (or very small) in the steady Observer
state operation of the synchronous generator, and nonzero xˆ ( t )
xˆ& ( t ) = A xˆ ( t ) + Bu ( t )
during transients. However, it is not possible to measure the yˆ (t ) = C xˆ ( t ) + Du ( t )
damper currents directly using physical instruments, even in
the case that the damper windings are not fictitious. There- yˆ ( t )
V. OBSERVERS AND OBSERVABILITY The concept of an observer for a dynamic process was in-
In general, a linear system may be represented in a state troduced by Luenberger in 1966 [6]. Luenberger recognized
space realization of the form, that often the full state vector of a system is not available, ei-
x& (t ) = Ax(t ) + Bu (t ) ther because there is no access to all the states or because the
y (t ) = Cx (t ) + Du (t ) , (16) instrumentation required to measure all the states is very ex-
pensive. Therefore, it is desired to obtain an approximation of
where x(t) ∈ ℜ
n
is the vector of the states of the system, the unavailable states so as to enable the design of feedback or
u(t) ∈ ℜ is the input to the system, and y(t) ∈ ℜ is the out-
m p other control laws even in cases where the entire state vector is
put of the system. For the purposes of the observer design, A unknown.
One of the variations of the Luenberger observer is the
∈ ℜ nxn , B ∈ ℜ nxm , C ∈ ℜ pxn , and D ∈ ℜ pxm must be time identity observer [5], [7]. Let the system under consideration
invariant and known except in the case of adaptive observers, be of the form,
where it is possible to estimate both the states and some of the x& (t ) = Ax(t ) + Bu (t )
parameters of the system simultaneously. y (t ) = Cx (t ) , (17)
Observers are dynamic systems that use current and past
values of the plant input, output, and known states, to generate where all the matrices are as defined in (16) and are time-
an estimate of the unknown (unmeasurable) states. The plant invariant and known. The system is assumed to be n dimen-
parameters are required to be known and the plant should be sional and the output vector is p dimensional. It is also neces-
observable. A simple test for observability is the requirement sary for the system to be completely observable. The identity
that the subsystem of (16) described by (A, C) has full column observer is of the form,
rank [20]-[22]. The synchronous generator model of (9) can z& (t ) = Az (t ) + E[ y (t ) − Cz (t )] + Bu (t ) , (18)
be shown to be observable.
where E ∈ ℜ , and is selected so that the eigenvalues of A-
nxp
Various observer configurations have been proposed in the
literature and are mainly categorized into open loop observers EC have negative real parts so that the error is reduced to zero
and closed loop observers. Open loop observers are simple in in a finite time.
construction, but since the complete dynamics of the plant are The error tends to zero at a rate determined by the domi-
rarely known exactly, no control is provided over the error nant eigenvalue of A-EC, and the state of the observer will
generated by the observer [4]. A general open loop observer converge to the state of the observed system. A good practice
can be seen in Fig. 5. is to select the eigenvalues of A-EC to be more negative than
An improvement over the open loop observer is the closed the eigenvalues of the observed system so that the conver-
loop observer. A schematic diagram of such an observer can gence is fast enough. However, the eigenvalues cannot be
6
moved arbitrarily towards minus infinity, since that will force VI. DEVELOPMENT OF AN OBSERVER FOR A ROUND ROTOR
the observer to act like a differentiator and be extremely sensi- SYNCHRONOUS GENERATOR
tive to noise [7]. An identity observer is shown in Fig. 7. The Usually, available data for synchronous generators are the
identity observer estimates the whole state vector. stator phase currents and voltages at the terminals of the ma-
chine, and the field voltage and current. If damper current
B
measurements are required in a certain application, then it is
u (t )
B
I x (t )
C
y (t )
E
I z (t )
C
Cz (t )
necessary to estimate the damper currents by means of an ob-
s
s
server. The general concept of an observer is as follows: cer-
A A tain states of a physical system may be difficult to measure or
calculate. These unobserved states may nonetheless be needed
to calculate an estimate of the machine parameters. An ‘ob-
Fig. 7. Luenberger identity observer for a linear system server’ is a dynamic system that is constructed so that the un-
observed states may be estimated. The observer is adaptive:
An improvement over the identity observer is the reduced parameters of the observer are adjusted methodically so that
order observer. Since usually some of the states are measur- the output of the machine simulation agrees with the actual
able, it is possible to construct an observer of order n-p rather
measured machine output.
than n (p are the measured states and n is the dimension of the
For the construction of the observer, the last three equations
full state). This provides better performance and uses a dy-
of the synchronous generator model in (9) can be rearranged
namic system of lower dimension, which improves the speed
of the observation. Such an observer can be seen in Fig. 8. As so as to obtain expressions for the damper winding currents.
before, the system matrices need to be time invariant and The three equations are given by,
known. 1 1 1
− v D = 0 = − rD i D − L AD i d′ − L AD i F′ − ( L AD + l D )i D′ (19)
ωB ωB ωB
D 1 1 1
u (t ) plant − v G = 0 = − rG i G − L AQ i q′ − ( L AQ + l G )iG′ − L AQ iQ′ (20)
I x (t ) y (t ) ωB ωB ωB
B C
s 1 1 1
− v Q = 0 = −rQ i Q − L AQ i q′ − L AQ iG′ − ( L AQ + l Q )iQ′ (21)
ωB ωB ωB
A
where for the purposes of the development of the observer the
D current derivatives are replaced by the forward difference
e (t )
E formula,
i (t + ∆t ) − i (t )
I xˆ (t ) i ′(t ) ≈ . (22)
B s
C ∆t
yˆ ( t )
Solving (19) for iD, an expression can be obtained in terms
A-EC of known measurements and the value of iD at the previous
observer D step.
Fig. 8. Luenberger observer for a linear system [20] rD ω B ∆t L AD ∆t
i D (n + 1) = (1 − )i D (n) − (i d′ (n) + i F′ (n)) (23)
L AD + l D L AD + l D
Luenberger observers have the advantage that they are
The quadrature axis damper winding currents iG and iQ can
simple to construct and cheap. They can be simplified to a
be obtained by the simultaneous solution of (20) and (21) and
reduced dimension observer so as not to estimate the whole
state vector if it is already available. Luenberger observers are given by,
can be employed within many modern control strategies and ( L AQ + l Q )rG ω B ∆t L AQ rQ ω B ∆t
i G (n + 1) = 1 − i G (n) + i Q ( n)
are applicable to both continuous and discrete systems. Fi- k AQ k AQ
nally, if they are well designed, they do not suffer from noise L AQ ∆t L AQ l G
amplification and are ideal for applications where the model of + − 1i q′ (n) (24)
( L AQ + l G ) k AQ
the system is well known.
On the other hand, if the model is not well known, the re- ( L AQ + l G )rQ ω B ∆t L AQ rG ω B ∆t
duced order Luenberger observer will not perform as desired. i Q (n + 1) = 1 − i Q ( n) + i G ( n)
k AQ k AQ
The identity observer will offer better results, but still the un-
L AQ l G ∆t
certainty that always exists in the parameters will cause some − i q′ (n) , (25)
errors in the observation. Also, care must be taken in the se- k AQ
lection of matrix E that specifies the rate of convergence of the where k AQ = ( L AQ + l Q )( L AQ + l G ) − L2AQ .
estimation. If the eigenvalues of A-EC are very small (to-
Equations (23)-(25) enable the calculation of the damper
wards minus infinity), the observer will act as a differentiator
currents. All parameters can be accurately calculated using
and the system noise will be amplified. Finally, Luenberger
manufacturer’s data, while the time varying quantities are
observers cannot be applied in case A, B, C, and D are not
available measurements. The only ambiguity in the observer
completely known and time invariant.
equations is the value of iD(0), iG(0), and iQ(0). These are
7
needed to initiate the observation process. Nevertheless, the conditions assumed to be zero. All estimated damper currents
initial conditions can be assumed to be zero without loss of match the simulated damper currents and are equal to zero to
accuracy as will be shown in the two case studies in the next five decimal places, as expected from a synchronous generator
section. that operates in steady state. The error between the estimated
and simulated currents can be calculated using the formula,
VII. APPLICATIONS USING A DAMPER WINDING CURRENTS i simulated − iobserved 2
OBSERVER %error = × 100 , (26)
i simulated 2
The foregoing discussion of damper winding observers and
magnetic saturation are presented with a view to model and where . 2 denotes the 2-norm (square root of the sum of the
identify parameters for synchronous generators. In this sec- squares of all the elements). The errors in the three damper
tion, a brief illustration of the type of application of the ob- winding currents iD, iG, and iQ for the steady state data case are
servers and saturation models is given. For this purpose, a 0.16%, 0.28%, and 0.14% respectively.
synchronous generator was simulated using the Electromag- In the second simulated case study, transient data were con-
netic Transients Program (EMTP) both in the steady state and sidered. A permanent line to line fault was applied at 0.25
in the transient mode. EMTP is suitable for testing the estima-
seconds between phases b and c. The observed damper cur-
tion algorithm because data from the simulation are free of
rents as compared to the EMTP simulated damper currents for
noise, and one has access to all machine parameters and sig-
each axis can be seen in Figs. 9, 10, and 11. As can be seen
nals. Furthermore, EMTP provides the simulated damper cur-
rents in the direct and quadrature axes. Therefore, a means of from the graphs, the estimated damper currents match the
testing the observer is provided. The observer cannot be simulated damper currents. The percent errors between the
tested using actual measurements from a synchronous genera- simulated and estimated damper currents can be calculated in
tor, since damper currents are unmeasurable. the same way as in the steady state case. The errors in the
The machine under consideration is a cross-compound gen- three damper winding currents iD, iG, and iQ for the transient
erator located in the southwest U.S.A. The generator contains data case are 0.03%, 0.63%, and 0.16% respectively. Since
a high pressure unit rated at 483 MVA and a low pressure unit the difference between the simulated and estimated signals is
rated at 426 MVA. Table I shows the parameters for the high not discernible in Figs. 9, 10, and 11, a portion of iQ as shown
pressure generator as calculated by manufacturer’s data. in Fig. 11 was magnified to visualize the difference between
These parameters are used in the EMTP program to obtain the the two signals. This difference is shown in Fig. 12.
required measurements. At this point it suffices to state that
the measurements required are the three phase stator voltages
and currents, and the DC field voltage and current. The meas-
urements are assumed to be error free for the simulation tests.
The tabulated parameters are unsaturated values at 40 ˚C.
TABLE I
OPERATIONAL PARAMETERS FOR A 483 MVA SYNCHRONOUS GENERATOR