8.1 Introduction
The presence of high voltage power electronic equipment in the transmission system has important consequences on the stability of the power system. However, the conventional multi-machine stability assessment described in the last chapter is inadequate to represent the transient behaviour of some power electronic devices. This problem is particularly important in the case of HVDC converters due to the occurrence of inverter commutation failures that often follow a large disturbance. Only the use of electromagnetic transient programs, described in Chapter 6, can provide accurate information of the power electronic response during and immediately following clearance of the disturbance. Although machine rotor dynamics, as discussed in Chapter 7, could be included in the electromagnetic transient program of Chapter 6 , this is not a practical proposition in the presence of power electronic switching, considering the different time constants involved. For instance, electromagnetic-transient simulations require steps of (typically) 50 ps, whereas the stability programs use steps at least 200 times larger. To reduce the computational requirements, some programs [ 11 contain two separate modes. An instantaneous mode is used to model components in three-phase detail with small time steps in a similar way to the EMTPEMTDC programs [2]. The alternative is a stability mode based on r.m.s. quantities at fundamental frequency only, with increased time-step lengths. The user can switch between the two modes as required while running but, in either mode, the entire system must be modelled in the same way. Thus, when using the instantaneous mode, a system of any substantial size would still be very computationally intensive. A more efficient solution is described here [3,4] which takes advantage of the computationally inexpensive dynamic representation of the a.c. system in the stability program, and the accurate transient simulation of the power electronic devices. The slow dynamics of the a.c. system are adequately modelled by the stability program while the fast dynamic responses of the HVDC and FACTS devices are accurately represented by electromagnetic transient simulation. This hybrid combination is essential to predict the first few swings in stability studies.
298
Finally, quasi steady-state models of the power electronic devices are also developed in this chapter for use in longer stability studies as well as system response to small perturbations.
7 -
299
detailed
Loadflow
Stability input
TSlEMTDC hybrid
Outflow
c-y
Interface
models produce the same results within a predefined tolerance and over a set period, the complete system can then be reconnected and used by TS,and the EMTDC representation terminated. This allows better computational efficiency, particularly for long simulation runs.
8.2.1
Data flow
Data for the EMTDC solution entered in the programme database via the PSCAD graphics. Equivalent circuits are used at each interface point to represent the rest of the system not included in the detailed model. This system is then run until steady
300
state is reached and a snapshot is taken. This snapshot holds all the relevant data for the various system components and is used as the starting point when interfacing the detailed model with the stability programme. The stability program is initialized conventionally through power-flow results via a data file. An interface data file is also read by the TSE hybrid that provides the number and location of interface buses, analysis options, and timing information. The data flow diagram is shown in Figure 8.3.
301
Yes
i
Solve EMTDC Extract information from detailed solution and include in stability programme Solve stability equations
T = T + step length
1 1 I
c=
Output results Is T = end time?
No
The EMTDC programme represents system 1 by a Thevenin equivalent (El and as shown in Figure 8.6. The simplest ZI is an R-L series impedance, representing the fundamental frequency equivalent of system 1. This impedance can be derived from the results of a power flow and a fault analysis at the interface bus. The power flow 'provides an initial current through the interface bus and the initial interface bus. A fault analysis is also used to determine the fault current through the
zl)
302
f
Figure 8.5 Hybrid interface
System 1
I,
System 2
interface for a short circuit fault to ground. The equivalent Thevenin source Thevenin impedance ZI values shown in Figure 8.7, are obtained as follows: Using the power flow solution:
El and
El = inZl + V ,
and using the fault circuit:
El = I F Z ~ .
can then be found from either Equation (8.1) or Equation (8.2). and During a transient, the impedance of the synchronous machines in system 1 will change. The net effect on the fundamental power in or out of the equivalent circuit, however, can be represented by varying the source and keeping ZI constant.
303
Figure 8.7
Derivation of Thevenin equivalent circuit (a) Power-flow circuit. (b) Fault circuit
However, the waveforms obtained from EMTDC involve frequencies other than the fundamental and, therefore, a frequency-dependent equivalent circuit (as described in section 6.4.3)is used to represent the a.c. system. Regarding the equivalent circuits sources, information from the EMTDC model representing system 2 (in Figure 8.6) is used to modify the source of the equivalent circuit of system 2 in the stability programme. Similarly, data from TS are used to modify the source of the equivalent circuit of system 1 in EMTDC. These equiva. 5 ) . From lent sources are normally updated at each TS step length (refer to section 8 Figure 8.6,if both ZI and 2 2 are known, additional information is still necessary to determine update values for the sources 7 , and El. The selection and processing of interface parameters to derive this information is discussed in section 8.4. Moreover, the stability programme requires only positive sequence data so data from the three a.c. phases at the interface(s) is analysed and converted to positive sequence, i.e. for the positive sequence voltage:
v,,
p, + ZVb + Z*V,),
= 1L 120").
(8.4)
where
V,,
- - V b , V,
= phase voltage,
304
Similarly, the positive sequence voltage from TS is converted to three phase data as follows:
Va
= Vps,
(8.5)
Some transient stability programmes have the additional capability of modelling negative and zero sequence networks. If this option is utilized, sequence data from a stability programme can then be converted to three phase through the following formulae:
Va
v b
= Vps
where
(8.8)
(8.9)
(8.10)
Similarly, any unbalance in system 2 can be accommodated in the transient stability programme.
El = TiZ1
+V ,
v = i2Z2,
i 2 = i,
+ 7,.
v =z1z2 + icZ2.
(8.14)
E~ = IIzIL(eII
+ ezl)+ w e V
sin(&
=I121cos(6Il + & l ) + j l l Z l
(8.15)
305
and
6 1 1 = ev - 9,
(8.16)
where rb is the displacement angle between the voltage and the current. Thus, Equation (8.15) can be written as
EI = IlZl cos(8v
=IIZl(cos8vcosB-sinevsinB)+ vcosev
B = ezI- 9.
If then equating the real terms:
Ei = E l , + j E l i ,
E l , = (7,Zl cosB+ V)cos6v
+ (-1121sinB)sin&,
(8.18)
is known and constant throughout the simulation. where From the EMTDC results, the values of V , I and 4 are also known and, hence, so is j3. El can be determined in the TS phase reference frame from knowledge of and the previous values of interface current and voltage from TS, through the use of Equation (8.1 1). From Equation (8.18), making
A = IlZi
C O S ~
zl
zI
+ V,
(8.19) (8.20)
B =-IIZ~ sin B,
and remembering that
A cos ev
+ B sin ev = J F T 3 c o s ( e v
*),
(8.21)
where
the voltage angle Bv in the TS phase reference frame can be calculated, i.e.:
(+= tan-'
v
22
)I:[
(8.23)
Lerl,
(8.24)
306
In a similar way, data from the transient stability programme simulation can be used to calculate a new Thevenin source voltage magnitude for the equivalent circuit of system 1 in the EMTDC programme. Knowing the voltage and current magnitude at the TS programme interface and the phase difference between them, by a similar analysis the voltage angle in the EMTDC phase reference frame is:
-I,
COS~,
4 = ev -ell,
+ = tan-'
I$[
(8.29)
Knowing the EMTDC voltage angle 8v allows calculation of the EMTDC current angle from Equation (8.28). The magnitude value of El can then be derived from Equation (8.1 1).
= pf
ph.
(8.30)
The r.m.s. power can be extracted directly from the waveforms and, therefore, Fourier transform or curve fitting methods are not necessary, i.e.: (8.31) This method greatly reduces the computing time necessary for the data extraction from EMTDC. The choice of r.m.s. power was made on the basis that harmonic power flow will result only from in-phase components of harmonic voltage and current. The assumption was made that if a system contains only a low resistive component, then the harmonic power flow is not significant [6]. This, however, is not valid for every situation and, particularly, at the inverter end of an HVDC link, the effect of the resistive component of the network is not insignificant.
307
Certain harmonic frequencies in a network may also be parallel resonant or close to parallel resonance and exhibit more resistance than reactance. The presence of a transient may excite the resonant frequency and greatly affect the results. It is important, therefore, to model the fundamental power flow accurately. Another factor to consider is that the direction of fundamental power may not necessarily be the same as the direction of harmonic power. With an HVDC link, while fundamental power is drawn into a rectifier, much of the harmonic power flow will be in the reverse direction. The r.m.s. real power measured is the difference between these two powers and is, therefore, not entirely representative of the fundamental power load of the HVDC link. Conversely, at the inverter end the r.m.s. real power will include harmonic power flow into the a.c. system. This may exaggerate the amount of true fundamental power from the inverter. As an illustration of the problem, Figure 8.8 shows an EMTDC simulated result of a single phase fault at the inverter end of the CIGRE HVDC benchmark model [7] The instantaneous power is shown, along with the total r.m.s. power over discrete one cycle intervals. For hybrid interfacing purposes, data is transferred at discrete intervals equal to the stability programme time-step and these are significantly larger than the interval between the discrete output points constituting the instantaneous power. The r.m.s. power is then taken over a fundamental period to represent its use in the hybrid model. Figure 8.9 shows an analysis of the single-phase fault comparing fundamental frequency power with the total r.m.s. power. The fundamental frequency power was derived using the curve fitting method described in section 8.5 to extract both fundamental voltage and current. The comparison shows that, particularly during the
Instantaneous power
lo00
800 600
F 3
400
200
0
1 a -200
-400
-600
-800
-1000
, I I I
.O
308
1000
900 -
- 700 800
600-
z i
500400200 -
$ 300100
I
1
0.04
0.08
0.12
0.16
0.20
0.24
0.28
0.32
0.36
0.40
fault time, there exists a significant amount of Ph or harmonic r.m.s. power in the total r.m.s. power. When the three phase network is unbalanced the fundamental frequency real power consists of positive, negative and zero sequence components, i.e.: Pf = Pfps
+ Pfns + Pfzs,
(8.32)
and the negative and zero sequence powers cause additional power loss in the network. Figure 8.10 shows the sequence components of the fundamental r.m.s. power of Figure 8.9. Fundamental frequency negative sequence currents, in the presence of damper windings, can produce a braking torque which will retard the rotor [8]. Damper windings, however, also serve to lower the negative sequence impedance of a machine which in turn reduces the negative sequence voltage [9]. Which of these two opposing effects is dominant depends on the resistance of the damper windings. High resistance windings cause the braking torque to be the significant effect. The braking power of the negative sequence current is:
(8.33)
where
309
1000
-.
900
800
z
a
8 I
'
. . , . . . r
-
0
-100 -200
- ------ ---I
Figure 8.10
rn
The negative sequence resistance can be approximated from the rotor and the armature resistance, i.e.: (8.35) rn % ( ~ s
+i ~ r ) ,
where
R, = armature resistance.
Retardation of the rotor can also be caused by d.c. components in the armature windings. Three phase faults at or near machine terminals can cause d.c. components of short circuit armature current which can have a definite braking effect on the machine [9].The braking power in this case is:
(8.36) (8.37)
where
I , = the effective very low frequency value of the armature current,
idc
The total r.m.s. power then is not always equivalent toeither the fundamental frequency power or the fundamental frequency positive sequence power. A comparison of these
310
1000
900 -
800
700
600-
5 5003
g m
400300-
200
100 -
I c . . *
0-100-
7.-_ I . . .
-200 0.00
d
I
I t I
0.04
0.08
0.12
0.16
0.20
0.24
0.28
0.32
0.36
0. 0
Figure 8.11 Comparison of total r.m.s. power, fundamental frequency power and fundamental frequency positive sequence power
three powers is shown in Figure 8.11. The difference between total r.m.s. power and the positive sequence power can be seen to be highly significant during the fault. The most appropriate power to transfer from EMTDC to TS is then the fundamental frequency positive sequence power. This, however, requires knowledge of both fundamental frequency positive sequence voltage and fundamental frequency positive sequence current. These two variables contain all the relevant information and, hence, the use of any other power variable to transfer information becomes unnecessary.
(8.38)
where
+=d o .
8.5
311
(8.39)
Letting C1 = A cos(wTo) and C2 = A sin(wT0) and if sin(wt) and cos(ot) are represented by functions F I ( t ) and F2(t), respectively, then:
y(t> = CI FI ( t )
+ C2F2(t).
(8.40)
F l ( t ) and F 2 ( t ) are known if the fundamental frequency w is known. However, the amplitude and phase of this frequency generally need to be found, so the equation has to be solved for C I and C2. If the signal y ( t ) is distorted, then its deviation from a sinusoid can be described by an error function E :
~ ( t= ) y(t)
+E .
(8.41)
For a least squares method of curve fitting, {he size of the error function is measured by the sum of the individual residual squared values such that:
i= 1
+ iAr).
(8.43)
E=
C ( x ; - ~ I ~ l ( t-i ~) 2 ~ 2 ( t i ) ) ~ ,
i= I
where the residual value r at each discrete step is defined as: In matrix form: (8.45)
or
The error component can be described in terms of the residual matrix as follows:
E = [rlT[r]
312
(8.48)
and hence
n-1
(8.50)
Similarly:
and (8.5 1)
(8.52)
From these matrix element equations, CIand C2 can be calculated recursively using
sequential data.
313
2oo
-5.00 0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0.20
Time (s)
314
illustrated in Figure 8.13 is selected for the information exchange. It shows the use of the previous time-step data when passing information from TS to EMTDC but present time-step data when passing information from EMTDC to TS. When the two programs are run concurrently, TS passes its information to EMTDC. EMTDC then runs to t + A t , and the information gathered over this present time-step given back to TS at time t . TS now runs to t Ar, using equivalent circuit information derived for that particular time step. However, to cater for TS step lengths below the fundamental period, the above method is modified as shown in Figure 8.14; it relates to the case when the step length is one half of the fundamental period. Following the sequential numbering on Figure 8.14, at a particular point in time, the EMTDC and TS programs are concurrent and the TS information from system 1 is passed to update the system 1 equivalent in EMTDC. This is shown by the arrow marked 1. EMTDC is then called for a length of half a fundamental period (arrow 2) and the curve fitted results over the last full fundamental period processed and passed back to update the system 2 equivalent in TS (arrow 3). The information over this period is passed back to TS at the mid-point of the EMTDC analysis window which is half a period behind the current EMTDC time. TS is then run to catch up
-4
etc.
I
I I
1-1
TS
-2
6
4 c
II II
Figure 8.14
315
Stabiliity step length
*
t
I I
- /-I
8
etc.
TS
v p *
r
t
++
II
to EMTDC (arrow 4), and the new information over this simulation run used to again update the system 1 equivalent in EMTDC (arrow 5 ) . This protocol continues until any discontinuity in the network occurs. When a network change such as a fault application or removal occurs, the interaction protocol is modified to that shown in Figure 8.15. The curve fitting analysis process is also modified to avoid applying an analysis window over any point of discontinuity. The sequential numbering in Figure 8.15 explains the flow of events. At the fault time, the interface variables are passed from TS to the system 1 equivalent in EMTDC in the usual manner, as shown by the arrow marked 1. Neither system 1 nor system 2 has yet been solved with the network change. The fault is now applied in EMTDC which is then run for a full fundamental period length past the fault application (arrow 2) and the information obtained over this period passed back to TS (arrow 3). The fault is now also applied to the TS program which is then solved for a period until it has again reached EMTDCs position in time (arrow 4). The normal interaction protocol is then followed until any other discontinuity is reached. A full period analysis after the fault is applied is necessary to accurately extract the fundamental frequency component of the interface variables. The mechanically controlled nature of the a.c. system implies a dynamically slow response to any disturbance and so, for this reason, it is considered acceptable to run EMTDC for a full period without updating the system 1 equivalent circuit during this time.
316
An alternative approach has been proposed [ 101 where the interface location is extended out from the converter bus into the a.c. system. This approach maintains that, particularly for weak a.c. systems, a fundamental frequency equivalent representing the a s . system is not adequate at the converter terminals. In this case, the extent of the a.c. system to be included in the d.c. system depends on phase imbalance and waveform distortion. Although the above concept has some advantages, it also suffers from many disadvantages. The concept is proposed, in particular, for weak a.c. systems. A weak a.c. system, however, is likely to have any major generation capability far removed from the converter terminal bus as local generation serves to enhance system strength. However, when the generation is far away from the converter, then the distance required for an interface location to achieve considerably less phase imbalance and waveform distortion is also likely to be significant. The primary advantage of a hybrid solution is in accurately providing the d.c. dynamic response to a transient stability program, and in efficiently representing the dynamic response of a considerably sized a.c. system to the d.c. solution. Extending the interface some distance into the a.c. system, where the effects of a system disturbance are almost negligible, diminishes the hybrid advantage. If a sizeable portion of the a.c. system requires modelling in detail before an interface to a transient stability program can occur, then one might question the use of a hybrid solution at all and instead use a more conventional approach of a detailed solution with a.c. equivalent circuits at the system cut-off points. Another significant disadvantage in an extended interface is that ax. systems may well be heavily interconnected. The further into the system that an interface is moved, the greater the number of interface locations required. The hybrid interfacing complexity is thus increased and the computational efficiency of the hybrid solution decreased. The requirement for a detailed representation of a significant portion of the a.c. system serves to decrease this efficiency, as does the increased amount of processing required for variable extraction at each interface location. The advantages of using the converter bus are:
0 0 0 0
The detailed system is kept to a minimum. Interfacing complexity is low. Computational expense is minimized. Converter terminal equipment, such as filters, synchronous condensers SVCs, and can still be modelled in detail.
The major drawback of the detailed solution is in not seeing a true picture of the ax. system, since the equivalent circuit is fundamental frequency based. Waveform distortion and imbalance also make it difficult to extract the fundamental frequency information necessary to transfer to the stability program. The problem of waveform distortion for transfer of data from EMTDC to TS is dependent on the accuracy of the technique for extraction of interfacing variable information. If fundamental frequency quantities can be accurately measured under distorted conditions, then the problem is solved. Section 8.5 has described an efficient curve fitting algorithm to extract the required information from distorted waveforms. It has
317
Rectifier
1 Inverter
ESCR=2.0
h + @
been shown that, using the technique described, fundamental frequency quantities can be accurately measured. Moreover, a simple fundamental frequency equivalent circuit is insufficient to represent the correct impedance of the a.c. system at each frequency. Instead, this can be achieved by using a fully frequency-dependent equivalent circuit of the a.c. system [ l 11 at the converter terminal instead of just a fundamental frequency equivalent. A frequency-dependent equivalent avoids the need for modelling any significant portion of the ax. system in detail yet still provides an accurate picture of the system impedance across its frequency spectra. To show the effect of a frequency-dependent equivalent, the test system of Figure 8.16 is used with an inverter effective short circuit ratio of 2. The d.c. link is represented by the CIGRE benchmark model [7]. A three-phase solid short circuit fault is applied at the inverter terminal. Three cases are considered, the first being the entire system represented by the detailed solution. The second and third cases relate to the hybrid solution interfaced at the converter bus, one case with a fundamental frequency Thevenin representation of the stability program in the detailed solution, and the other with a frequency-dependent equivalent. The results of the inverter terminal voltages for the three cases are plotted in Figure 8.17, and they show that the benchmark EMTDC case and the frequencydependent equivalent case are identical. The fundamental frequency equivalent case (Figure 8.17(b)) shows more distortion and prolonged effects from the disturbance than the benchmark EMTDC case. These results show the inadequacy of a fundamental frequency equivalent at the converter terminal, whereas the use of a frequency-dependent equivalent is perfectly adequate.
318
-1.4
I
,
(a) EMTDC
1.4 1.2 1 .o 0.8 0.6 0.4 0.2
i
&
g) 0.0
-0.2
-0.6 -0.8 -1 .o -1.2 -1.4 I
>o -0.4
0.96
1.00
1.02
1.04
1.06
1.08
1.10
1.12
1.14
1.16
1.18
319
Event Isolate TS network ready to interface with EMTDC Initial time through interfacing routine Normal interaction between TS and EMTDC System network change signalled-run EMTDC for a full period past the fault and return information to TS TS catching up with EMTDC after a system network change
No
Yes
1
I
Compare variables
I
s = Ts+stepl
I
No 1
1 and2
Ts=Ts+steol
Figure 8.18 TSE hybrid algorithm. [step 11 TS program step length. [Ts] TS step length counter. [Tend] Minimum finish time for EMTDC. [K] Time counter for EMTDC and TS comparison of variables. [Tc] Time over which comparison must be within tolerance
320
eriod of TI2
variables to TS sys. 1
lr
+ = + &
K=K+stepl
I No
i
*
I
C
Figure 8.18 (Continued)
8.8
321
IC
stop
f
Pass TS,System 1
variables to EMTDC
%
Ts=Ts+step 1
Ts=O
Figure 8.18
(Continued)
into systems 1 and 2, and Norton equivalent circuits set up in each system. The state is now set to 1, representing normal interfacing conditions, and TS is solved until its time is concurrent with EMTDC. Under the normal interfacing of state 1, variable information is given to EMTDC from TS to update its equivalent circuit. The equivalent circuit of system 1 in the TS model of system 2 is updated in the same way. EMTDC is then called for one half a period and the information measured over the last fundamental period of its simulation . . . . . . , used to update the equivalent circuit or system L in 15. In the interface data file, a time can be specified as a minimum termination time for EMTDC, after any network disturbance is cleared. Once this time is reached in the program, the interface variables of both system 2 models are compared at each TS time-step. If they correlate within a predefined tolerance over a set period, then EMTDC can be terminated and the TS representation of system 2 once again takes river nurino the vsriahleq are b the ..." nerinrl y ,. . . , . . thpv -.."J rirp hpino -... rhprkpd -..--.-..-,if _- at anv ,time --.-._-.. -_ - outside the specified tolerance, the checking period time is restarted.
a .
r n "
" 1 1 1 .
Y"1 1 . .
I.w"
I --f
-_---_-I
322
When a network admittance change is about to occur, such as a fault being applied or removed, the interfacing routine is entered with a state value of 2. At this point in time, TS is concurrent with EMTDC. The TS interface variables are passed to EMTDC, which is then run for a full period from the network discontinuity time. It is not accurate to apply the analysis window over the network discontinuity, so the staggered window process must be restarted. The variables from EMTDC are passed back to TS at the time of the disturbance and the state is changed to 3. State 3 allows TS to catch up to EMTDC before the normal interfacing procedure of state 1 recommences.
Livingston
Aviemore
Figure 8.19 Test system on the a.c. side of the rectifier station
323
3 2000r
3000
o 2500-
.-
6
m
t
ii
1500500 -
E 1000-
8 2
1.20
7 1.00
0.800.603 0.400 0 0.20..... 0.00 I I I I I I I I 1.50 1.65 1.80 1.95 2.10 2.25 2.40 2.55 2.70 2.85 3.00 Time (s)
C
Figure 8.21 d.c. current response to a three-phase short circuit at the rectifier end
Following the power ramp, the a.c. current returns to its nominal value and after a small oscillatory settling of the controllers so do the d.c. voltage and current.
8.9.2
The results of the EMTDC and TS simulations show substantial differences and thus their independent use is inadequate to model the transient stability of the a.c. system. Figure 8.22(a) displays the rectifier terminal a.c. voltage levels calculated with the TS and TSE alternatives. The main difference between them is a transient overshoot in the TSE solution immediately after fault clearance. The rectifier d.c. currents, displayed for the three solutions in Figure 8.22(b), show a very similar variation for the TSE and EMTDC solutions, except for the region between I = 2.03 s and t = 2.14 s but the difference with the TS only solution is very pronounced. Figure 8.23 compares the fundamental positive sequence real and reactive powers across the converter interface for the TS and TSE solutions. The main differences in real power occur during the link power ramp. The difference is almost a direct relation to the d.c. current difference between TS and TSE shown in Figure 8.22(b). The oscillation in d.c. voltage and current as the rectifier terminal is de-blocked is also evident.
324
E (TS variable)
0.80
0.60 0.40 0.20 0.00
z
6
0
--
_ I
0.800.600.40 0.20 0.00 I I I I I I I I 1.50 1.65 1.80 1.95 2.10 2.25 2.40 2.55 2.70 2.85 3.00
(b)
Time (s)
Figure 8.22 Hybrid response comparisons. (a) Rectifier end a x . voltage. (b) Rectifier end d.c. current
1200 1000
- TSE
TS
I
5 8
600 400
0
I I
200
z. p 8
-200-
--
..........
...............
~
-400-600-800-
325
5 s
a 0 ,
-30 -50
I
I ~.,~,-I~,.R-,,-~--.
1 ,
--
-_-.- - - - - - -
lo-10-
pi F, !..,/*\ ,.,..-. . ,i \. !. ..
L........
......................................................................
Figure 8.24
As for the reactive power Q, prior to the fault, a small amount is flowing into the system due to a surplus MVArs at the converter terminal. The fault reduces this power flow to zero. When the fault is removed and the a.c. voltage overshoots in TSE, the reactive MVArs also overshoot in TSE and since the d.c. link is shut down, a considerable amount of reactive power flows into the system. Finally, Figure 8.24 shows the machine angle swings with respect to the Clyde generator (see test system of Figure 8.19). These indicate that the system is transiently stable.
The implementation of delay angle control is instantaneous. The transformer tap position remains unchanged throughout the stability study unless otherwise specified. The direct current is smooth, though its actual value may change during the study.
326
; .
'd*
'd
supply voltage without filtering. Rectifier loads can, therefore, be modelled as a single equivalent bridge with a sinusoidal supply voltage at the terminals but without representation of passive filters. This model is shown in Figure 8.25. Rectifier loads can utilize a number of control methods which can be modelled using a controlled rectifier with suitable limits imposed on the delay angle (a)[12].
Static loads Operating under constant current control, the d.c. equations are:
(8.53)
(8.54)
where A is the constant current controller gain and I d s is the normal d.c. current setting as shown in Figure 8.26. Constant current cannot be maintained during a large disturbance as a limit of delay angle will be reached. In this event, the rectifier control specification will become one
327
(8.55)
Protection limits and disturbance severity determine the rectifier operating characteristics during the disturbance. Shutdown occurs if I d reaches a set minimum or zero and the voltage Vload will cause shutdown before the a.c. terminal voltage reaches zero. The action of the rectifier load system is thus described by Equations (3.24), (3.30), (3.34), (3.36), (3.37), (8.53) and either (8.54) or (8.55).
Dynamic loads The basic rectifier load model assumes that current on the d.c. side of the bridge can change instantaneously. For some types of rectifier loads this may
be a valid assumption, but the d.c. load may well have an overall time constant which is significant with respect to the fault clearing time. In order to examine realistically the effects which rectifiers have on the transient stability of the system, this time constant must be taken into account. This requires a more complex model to account for extended overlap angles, when low commutating voltages are associated with large d.c. currents. When the delay angle (a)reaches a limiting value, the dynamic response of the d.c. current is given by (8.56) V d = IdRd Vload Ldpld,
where 4 represents the equivalent inductance in the load circuit. Substituting for using Equation (3.24) gives
Vd
where T d c = L d / R d . The controller time constant may also be large enough to be considered. However, in transient stability studies where large disturbances are usually being investigated, faults close to the rectifier load force the delay angle (a)to minimum very quickly. Provided the rectifier load continues to operate, the delay angle will remain at its minimum setting throughout the fault period and well into the post-fault period until the terminal voltage recovers. The controller will, therefore, not exert any significant control over the d.c. load current. Ignoring the controller time constant can, therefore, be justified in most studies.
Abnormal modes of converter operation The slow response of the d.c. current
when a large disturbance has been applied to the ax. system can cause the rectifier to operate in an abnormal mode. After a fault application near the rectifier, the near normal value of d.c. current ( I d ) needs to be commutated by a reduced a.c. voltage. This causes the commutation angle ( b )to increase and it is possible for it to exceed 60". This mode of operation is beyond the validity of the equations and to model the dynamic load effects accurately it is necessary to extend the model.
328
The full range of rectifier operation can be classified into four modes [13]:
Mode 1 -Normal operation. Only two valves in the bridge are involved in simultaneous commutation at any one time. This mode extends up to a commutation angle of 60". Mode 2 -Enforced delay. Although a commutation angle greater than 60" is desired, the forward voltage across the incoming thyristor is negative until either the previous commutation is complete or the firing angle exceeds 30". In this mode, p remains at 60" and a ranges up to 30". Mode 3 -Abnormal operation. In this mode, periods of three-phase short circuit and d.c. short circuit exist when two commutations overlap. During this period, there is a controlled safe short circuit which is cleared when one of the commutations is complete. During the short-circuit periods, four valves are conducting. Commutation cannot commence until 30" after the voltage crossover. Mode 4 -Continuous three-phase and d.c. short circuit caused by two commutations taking place continuously. In this mode, the commutation angle is 120" and the a.c. and d.c. current paths are independent.
The waveforms for these modes are shown in Figure 8.27, and Table 8.2 summarizes the conditions for the different modes of operation. Equations (3.24) and (3.30) do not apply for a rectifier operating in Mode 3 and they must be replaced by:
(8.58)
and
Id
45xc
(8.59)
Fourier analysis of the waveform leads to the relationship between ax. and d.c. current given by Equation (3.32), where the factor k is now:
(8.60)
where
a ' = CY - 30"
and
y' = y
( p ) is shown in Figure 8.28.
+ 30".
(8.61)
A graph showing the value of k for various delay angles (a) and commutation angles
Identification of operating mode The mode in which the rectifier is operating can be determined simply by use of a current factor K I . The current factor is defined as:
(8.62)
Substitution in this, using the relevant equations, yields limits for the modes.
329
Figure 8.27 Rectifier voltage waveforms showing different modes of operation. (a) Mode 1, g < 60"; (b) Mode 2, p = 60" with enforced delay al; (c) Mode 3, p z 60" with short-circuit 2 period a
Table 8.2 Rectifier modes of operation
Mode
1
Firing angle
0" 5 a 5 90" 0" 5 a 5 30"
Overlap angle
0" 5 /A 5 60"
60" 60" 5 p < 120" 120"
2 3
4
330
1.2
Mode 1:
K I 5 COS(~O"- a )
and
(8.63)
Mode 3:
43
2
43
2
2
(8.65)
Mode 4:
K I = - when a 5 30,
43
43
(8.66)
This can be demonstrated in the curve of converter operation shown in Figure 8.29. It can thus be seen that the mode of operation can be established prior to solving for the rectifier load equations at every step in the solution.
8.10.2
d.c. link
Provided that it can be safely assumed that a d.c. link is operating in the quasi steadystate (QSS) Mode 1, the equations developed for converters in Chapter 3 can be used. That is, the converters are considered to be controllable and fast acting so that the
331
Delay angle, a
normal steady-state type of model can be used at each step in the transient stability study. The initial steady-state operating conditions of the d.c. link will have been determined by a load flow and, in this, the control type, setting and margin will have been established. During the solution process at each iteration, the control mode must be established. This can be done by assuming Mode 1 (i.e. with the rectifier on C.C. control) and, by combining Equations (3.39), (3.45), (8.53) and (8.54), a d.c. current can be determined as: Idsr - [(3fi/n)aiVtermi cos Yicl/Ar (8.67) I d mode I = 1 (Rd - (3/n)xci)/Ar Assuming this current to be valid, then d.c. voltages at each end of the link can be calculated using Equations (3.24) and (3.39). The d.c. link is operating in mode 2 (i.e. with the inverter on C.C. control) i f
(8.68)
For constant power control, under control mode 1, the d.c. current may be determined from the quadratic equation:
where
Pds
pds = (Pdsr
+ pdsi)/2*
(8.7 1)
332
Table 8.3 Current setting for constant power control from quadratic equation
Id I Id2
Within
Outside
Within
Id I
Id2
Greater Greater
Less Less
Within = within the range / d m i n to / d m a x ; Outside = outside the range l d m i n to l d m a x ; Greater c Greater than I d m a x ; Less = Less than I d m i " .
The correct value for Idm&I can then be found from Table 8.3. Control mode 2 is determined using Equation (8.68) and in this case the following quadratic equation must be solved: k r z : mode 2 - k v l d mode 2 - P d marg - p d s = 0, (8.72) where (8.73) (8.74) If the link is operating under constant power control but with a current margin then for control mode 2:
It is possible for the d.c. link to be operating in control mode 2 despite satisfying the inequality of Equation (8.68). This occurs when the solution indicates that the rectifier firing angle (a,)is less than the minimum value ( a r m i n ) . In this case, the delay angle should be set to its minimum and a solution in mode 2 is obtained. It is also possible that when the link is operating close to the changeover between modes, convergence problems will occur in which the control mode changes at each iteration. This can easily be overcome by retaining mode 2 operation whenever detected for the remaining iterations in that particular time step.
d.c. power modulation It has been shown in the previous section that under the constant power control mode, the d.c. link is not responsive to a.c. system terminal conditions, i.e. the d.c. power transfer can be controlled disregarding the actual a.c. voltage angles. Since, generally, the stability limit of an ax. line is lower than its thermal limit, the former can be increased in systems involving d.c. links, by proper utilization of the fast converter controllability. The d.c. power can be modulated in response to a.c. system variables to increase system damping. Optimum performance can be achieved by controlling the d.c. system
333
so as to maximize the responses of the a.c. system and d.c. line simultaneously following the variation of terminal conditions. The dynamic performance under d.c. power modulation is best modelled in three separate levels [14]. These levels, illustrated in Figure 8.30, are (a) the a.c. system controller, (b) the d.c. system controller and (c) the ax.-d.c. network. (a) The ax. system controller uses ax. andor d.c. system information to derive the current and voltage modulation signals. A block diagram of the controller and a.c.-d.c. signal conditioner is shown in Figure 8.31.
(b) The d.c. system controller receives the modulation signals AI and A E and the
steady-state specifications for power PO current I 0 and voltage Eo. Figure 8.32(a) illustrates the power controller model, which develops the scheduled current setting; it is also shown that the current order undergoes a gradual increase during restart, after a temporary blocking of the d.c. link. The rectifier current controller, Figure 8.32(b), includes signal limits and rate limits, transducer time constant, bandpass filtering and a voltage dependent current order limit (VDCOL). The inverter current controller, Figure 8.32(c), includes similar components plus a communications delay and the system margin current ( Z , , , ) . Finally, the d.c. voltage controller, including voltage restart dynamics, is illustrated in Figure 8.32(d). (c) The d.c. current I d and voltage E d derived in the d.c. system controller constitute the input signals for the a.c.-d.c. network model which involves the steady state
ffY
I _
(a)
(b)
(4
Figure 8.30 a.c.-d.c. dynamic control structure: (i) a.c. system controller, (ii) d.c. system controller, (iii) ax.-d.c. network
334
3~
'07
N(P)
D(P)
Figure 8.32 d.c. system controller. (a) Power controller; (b) Rectifier current controller; (c) Inverter current controller; (d) d.c. voltage controller
solution of the d.c. system (neglecting the d.c. line dynamics which are included in the d.c. system controller). Here the actual a.c. and d.c. system quantities are calculated, i.e. control angles, d.c. current, voltage, active and reactive power. The converter a.c. system constraints are the open circuit secondary voltages Ear and Eai.
335
From the initial load flow, nominal bus shunt admittance (yo) can be calculated The for the rectifier. This is included directly into the network admittance matrix [Y]. current injected into the network in the initial steady state is, therefore, zero. In general:
where
and
?;*
30
Yo=Ivrermo12
(8.77)
(8.78)
The static load rectifier model does not depart greatly from an impedance characteristic and is well behaved for low terminal voltages, the injected current tending to zero as the voltage approaches zero. Figure 8.33 compares the current due to a rectifier with that due to a constant impedance load. As the injected current is never large, the iterative solution for all ax. conditions is stable. When the rectifier model is modified to account for the dynamic behaviour of the d.c. load, its characteristic departs widely from that of an impedance. Immediately after a fault application, the voltage drops to a low value but the injected current magnitude does not change significantly. Similarly, on fault clearing, the voltage recovers instantaneously to some higher value while the current remains low.
Ip
t
I
0.5
Vterm
1.o
linj
~F.L.
Figure 8.33 Difference between impedance and static load rectifier characteristic (for vload # 0)
336
When the load characteristic differs greatly from that of an impedance, a sequential solution technique can exhibit convergence problems, especially when the voltage is low. With small terminal voltages, the a.c. current magnitude of the rectifier load is related to the d.c. current but the current phase is greatly affected by the terminal voltage. Small voltage changes in the complex plane can result in large variations of the voltage and current phase angles. To avoid the convergence problems of the sequential solution, an alternative algorithm has been developed [ 151. This combines the rectifier and network solutions into a unified process. However, it does not affect the sequential solution of the other components of the power system with the network. The basis of this approach is to reduce the a.c. network, excluding the rectifier, to an equivalent Thevenin source voltage and impedance as viewed from the primary side of the rectifier transformer terminals. This equivalent of the system, along with the rectifier, can be described by a set of non-linear simultaneous equations which can be solved by a standard Newton-Raphson algorithm. The solution of the reduced system yields the fundamental a.c. current at the rectifier terminals. To obtain the network equivalent impedance, it is only necessary to inject 1 p.u. current into the network at the rectifier terminals while all other nodal injected currents are zero. With an injected current vector of this form, a solution of the nodal network Equation (7.48) gives the driving point and transfer impedances in the resulting voltage vector: (8.79) [Z] = [V']= [ T ] l [ 7 ; , 1 ,
(8.80)
The equivalent circuit shown in Figure 8.34 can now be applied to find the rectifier current ( 7,) by using the Newton-Raphson technique. The effect of the rectifier on the rest of the system can be determined by superposition: (8.81) [VI= [TI [ZII,,
where
[ T I = [r]-"l;j]
(8.82)
and [7Ynj]are the injected currents due to all other generation and loads in the system. If the network remains constant, vector is also constant and thus only needs re-evaluation on the occurrence of a discontinuity. Thus, the advantages of the unified and sequential methods are combined. That is, good convergence for a difficult element in the system is achieved while the programming for the rest of the system remains simple and storage requirements are kept low.
[z]
337
'I
Rd
vd=0
Vload
The equivalent system of Figure 8.34 contains seven variables ( V , e r m , I , , 8, @, a, and I d ) . With these variables, four independent equations can be formed. They are Equation (3.24) and:
Vd
Vl!@- V t e r m L 8 - Z,hl!(IpL$'
Vdld
= 0,
(8.83) (8.84)
- &aVtermlp
cOS(6 - 9) = 0.
Equation (8.83) is complex and represents two equations. Substituting for V d and I , using Equations (3.34) and (8.53) reduces the number of variables to five. A fifth equation is necessary and with constant current control, i.e. with the delay angle (a) within its limits, this can be written as
Id
- I d s p = 0.
(8.85)
Equation (3.24), suitably reorganized, and Equations (8.83)-(8.85)represent [ F ( X ) ]= 0 of the Newton-Raphson process and
(8.86) [XIT = [Er, 8, a,I d ] . When the delay angle reaches a specified lower limit (omin), the control specification, given by Equation (8.85) changes to
$9
a - amin = 0.
(8.87)
Equation (3.24) is no longer valid. The d.c. current ( I d ) is now governed by the differential Equation (8.57). If the trapezoidal method is being used, this equation can be transformed into an algebraic form similar to that described in Chapter 7 . Equation (3.24) is replaced by:
Id
= kaE, cosa - kb = 0.
(8.88)
The variables ka and kb contain information from the beginning of the integration step only and are thus constant during the iterative procedure:
ka = h / ( 2
+ kc
h),
(8.89)
338
where
(8.91)
t represents the time at the beginning of the integration step and h is the step length.
Commutation angle p is not explicitly included in the formulation, and since these equations are for normal operation, the value of k in Equation (3.35) is close to unity and may be considered constant at each step without loss of accuracy. On convergence, p may be calculated and a new k evaluated suitable for the next step. In Mode 3 operation, the value of k becomes more significant and for this reason the number of variables is increased to six to include the commutation angle p. The equations [ F ( X ) ] = 0 for the Newton-Raphson method in this case are:
VLp - VtemL8 -uVterm
7I
(8.92)
Jz
A cOs(8 - @ ) f ( p ) - -aVterm
7I
Id
cosa!'
x c + -3In d
= 0,
- ka
cos(a!
= 0,
- CY,ln = 0.
Although k can be calculated explicitly, a linearized form of Equation (8.60) obtained for 01 = 30" can be used to simplify the expression. In the range 60" < 1 < 120", the value of k can be obtained from:
f(1) = 1.01 - 0.0573 1,
(8.97)
where p is measured in radians. In Mode 4, the a s . and d.c. systems are both short circuited at the rectifier and operate independently. In this case, the system equivalent of Figure 8.34 reduces to that shown in Figure 8.35. The network equivalent can be solved directly and the d.c. current is obtained from the algebraic form of the differential Equation (8.57).
Dynamic
element
339
d.c. links The problems associated with dynamic rectifier loads do not occur when the d.c. link is represented by a quasi steady-state model. Each converter behaves in a manner similar to that of a converter for a static rectifier load. A nominal bus shunt admittance (Lo) is calculated from the initial load flow for both the rectifier and inverter ends and injected currents are used at each step in the solution to account for the change from steady state calculated from Equation (8.76). Note that the steadywill have a negative conductance value as state shunt admittance at the inverter (YOoi) power is being supplied to the network. This is not so for a synchronous or induction generator as the shunt admittance serves a different purpose in these cases.
Although electronically produced, the dead band may be considered as a physical linkage problem, as shown in Figure 8.38(a). In this example, the input ( x ) and output ( y ) move vertically. The diagram shows the initial steady-state condition in which x and y are equal. The input ( x ) may move in either direction by an amount Db/2 before y moves. Beyond this amount of travel, y follows x, lagging by &/2 as depicted in Figure 8.38(b). The effect of a dead band can be ignored by setting Db to zero.
340
I. a -
I I a 1
II
I
I I
t
v
I I I
3
Calculate mismatches Convergence Yes given mode of operation lter + 11ter+ll
I
I
,
I
I
I
i
I
I ( %
Calculate rectifier variables
I
I
I II I I
I ! (Convergence?)
lter > 15
t
A.C. current on [ V ]
I
I
Figure 8.36 Unified algorithm flow diagram
Stepped output permits the modelling of SVS when discrete capacitor (or inductor) blocks are switched in or out of the circuit. It is usual to assume that all blocks are of equal size. During the study, the SVS operates on the step nearest to the control setting. Iterative chattering can occur if the control system output (B3) is on the boundary between two steps. The simplest remedy is to prevent a step change until B3 has from the mean setting of the step. moved at least 0.55BSep
8.1 1
Initial controlling vohage ( p u ~
341
"9v-t
Dead band
Control system
1x r + z 4
(a) Movement
Movement
input x
\n
Figure 8.38 Dead band analogy and effect. (a) Physical analogy of dead band; (b) The effect of a dead band on output
The initial MVAR loading of the SVS should be included in the busbar loading schedule data input. However, it is possible for an SVS to contain both controllable and uncontrollable sections (e.g. variable reactor in parallel with fixed capacitors or vice versa). It is the total MVAR loading of the SVS which is, therefore, included in the busbar loading. Only the controllable part should be specified in the SVS model input and this is removed from the busbar loading leaving an uncontrollable MVAR load which is converted into a fixed susceptance associated with the network. Note that busbar load is assumed positive when flowing out of the network. The sign is, therefore, opposite to that for the SVS loading. In order to clarify this, consider an overall SVS operating in the steady state, as shown in Figure 8.39(a). The busbar loading in this case must be specified as -50 MVAR and it may be varied between -10 MVAR and -80 MVAR provided the voltage remains constant. The SVS may be specified in a variety of ways, some more obvious than others, the response of the system being identical. Three possible specifications are given in Table 8.4. In the first example, the network static load will be +20 MVAR, while in the second case the static load will be zero. The third example may be represented by an overall SVS, as shown in Figure 8.39(b).
342
12z
Initial loading
70 50 -30
80 MVAR
Figure 8.39 Example of an overall SVS controllable and uncontrollable sections. (a) Example of overall SVS using controllable capacitors; (b) Alternative to overall SVS in (a) using a controllable reactor
Table 8.4 Examples of MVAR loading specification for SVS shown in Figure 8.39(a)
Example
1
Maximum limit
100
Minimum limit
30
10
2
3
80
0
-70
It is convenient, when specifying the initial steady-state operation, to use MVAR. However, this is a function of the voltage and, hence, all MVAR settings must be converted to their equivalent per unit susceptance values prior to the start of the stability study.
8.12 REFERENCES
343
where
Y =0
+jB4.
Although not necessary for the solution process, the MVA output from the SVS into the system is given by:
(8.102)
8.12 References
1, Kulicke, B. Netomac digital program for simulating electromechanical and electromagnetic
transient phenomena in a x . systems, (198 1). Siemens Aktienngesel-lschaft, E15/1722- 101. 2. Woodford, D A, (1985). Validation of digital simulation of d.c. links, IEEE Transactions on Power Apparatus and Systems, PAS-104 (9), pp. 2588-2595. 3. Anderson, G W J, Arnold, C P, Watson, N R and Arrillaga, J, (1995). A new hybrid a.c. -d.c. transient stability program, International Conference on Power Systems Transients (IPST), pp. 535-540. 4. Anderson, G W J, (1995). Hybrid simulation of a.c.-d.c. power systems, Ph.D. thesis, University of Canterbury, New Zealand. 5 . Eguiluz, L I and Arrillaga, J, (1995). Comparison of power definitions in the presence of waveform distortion, International Journal of Electrical Engineering Education, 32 (2), pp. 141-153. 6. Heffernan, M D, Turner, K S, Arrillaga, J and Arnold, C P, (1981). Computation of a.c. -d.c. system disturbances -Parts I, I1 and 1 1 1 , Transactions on Power Apparatus and Systems, PAS-100 (1 l), pp. 4341 -4363. 7. Szechtman, M, Weiss, T and Thio, C V, (1991), First benchmark model for HVDC control studies, ELECTRA (133, pp. 55-75. 8. Clarke, E, (1950). Circuit Analysis o f a.c. Power Systems- Vol. 11, John Wiley and Sons, New York. 9. Kimbark, E W, (1968). Power System Stability- Vol. IIIs Synchronous Machines, Dover Publications, New York. 10. Reeve, J and Adapa, R, (1988). A new approach to dynamic analysis of a.c. networks incorporating detailed modelling of d.c. systems -Parts I and 11, IEEE Transactions, PD-3 (4), pp. 2005-2019. 1 I . Watson, N R, (1987). Frequency-dependent a.c. system equivalents for harmonic studies and transient converter simulation, Ph.D. thesis, University of Canterbury, New Zealand. 12. Arnold, C P, Turner, K S and Arrillaga, J, (1980). Modelling rectifier loads for a multimachine transient-stability programme, IEEE Transactions on Power Apparatus and Systems, PAS-99 (l), pp. 78-85. 13. Giesner, D B and Arrillaga, J, (1970). Operating modes of the three-phase bridge converter, International Journal of Electrical Engineering Education, 8, pp. 373-388. 14. IEEE Working Group on Dynamic Performance and Modeling of d.c. Systems, (1980). Hierarchical structure. 15. Turner, K S , (1980). Transient stability analysis of integrated a.c. and d.c. power systems, Ph.D. thesis, University of Canterbury, New Zealand. 16. CIGRE Working Group 31-01, (1977). Modelling of static shunt VAR systems for system analysis, ELECTRA (Sl), pp. 45-74.