Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

270

IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, VOL. PAS-90, NO. 1, JANUARY/FEBRUARY 1971

Digital Calculation of Impedance for


Transmission Line Protection
BARRY J. MANN,
MEMBER, IEEE, AND

I. F. MORRISON,

MEMBER, IEEE

Abstract-The paper proposes a method of distance-type protection suitable for on-line digital computer protection of transmission lines. The basic principle is the predictive calculation of peak fault current and voltage from a small number of sample values. The magnitudes of various sources of error are discussed. The results of experimental work on a model transmission line and on a highvoltage transmission system are described. The calculation times for implementation in real time are estimated.

THEORY
Basic Principles The method of calculation of line impedance involves the predictive calculation of peak current and peak voltage, the impedance being determined by division of peak voltage by the peak current. A digital computer sampling a sinusoidal waveform can determine the peak values as they occur. However it is necessary in the interests of time to determine the peak values before their occurrence; i.e., to predict the peak value of the waveform from any given sample. Consider
v=

INTRODUCTION

THE growing complexity of the control requirements on nmodern power systems is resulting in increasing attention being paid to the possibility of allocating the functions of highvoltage substations to on-line digital computers [1]-[3]. Such computers offer attractive compactness and flexibility. Of all the modern substation functions, the protection of transmission lilnes and substation equipment would appear to be the most demanding in terms of computer speed, cycle time, and peripheral equipment. Rockefeller [2], in the basic work in this field, proposes a complete group of programs for protection by digital computer. This group contains programs for the protection of transmission lines as well as transformers and bus protection and also includes a valuable account of executive routines for detection of faults and the organization of subsequent protection operations. The technique of transmission line protection proposed by Rockefeller is essentially an adaptation to digital computer of earlier analog zone packaged line-drop compensation relaying. The method of transmission line protection intended for digital computer application proposed here attempts to start from basic l)rinciples. A small number of current and voltage samples are used to estimate numerically the peak value of current and voltage. From these the transmission line impedance can be calculated anld fault conditions detected and measured by the drop in impedance. The inherent numerical errors are assessed and the method is applied to fault data obtained both from a model line and from a high-voltage transmission system. An estimate is then made of calculation time on a typical processcontrol computer of the general type that could be used in eventual applications.
Paper 70 TP 165-PWR, recommended and approved by the Power System Relaying Committee of the IEEE Power Group for presentation at the IEEE Winter Power Meeting, New York, N. Y., January 25-30, 1970. Manuscript submitted September 8, 1969; made available for printing December 5, 1969. This work was supported by the Electrical Research Board and the Electricity Commission of New South Wales, Australia. The authors are with the School of Electrical Engineering, University of New South Wales, Kensington, N. S. W., Australia.

Vpk sin cot

(1)

where Vpk is the unknown quantity and v is a typical sample value. Since it is not proposed to synchronize the sampling to the sinusoid, the value of sin cot is also unknown. Taking the derivative of (1)

U' = &)Vpk COS cot

(2)

and assuming that v' can be determined, the peak value of the sinusoid is determined from

Vpk2

V2 +

(3)

If (1)-(3) are applied to transmission line current and voltage, the modulus of the line impedance can be determined. Furthermore, the phase difference between the voltage and current waveforms can be determined from 0
=

arctan

(i

arctan (,)

(4)

enabling complete impedance determination (see Appendix II for derivation).


Power System Application The possible existence of an exponentially decaying dc transient on the current and voltage signals of a high-voltage system causes some difficulties in application of the above principles. Iron-cored current transformers with resistive burdens transmit the full amount of offset in the primary fault current. This component often takes more than 5 cycles to decay on the present systems with high X/R ratios. The voltage can also have an offset component, but this is considerably smaller than that on the current waveform. The relevant expressions are well known (see [4], [5], for example) and are repeated here for

MANN AND MORRISON: DIGITAL CALCULATION OF IMPEDANCE

271

convenience:

This yields an estimate VO of V at time to, from (3)


a

ii

Ilpk[5Sin (cat +

- eRlt/L1 sin e)
-

(a

4)1)]

(5)
(6)

vl= Vlpk [sin (ct +

a - 41

+ 4L)

Vo2

v02 + (At |

)-

Vpk2

1-

sin2 Coto

- * *

sin (a

4)) sin (4)


sin 41

4L)

e-Rlt/Ll].

hence

In these equations, a is the phase of fault incidence, the subscript 1 refers to primary (system) quantities, and L refers to line quantities. The dc offset in the current signal can be removed using mimic impedances in the current transformer (CT) secondary. If an ideal CT is connected to a secondary burden having the same X/R ratio as the primary circuit, then the voltage across the burden will be purely sinusoidal. Applying (5) to a CT of secondary burden R2 and L2
v V =

V0

Vpk

(t) sin2 ,to

. .

.)

(12)

(iR2 + L2dt)
a

(7)

N2
N,

11Z2 sin (ct +


+ X2

-1

+ 02)

z2

1R _ R2) X2/ e,-Rit/Li sin (aXxI

4)1)]

(8)

where subscript 2 refers to CT secondary. For more detailed discussion see [6]. Note that the coefficient of the exponential term is zero if
X1

X2

R,

R2

(9)

Exact cancellation of the de offset is not possible for all faults. The primary X/R to be matched is that of source plus line to the fault point, and since in general the source X/R is not equal to the line X/R, the overall primary X/IR is a variable quantity, dependent on how far along the line the fault occurs. This problem may be avoided by matching the secondary burden to the primary X/R composed of the source plus, say, 90 percent of the line impedance. It is for faults near the end of the line (zone 2) that the most accurate impedance calculations are required for discrimination purposes, and for these faults the transient component will be almost completely removed. For faults closer in, offset will be drastically reduced but not entirely removed. WTith this provision for the removal of offset dc components, the method proposed here involves the use of a digital computer to calculate numerically the impedance of a faulted transmission line from the predicted peak values of fault current and voltage.

Error Theory Since it is proposed to use numerical differentiation, it is necessary to check the errors involved. Assume that the waveforms are sampled at an interval of At, with actual sampling times being , , tklb, , and corresponding sampled values being , vk), vk1, . Take to as being halfway between tk and tk+l. Then
vO
v0
=

For a At of 0.5 ms, (12) gives a maximum error in Vo on a 50-Hz system of 0.15 percent, which shows that numerical errors are at least small in theory. Numerical analysts list several ways of calculating derivatives e.g., [7]-[9]. In general these methods are series expressions of forward, backward, or central differences. The actual formulas for numerical differentiation can be obtained by differentiation of the appropriate interpolation formula. Isaacson and Keller [9] show that interpolation errors are least near the center of the interval of interpolation and that an analogous result is true for numerical differentiation. Thus in theory it seems advisable to use formulas of the central difference type. However, for realtime implementation it may be advantageous to use backward differences because existing samples could be used for immediate derivative calculation; see Appendix III for details. Sampling rate is another variable. Presence of noise will mitigate against high sampling rates, as will the accumulation of too much data for an already overworked computer to handle. On the other hand, the need for rapid fault detection implies a reasonably high rate. Other effects, such as CT saturation, may be of importance; if a CT can saturate and if the fault detection process requires n samples, then these n samples would have to be taken before saturation occurs. At this stage, however, it was decided to select the sampling interval on the basis of truncation and round-off errors incurred in the derivative estimation. Ralston [7] shows that the roundoff error is proportional to the maximum round-off and inversely proportional to the interval of differentiation. The truncation error, on the other hand, is proportional to a power of At depending upon the number of points used in the differentiation formula (see Appendix III). A suggestion for a choice of interval is to make the upper bounds of these two errors equal. In the present case we find that for a three-point formula an interval of 0.5 ms is near optimum and the upper bounds of those errors are in the vicinity of 6 X 10-3. For a five-point formula using a 0.5-ms interval it was found that the sampled quantities would have to have at least 10 significant bits to avoid round-off errors becomlng excessive. These considerations led to the tentative selection of a sampling interval of 0.5 ms, i.e., 40 samples per cycle on a 50-Hz system. In order to estimate the errors arising from the numerical approximation described above, it was decided to apply the method to theoretical data, i.e., data free from system or instrumentation noise. From (5) and (6) the normalized expressions required to generate the secondary current and voltage signals can be shown to be
i-sin (wt + _ , + $,s)8+ a -)+ i=sin(wt
= sin

_sin (a

4) sin () - Rs) et/L e


sin

(Vk + Vk+l)
1

(10)
v

(13)
(ct + a-4 + 4L) sin (a-4) sin ()-4L) e -Rt/L (14)
sin
4)

-A-

At

k+l

V70.

(11)

272

IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, JANUARY/FEBRUARY 1971

Values of these expressions at 0.5-ms intervals over a 50-ms period were generated off-line. Constants occurring in these equations were chosen to coincide with measured values from the model transmission line mentioned in later sections (X/R = 10, XL/IRL= 7). Data were generated for maximum, half, and zero offset for the cases of matched and mismatched (X2/R2 = 6) CTs. The program to analyze these data computes i', V', Ipk and Vpk, and Z for each point. The mean value of Ipk, Vpk, and Z is calculated for the 100 points and the "variance"2 about the mean noted also. Since the quantities involved are not random variables, this is not a true variance in the statistical sense, but it does assist in the overall estimation of errors. The data generated with no offset had maximum errors for Z of less than 0.3 percent from the mean, verifying the accuracies predicted. The data sets with some offset remaining which, although they provided maximum errors of 10 percent in Vpk, lpk, calculated Z to within 4 1.5 percent. This result arises from the fact that the current and voltage signals over the matched CT secondaries are almost in phase and the errors in Ipk, Vpk partially cancel. On this theoretically generated data, no useful distinction could be made between use of backward and central differences. Inclusion of the second terms in the series resulted in marginally smaller errors. Similar observations apply to the phase angle determination of (4). Data generated for the case of zero offset measured the angle to accuracies of the order of 0.10. Effects due to the current and voltage signal being approximately in phase were observed here also, maximum errors for individual voltage or current phase being around 30, whereas the maximum phase error for impedance is less than or equal to 10.

CONTROL UNIT
STOP START

C.T.

SOURCE
50Hz

LINE
0|V.T.

LOAD

Fig. 1. Model transmission line.

The A/D converter used was the DEC-138 [10], [11]. The earlier consideration of the accuracy indicated a 10-bit digital output as being desirable, and this matched the accuracy available at the analog input. In the absence of a sample and hold unit, rate of change errors in the converter were encountered. It was not found possible to obtain accuracy to 10 bits at the maximum rates of change; even with the fastest setting of the switching point selector, only 8 bits are significant near the zero crossings. Experimentally it was found that the A/D converter introduced maximum errors of A 2.5 percent in the calculated peak values with a variance, as previously defined, of 1.2 percent for 100 samples. Thus, with the equipment used, errors of -4 percent in peak voltage and current values were anticipated from sources other than noise on the input signal.

EXPERIMENTAL RESULTS A lumped-element model transmission line was used to acquire data simulating a faulted line (Fig. 1). Faults of any nature may be applied by appropriate connections to the secondary of the breaker shown. The load was selected to give a ratio of fault to prefault current of about 20:1. The current and voltage signals from a sequence of faults of different phase of incidence were recorded on a high-performance magnetic tape recorder. The data was then played back into the (15) Vn-I = 6 (5vn-1 + 2Vn - Vn+l) 6 analog-digital (A/D) converter of a Digital Equipment Corporation (DEC) PDP-8 computer. This small general-purpose (16) Vn = (Vn-1 + Vn + Vn+l) computer sampled the current and voltage waveforms at 0.5-ms 3 intervals for 2 cycles before and 5 cycles after each fault and output these data onto paper tape. The data could then be (17) Vni = -1 (- Vn-l. + 2vn + 5Vn+i). processed off-line by an IBM 360/50 using the programs de6 veloped in earlier sections. Ralston recommends the central expression be used wherever The primary focus of the work described was to establish the The first and third expressions use only forward and backpossible. possibility of using the method for the protection of transmission lines. Much attention was directed in the experimental work to ward points, respectively, and would thus be suitable for the the estimation of the error contribution from various sources, first or last points in the record. In our case, if the central expression (the second) has only with a view to isolating the errors inherent in the method and better smoothing effects than the backward (the marginally those that can be anticipated in the power system context. Error sources requiring particular attention were the A/D third), it may be advisable to use the backward expression, converter, the tape deck, and the computer interface. These since after any given sample is taken a smoothed value can be were evaluated by selectively omitting them from the experi- generated with data already in hand. mental equipment, and by using a known 50-Hz sine wave the variation of the error in the estimates of peak values was noted. Results from Model Transmission Line First, it was found that the tape deck and interface equipment Data from the model line were converted to a form usable by contributed less than 1 percent to the maximum errors, this being the IBM 360/50 as described previously. Basically, two types of consistent with the standard of equipment used. program were run. The first processed the 100 samples over the
1

Digital Smoothing To avoid excessive error due to this noise it was decided to investigate the use of digital smoothing of the sampled data. To select the degree of approximation, it was noted that maximum errors seem more likely to occur on the straight segment of the sine wave near the zero crossings, because here the differential term is the laiger contribution to (1), the expression for Vpk. A straight-line or first-degree approximation is thus selected. Furthermore, no more than three-point smoothing seems advisable since the process would incorporate too large a time delay into the overall calculations. With these requirements, least squares formulas [7] give the following expressions for digitally smoothed sample values:

MANN AND MORRISON: DIGITAL CALCULATION OF IMPEDANCE

273

1) 2) 3) 4) 5) 6) 7)

Test Runs Central difference-first term only Central differencesecond term Backward difference-second term Repeat of runs 1-3 with digital smoothing Repeat of run 2 with end point smoothing

Variance (percent)

TABLE I POSTFAULT RESULTS, MODEL LINE I V Maximum Maximum Errors Variance Errors about Mean about Mean (percent) +6.4, -12.9
4.4

z
Variance (percent) 6.0 7.6
6.1 3.7 4.0 4.0 Errors about Mean

Maximum

3.6
4.0 6.8 2.9 3.0 6.5

+11.5, -9.8

+19.5, -10.6

4.0

+7.6, -14.3 +14.5, -13.2 +3.8, -9.6 +4.1, -10.2 +12.1, -10.0 +7.8, -13.9

5.7
6.8 2.3 2.6 5.7
-

5.5

+15.8, -12.5 +15.8, -12.9 +5.7, -4.9 +6.5, -5.6 +13.6, -7.9 +14.1, -13.1

7.3

+24.6, -15.0 +21.6, -10.0 +12.9, -5.9 +14.4, -7.0 +14.3, -5.6 +26. 3, -13.0

PU

&,

-1
-A-A--X-- X--

-0-D-

-0-0- impedance smoothing Fig. 2. Fault A model line. Dotted sections of curves represent transition from prefault to postfault conditions; effect of digital smoothing readily seen from results. Note that reliable impedance calculation (i.e., within 4+ 10 percent) is possible for sixth postfault sample, i.e., 3 ms after fault.

voltage current impedance

-A-A-- X - X- -

-f-f-

voltage
current

-0-0- impedance smoothing Fig. 3. Fault B model line.

impedance

I'

pu
1
.5

V &
V
&
A

III
I

kts
2

\: i \ -

6
MS

FAULT

AFTER

-A-A-

_z_z_
-x-

--x

-0-0-

voltage impedance impedance smoothing


current

-fl-fl-

voltage
current

-A-AX
x
-0-0--

impedance impedance smoothing

Fig. 4. Fault C model line.

Fig.

5.

Fault D model line.

274

IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, JANUARY/FEBRUARY 1971

TABLE II POSTFAULT RESULTS,* ECNSW DATA


I

V
Maximum Errors about Mean +6.3, -7.5 +4.1, -5.0

Maximum Variance Errors Variance (percent) about Mean (percent) 5.3 10.8, -13.5 3.0 3.9 9.1, -7.5 1.9 * Results on second line reflect digital smoothing.

Variance (percent)
5.6 4.2

Maximum Errors about Mean 13.1, -13.7

8.8, -12.2

LLi

0
I

00

-O---0- current Fig. 6. Fault B model line, phase plot.

-X-X-

voltage

-A-A-- - X - -O--O--

0-EJ-

voltage current

impedance impedance smoothing

Fig. 7. Fault E, 132-kV ECNSW grid.

fault point to allow study of transient behavior. The second modulus. To observe the behavior of the angle measurements considered 100 samples in the postfault steady-state region to over the fault point see Fig. 6. allow more ready statistical treatment. The disturbances in voltage phase and current phase settle The tabulation of the results for one typical fault illustrates down by the sixth postfault sample also. This is representative the differences between the various numerical differentiations of all the results. Excluding the samples immediately around the and in the effects of digital smoothing (Table I). fault point, the maximum errors from the line of best fit were The different differentiation techniques do not give radically t50 for voltage and 70 for current, this being true for all different errors. The central difference expression with the runs. It is thus apparent that the angle of the impedance would second term is slightly worse than the expression with only the be accurate to within 12. first term included. The backward difference expression yields It should be noted that, in Figs. 2-6, the current i has been accuracies similar to the central difference expressions in the shifted in phase by 1800 for greater clarity. case of impedance values but not in the case of peak current Results from High-Voltage Grid and voltage, which are significantly worse. The effects of digital smoothing are quite marked. Comparing Several recordings were taken of single phase-to-ground runs 2, 5, and 7, it can be seen that the central expression for faults on the Electricity Commission of New South Wales smoothing improves accuracies considerably, whereas the end (ECNSW) grid. These faults were at 132 kV at the end of a 23point expression makes only a marginal difference. mile line that was open circuit prior to the fault. Overall maximum errors in Z from 800 calculations using a Representative results taken from one of these faults are series of 8 faults were +16 percent, -10 percent, with variance shown in Table II, which relates more than 100 samples to the of these calculations about 3.5 percent about the mean. Maxi- postfault steady state. mum errors in Vpk and lpk were always less than 10 percent of The magnitudes of errors are similar to those for data from the the mean. model line. Results from the transient study are shown in Figs. 2-5. The graphical plot (Fig. 7) of v, i, and Z over the fault is also The horizontal scale for each is time, and v, i, and Z are shown; shown. Z appears both with and without digital smoothing. Each of the No special provision was made while recording to match the figures represents a fault of different phase. CT secondary; thus there is some offset present in the current Results for the angle determination are similar to those for the signal. Since the X/R ratio of the overall source plus line at the

MANN AND MORRISON: DIGITAL CALCULATION OF IMPEDANCE

275

relaying point was about 5, this offset decays reasonably quickly. The offset is apparent on both the Z plot and the plot of current angle. The oscillatory noise superimposed on the voltage and current imposes quite a severe test on the method. The use of digital smoothing was absolutely necessary for these faults. In general the similarity between these results and those of the model transmission line confirmed the validity of the method and demonstrated its applicability in a power system environment.

LLJ
LL

C]
I)

Calculation Times The calculations described in previous sections were all conducted off-line. Since the method is intended for on-line implementation, calculation times for a small process computer have been established. The computer used was a DEC PDP8. This is a small generalpurpose computer with a 12-bit two's complement arithmetic. It has a 4K random address magnetic core with a cycle time of 1.5 us, and hardware multiply divide facilities [10]. The expressions for impedance and phase were calculated on the PDP8 and the calculations timed. Digital smoothing was not included since before any particular sample is stored it is envisaged that smoothing will be performed and the smoothed value stored. Time of calculation of the derivative was not included since this would probably be performed as the sampled values become available. The function calculated was
= a2 + b2 (18) c2 + d2 Since squaring removes all sign problems, unsigned multiplication and division routines could be used. Total time was 260 As. The expression for angle is
c a 0 = arctan b - aretan -d d

0r
00
0

AMS AFTER FAULT

-X-X-

Fig. 8. Fault E, 132-kV data, phase plot.

-OC{--O-

voltage
current

(19)

The calculation of arctangents is time consuming by exact means. To overcome this a program using a coarse table lookup was used-this divided the range of 0 to 450 into 32 intervals. Final calculation of arctan a/b was accurate to 10 and took 240 ,us. Thus overall calculation times for 4 were 480 ,s.
CONCLUSIONS

Second, a prior knowledge of the existence of a fault is not required since the sharp decrease in line impedance is in itself an indication of the presence of a fault. From Figs. 2-8 it can be seen that impedance and angle values of reduced accuracy are available in times considerably less than 5 ms, and these could well be used for starting more refined measurement techniques. A further possible fault detection application would be the continuous computation of Vpk at the various bus voltage levels in the substation, a sudden drop in voltage indicating the occurrence of a fault within one of the protection zones. Third, the experimental work described indicates the need for careful choice of A/D conversion equipment for the digital computer to be used in substation applications. The unit used in this case was not ideal either in accuracy or speed. Improved performance in these regards promise a further improvement in the overall accuracy of impedance calculation.
APPENDIX I DEFINITION OF TERMS

From the foregoing it can be seen that, for a faulted singlephase line, the proposed technique calculates the modulus of the impedance to within J 10 percent and the argument to within 150. Measurements to at least this accuracy should be available 5 ms after fault occurrence. Furthermore, it seems that the method is readily adaptable to the small, process-control type of computer and that the calculations can be performed in acceptably small times. The logic of extension of calculations to a three-phase line and to groups of these lines seems straightforward and is under active investigation at present. Several other comments arise from the work described. First, it is assumed that conventional iron-cored voltage and current transformers are the prime sources of fault data. Such data will be of acceptable accuracy since the A/D conversion devices place low voltampere requirements on the protective transformers, and common problems of saturation can be avoided. The accuracy of the method could only benefit from the use of any new type of transducer.

v, i

instantaneous phase voltage and current, respectively value of v, sampled at time tk Vk Vpk, lpk peak values of phase voltage and current, respectively v', it instantaneous values of first derivatives of phase voltage and current, respectively estimate of Vpk at time to Vo phase difference between voltage and current; i.e., impedance angle, tan-' wL/R where Z = R + jcoL phase of incidence of fault as measured from a1! positive-going zero crossing of the source voltage N2/N1 CT ratio

Vyk = (Yk Yk-1) bYk = (Yk+1/2 - Yk-112)


-

AYk

+ 2 (Yk+112 Yk-112).

276

IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, JANUARY/FEBRUARY 1971

Subscripts are best explained from Fig. 1: L quantities relating to line from relaying point to fault 1 CT primary quantities relating to overall source plus line 2 quantities measured in CT secondary circuit.

Using the first term only

hy k' = Yk - Yk-I
and with a second term

(28)
(29)

Unsubscripted quantities denote general waveforms, e.g., (1) refers to a sinusoidal voltage signal in general.
APPENDIX II DERIVATION OF (3) AND (4) To derive (3) from (1) and (2), first rewrite (2) as
co

hy7c

1(Y

Yk-2) -

Note that (25) and (26) for ye' involve sample values at times later than tk, whereas (28) and (29) do not. Thus both central and backward differences were considered as possible ways of performing the differentiation. ACKNOWLEDGMENT The authors wish to acknowledge the valuable collaboration of their colleagues, G. C. Dewsnap, A. D. Mclnnes, and G. H. Couch.
REFERENCES'
[1] I. F. Morrison, "Prospects of on-line computer control in transmission systems and substations," Elec. Eng. Trans., Inst. Eng., Australia., vol. EE3, pp. 234-236, September 1967. [2] G. D. Rockefeller, "Fault protection with a digital computer," IEEE Trans. Power App. Syst., vol. PAS-88, pp. 438-464, April 1969. [31 B. J. Mann, "Real time computer calculation of the impedance of a faulted single phase line," Elec. Eng. Trans., Inst. Eng., Australia, vol. EE4, pp. 26-28, March 1969. [4] P. Mathews, Protective Current Transformers and Circuits. London: Chapman and Hall, 19.55, ch. 6, 7. [51 P. Mathews and B. D. Nellist, "Transients in distance protection," Proc. Inst. Elec. Eng. (London), vol. 110, pp. 407-418, February 1963. [61 A. Wright, Current Transformers: Their Transient and SteadyState Performance. London: Chapman and Hall, 1968. [7] A. Ralston, A First Course in Numerical Analysis. New York:

Vpk cos cot.

(20)

Squaring and adding (1) and (20) gives

Vpk = Vpk2 (sin2 cot + cos2 cot)

v2 + (5)

which is (3). Equation (1) divided by (20) is co =Var tV- Vpk sin cot Vpk cos cot i.e., cot(-arytan , VI Vt Vpk COS COt'
If the current waveform is represented as

(21)

Ipk sin (cot+4)

(22)

then in a similar fashion

(cot + ') = arctan (V)


and (4) results from subtracting (21) from (23), i.e.,
arctan (
.

(23)
= arctan

Equations. London: Pergamon; Reading, Mass.: AddisonWesley, 1962. [9] E. Isaacson and H. B. Keller, Analysis of NVumerical Methods. New York: Wiley, 1966. [10] Digital Equipment Corporation, Small Computer Handbook, 1966-1967.

McGraw-Hill, 1965. [8] L. Fox, Numerical Solution of Ordinary and Partial Differential

[11]

, Logic Handbook, 1968 ed.

APPENDIX III

DIFFERENTIATION FORMULAS

Using the standard notations v, 6, and , for the operations of backward differencilng, central differencing, and averaging, respectively, the basic central difference expression for the derivative [8] is

Discussion
G. Hope, D. Bell, and S. Jura (University of Calgary, Calgary, Alta., Canada): The authors present a very valuable paper in the field of digital protection of transmission lines. Many universities and companies are now working very hard in this area. There were very few people five years ago who believed that it was possible to realize a digital protection system by using a small digital computer. The opinions are changing very slowly, but we can say that the number of believers is increasing, and this excellent paper shows further the possibility of proving the use of digital methods for power system protection. It is necessary to detect a fault very fast, and it seems to us that the method described in the paper is very suitable for this purpose because it makes possible the calculation of the waveform before it occurs. We are not sure, however, if the achieved accuracy is sufficient.

hYk' =(

-$

46 +

)Y Yk3-

(24)

where h = At and y stands for v or i. Using the first term only

hYk
and with a second term

I=

1 2

(Y k+I
e

Y k-1)

(25)

= -3A (y 2 hy1'

ye-i 1j YX-2}

(26)
(27)

For backward differences

hyk'=(V -

Y2)

Manuscript received February 12, 1970.

MANN AND MORRISON: DIGITAL CALCULATION OF IMPEDANCE

277
REFERENCES

The use of a coarse lookup table is a good solution for shortening of a calculation time. Nevertheless, for a large network, the capabilities of a small digital computer probably will not be sufficient. The use of hardware implementing instruction may be a great help. Comments on speed and core requirements necessary to protect several lines would be very useful. The model used has some limitations. When considering voltage levels of 735 and 500 kV, which constitute possibly the most important application of digital protection, high-frequency components due to line capacitance are introduced. We have doubts that the model of a transmission line without capacitance is sufficient. It is possible that a model change would change the results considerably. It is unfortunate that there are limitations to the A/D converter used. The effects of digital filtering are very strong. Is the method feasible when many lines are to be considered? It may be noted that some of the difficulty encountered is due to the desire to obtain the derivative. However, the derivative is used in (3) to obtain the peak value. If (2) is rewritten as e = fv dt then (3) becomes V,k2 = V2 + (,we)2. Would this not lead to some improvement? It is generally dangerous to try to obtain a derivative, and the filter used is to correct errors due to the equipment used. Will the system be able to discriminate for system swing and transient inrush due to transformers, etc.? Comments on the use of general-purpose computer with readonly storage would also be useful. The read-only method is wellsuited for the table lookup procedure for calculating arctangents. The authors' comments would be appreciated.

[12] M4. Ramamoorty, "High-speed relaying based on improved power system models," Ph.D. dissertation, Dep. Elec. Eng., University of Toronto, Ont., Canada, 1967.

M. Ramamoorty (Indian Institute of Technology, Kanpur, India): Having done some work in the area of application of digital computers for system protection, I intend to put forth the following observations on the paper. The authors have assumed sinusoidal waveforms in deriving the basic equations (3) and (4) for the calculation of impedance of the faulted line. Later on, regarding [12], the error involved in the calculation of the gradient using the central differences is discussed. However, if a computer is to be used for the purpose of protection, the concerned line is generally a long tie line requiring a highly sophisticated protection system to avoid any unnecessary opening of the line. For such cases the line capacitance cannot be neglected and in [13], which is not cited by the authors, the short-circuit current waveforms are shown to be highly distorted with traveling wave fronts. Fig. 9 shows a typical fault current waveform obtained from [13] and also shows high distortions in the first few cycles after the disturbance. Evidently if (3) and (4) are applied to this type of waveform the error will be much more than predicted. The application of a dc offset filter as mentioned in (9) will not be useful under these conditions.

Taylor (Bonneville Power Administration, Portland, Oreg.): The authors have made a valuable contribution to the numerical analysis aspect of digital computer fault protection. The protection engineer of the future will certainly have to be knowledgeable in the fields of computer programming, numerical analysis, and digital logic. Since a computer relaying system would probably only be justified in a multiterminal station, the first question that arises concerns the computer requirements and speed obtainable when perhaps 100 or more ac current and voltage quantities require mnonitoring. The authors' conclusion that a less accurate or a nondirectional "fault detecting" or "starting" calculation may be necessary is well taken. Ramamoorty [12] has suggested two methods of nondirectional fault detection to be followed by an impedance calculation utilizing 18 samples per cycle (60-Hz basis) rather than the 40 samples per cycle (50-Hz basis), which the authors suggested. Ain advantage of a digital computer impedance calculation is that the protection engineer can program any type of operating characteristic (impedance, mho, reactance, or others not in common use today). One undesirable characteristic of electromechanical or static distance relays is that their operating time increases for faults near their "reach" setting. Do the authors envision more accurate and hence slower impedance calculations for faults near its "reach" setting? The author's conclusion that a digital impedance calculation would benefit by nonconventional current transducers is true also for commercial static relays. Current transducers become more attractive at EHV and UHV and the electric utility industry should be encouraged to develop practical devices. We at BPA are very interested in digital computer fault protection not only to decrease fault clearing time, but also because of the advantages a digital computer in a multiterminal station offers to
Carson W.

0-

Fu
D

-20

Fig.

9. Sending-end current during fault on phase a (300 km, 5O).

1.6

1.2

Sudden load chan e

0.8

0.77 PU.D
V--

- _

Single phase fault


0.4

initiate such functions as breaker failure protection, transient blocking, out-of-step blocking, out-of-step tripping, generator dropping, load dropping, and series-capacitor switching. The functions are presently used at various places on the BPA system to maintain system stability.

0.29 Per Unit

_
,

_
TimeSeconds

3-Phase andZ-phas foult


0 0 0.00273
0.00834

0.0139 0.01668 Fig. 10. Time variation of impedance.

Manuscript received February 12, 1970.

Manuscript received February 24,

1970.

278

IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, JANUARY/FEBRUARY 1971

To smooth out the variations in the measured impedance, the authors have used 100 points at 0.5-ms intervals, which corresponds to a 2-cycle relay operation, which apparently is much slower than the protection schemes using transistorized relays. In [13] the impedance is evaluated by obtaining the fundamental amplitude from the sampled values of the voltage and current waveforms. Here the samples are taken at 180 apart corresponding to 20 points per cycle. Continuous evaluation of the impedance is made by the computer by adding one extra sample and deleting the first sample each time. Fig. 10 shows a typical plot of impedance variations for different types of faults [13]. This method of calculation is very useful because the trend of the impedance variation is also known, and for nearby faults the measured impedance falls within the relay setting in less than 1 cycle, thereby providing a rapid fault clearance. The calculation of the fundamental quantity is done as follows:

V/a92+ b,2 sin (cot + tan-l' a, =(t) \

xl a =- [fo +2fi cosxl + 2f2cos2x' + *+f0] 2

bL

xi

- [2 fi sin xl + 2 f2 sin 2 xl + 2

+ 2

f0-, sin (n- 1)xl]


(30)

f *, where x1 is the sample interval in radians, n = 27r/x', and fo,fi, fn, are the sampled values of the input. In the analysis carried out in [11], it is shown that reasonable accuracy is obtainied for all types of fault waveforms with n = 20. Thus (30) gives both the amplitude and phase angle of the fundamental from which the impedance is calculated. Since the values of cos x', * *, cos n xl and sin xl, * * *, sin (n - 1)xl, are constants for each set of samples, they can be precalculated and used as weighting factors for each of the samples obtained. Thus the total niumber of computations involved in this method will be less than that indicated in the paper, where averaging has to be done using 100 points. Different types of logics that can be performed on the computed impedance to obtain the quadrilateral tripping characteristic for the digital distance relay are also indicated in [13]. The impedance computed from the 20 sampled values for each cycle using (30) is found to be generally unaffected by the dc offset and other high-frequency components present in the transient waveforms. REFERENCES [13] M. Ramamoorty et al., "High-speed relaying based on improved power system models," CIGRE, Rep. 31-09, 1968. [14] M. Ramamoorty, V. R. Sastry, and R. Raghanan, "Restriking voltage in circuit breakers due to single and multipole switching in EHV systems," Paper 70 CP 596, presented at the IEEE Summer Power Meeting and EHV Conference, Los Angeles, Calif., July 12-17, 1970.

velop techniques requiring simple calculations, on the basis that as much main-frame time as possible should be available for other protection (or other substation) activities. The work described has been extended to a single three-phase line and then to a number of lines at the one bus-voltage level within the substation [15]. From this work emerges an assessment of desirable computer specifications and capabilities, and we would direct the discussers who raised these general points to that source. Broadly, for a substation of 12 lines and two voltage levels, i.e., 42 ac quantities, a small general-purpose computer with a 16-bit word, approximately 1-ps cycle time, minimum 8K store, direct memory access facility, and appropriate A/D conversion will be required. The use of a larger machine will conceivably allow greater accuracy to be achieved and should allow a range of other substation functions (some enumerated by Mr. Taylor) to be performed; but such a computer would cost more. Several discussers query the omission of capacitance from the model transmission line. We are aware of this omission and the need to evaluate the effects on our techniques. Although, we would intend these techniques to be suitable for forthcoming 500- and 750-kV levels, we also intend them to be suitable for the present voltage levels of 132 and 330 kV. At these voltages, conditions in Australia are such that the effects of distributed line capacitance are not usually significant. Arising from the question of the inclusion of capacitance are the effects of various types of noise on the input signals. We feel that the possible source and frequencies of noise in the relaying signals have not been adequately enumerated. Inaccuracies of impedance measurement introduced by the presence of noise need to be evaluated. At present we anticipate that analog or digital filtering of the input signals will be helpful, even at the expense of an additional calculation time in particularly difficult situations. Specifically we expect that the input bandwidth could usefully have its 3-dB point at about 200 Hz. The discussers also raise the question of the effects of transients other than those due to line capacitance and noise, such as system swing, sudden load changes, etc. It is intended that the method of fault detection [151 will discriminate against those arLd that the impedance calculation stage will not be reached for nonfault conditions. In reply to Mr. Hope, Mr. Bell, and Mr. Jura on the question of accuracy, we have mentioned above that this is determined by the need to complete the calculation in less than 10 ms. Reference [15] shows that for 1000 faults of all types, impedance magnitude is calculated with a variance of 4.3 percent, with 98.5 percent of values within 10 percent of the mean; the variance for phase is 2.90, with 99.1 percent of values with 70 of the mean. We have considered the possibility of obtaining greater accuracy by the use of integration rather than differentiation. It can be shown that for the sampling interval used (0.5 ms) and two approaches are identical. Using the notation of (10)-(12), if we write v = v sin cot, then

u=

Jk-at
co

vudt

V = --[cos cotk (1 - cos wAt) - sin cotk sin coAt].


Barry J. Mann and I. F. Morrison: The authors wish to thank the discussers for their contributions; they raise numerous interesting points. We are becoming aware that an increasing amount of work is being done on the proposed use of digital computers for protection purposes. Like Mr. Hope, Mr. Bell and Mr. Jura, we have found that attitudes seem to have changed in four or so years from outright rejection to grudging acceptance that digital computers may contribute to the protection field. In the development of the technique of impedance measurement we have been conscious of a conflict between the requirements of speed and accuracy. Broadly, we have sought a scheme that will give an impedance to within 10 percent in times comparable to those of the best available static relays, i.e., less than 1 half-cycle. We further sought to minimize the amount of special-purpose hardware required to achieve these aims; and we have attempted to deManuscript received March 30, 1970.

Thus
V2 = Vk For At = 0.5 ms
(Vk Sin coAt - U cos coAt t-12 cos oAt - l

(31)

sin At

cotAt (to at least four decimal places)


cos c.At
1
-

(cAt)2

and the integral u is approximated by u = vo At. Then (31) becomes V2 = Vk2 + (AV/cAt)2, which is the expression of (12). We agree with the remarks of Mr. Taylor on the potentially flexible choice of relay operating characteristics and on the need for a more accurate and slower calculation near the relay balance point. Reference [16] describes a further method of impedance calculation which, it is hoped, will provide greater accuracy of calcula-

IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, VOL.

PAS-90,

NO.

1, JANUARY/FEBRUARY 1971

279

tion in slightly longer times. The question of the range of available characteristics is also raised there. In general the authors are greatly encouraged by the interest shown by an organization of the size and technical preeminence of Bonneville Power in the use of digital computers for protection and other substation functions. The problem of distorted input signals has been mentioned earlier and those remarks apply to the waveform supplied by Dr. Ramamoorty. We have not had data of this general type available for our experiments to date. We are particularly indebted to Dr. PRamamoorty for bringing his own work to our attention. We had not been aware of it and look forward to familiarizing ourselves with it in some detail. The present remarks will, of necessity, be our initial impressions. The method he describes, and especially the interesting results of Fig. 10, show that not only is a single cycle of sample values required for calculations, but that a decisive change in computed impedance is not available for times in excess of 1 half-cycle. We contrast here the results of Fig. 2, which shows a stable value of impedance available 4 ms after fault occurrence. In general we expect that the ac-

curacy of impedance calculation will be improved if data can be taken over longer time intervals. Here there seems to have been some slight misunderstanding arising from our paper-although 100 postfault samples were typically processed, this was simply to study the behavior of the method in time. Only six postfault samples are required for calculation of impedance with the accuracies described; i.e., only information collected in 3 ms after fault occurrence is used in calculation. This is to be contrasted with Dr. Ramamoorty's procedure, which utilizes information over 1 complete cycle.
REFERENCES [15] B. J. Mann and I. F. Morrison, "Relaying a three-phase transmission line with a digital computer," IEEE Trans. Power App. Syst. (to be published). [16] A. D. McInnes and I. F. Morrison, "Real time calculation of resistance and reactance for transmission line protection by digital computer," paper p 1169, Elec. Mech. Eng. Trans., Inst. Eng., Australia (to be published).

Equivalent

Circuits for Single-Phase Hysteresis Motors


DENIS O'KELLY

Induction

and

Abstract-General steady-state equivalent circuits are derived for the family of single-phase motors with unequal contrarotating magnetic fields produced by one or two stator windings and/or stator saliency, and rotor torque due to hysteresis and/or eddycurrent action. The derivation of an equivalent quadrature motor, first shown by Kron, is extended to the range of motors under consideration. The idealized treatment in the paper permits hysteresis to be included in the linear analysis of single-phase motors.

INTRODUCTION

induction and hysteresis motors with one or two stator windings and a cylindrical rotor. In particular, no restriction is placed on the disposition of the stator winding(c) with respect to the axis of stator saliency, if present, and the representation of a tapped winding [17], [6] is considered. The treatment relies on linear theory: harmonics, end-winding, and slotting effects, etc., are ignored and the usual assumptions relevant to the Kron primitive machine are employed. All voltages, currents, and impedances are phasor quantities.
GENERAL EQUIVALENT CIRCUIT Consider a machine, shown schematically in Fig. 1, to have a cage winding on a cylindrical rotor and a stator with main and auxiliary windings inclined at angles a and B, respectively, to a line of magnetic symmetry defined as the direct axis. Each stator winding has a leakage flux, rv flux that links the rotor, and a common mutual flux that does not link the rotor. For the equivalent Kron primitive machine (Fig. 2) each stator winding is represented by an orthogonal coil pair and the rotor has two orthogonal coil pairs to represent the cage winding and hysteresis in the core material, respectively. A common mutual flux is assumed for each axis. With suitable assumptions for the representation of hysteresis in the rotor core, considered in the following section, the eight simultaneous equations for steady-state operation of the primitive machine (Fig. 2) may be transformed by well-known relationships [15] to rotating-field equations, in terms of forward and backward voltages and current (Appendix I) to give an equivalent circuit as shown in Fig. 3.

IT IS WELL KNOWN that unequal contrarotating magnetic fields in single-phase motors may be produced by split-phase wiindings, shaded lpoles, and/'or stator saliency [1]. Many equivalent circuits [2]-[5] employing these methods have been derived for single-phase induction motors, but with some exceptions [6], [7], the treatment is often restricted to one type of machine. Although the polyphase hysteresis motor has been analyzed [8]-[131, little has been published with particular reference to single-phase hysteresis motors [1]. Based onl Kron's work [14], [15] and the analysis of asymmetrical windings [6], [16], two types of general steady-state equivalent circuits are derived for the family of single-phase
Machinery Committee of the IEEE Power Group for presentation at the IEEE Winter Power Meeting, New York, N. Y., January 25-30, 1970. Manuscript submitted June 16, 1969; made available for printing October 15, 1969. The author is with the School of Electrical and Electronic Engineering, University of Bradford, Bradford, England.
Paper 70 TP 9-PWR, recommended and approved by the Rotating

You might also like