You are on page 1of 6

1769

IEEE TRANSACTIONS ON POWER APPARATUS AND SYSTEMS, VOL. PAS-87, NO. 9, SEPTEMBER 1968

/1.56

Zs= VL/C =

Nominial voltage: 12 volts rms line to nieutral.


Inductanices:
Li = 15.4 iHi,sectioii, L = 14.1 mH/section.
Capacitances: 4 iI- star:,
Cl - 0.186 RF/section, C,n = 1.51 jF/sectioii.
Resistanices: Ro = .3.2 Q, RI = 0.3 Q/section.

X 10-3/18.9 X 10-9

= 288 Q

Zso/Za,= 646/288= 2.25.


Wave speeds:

so=

l//Lo

= 1.12 X 105 mi/s

Si= 1/VLC = 1.84 X 105 mi/s

Toleranice of R,JL.C: O5 percenit.

sols= 0.61.
Wave travel times:
to = i/so = 180/1.12 X 105 = 1.61 ms

l/s1 = 180/1.84 X 105 = 0.975 ms.


Pulse width: to- t = 1.61 - 0.98 = 0.63 ms.
Period: to + t1 = 2.59 ms.

ti

B. Artificial Line of TNA


18 7r sections each representing 10 miles (16 km) of line.

Frequency scale: 1.

REFECRENCES

[11 D. E. Hedman, 1. B. Johnson, C. H. Titus, and D. D. Wilson,


[21
[31
[4]

[5]

Impedance scale: 1.

"Switching of extra-high voltage circuits: 11-Surge reduction


with circuit-breaker resistors," IEEE Trans. Power Apparatus
and Systems, vol. 83, pp. 1196-1205, December 1964.
L. Paris, "Basic considerations of magnitude reduction of switching surges due to line energization," IEEE Trans. Power Apparatus and Systems, vol. PAS-87, pp. 295-305, January 1968.
E. Maury, "Synchronous closing of 525 and 765 kV circuit
breakers: a means of reducing switching surges on tinloaded
lines," CIGRE, Rept. 143, 1966.
L. Norlin, "Single-pole reclosing: field tests in Sweden oni a 220
kV, 480 km transmission line," Rept. to CIGRE Study Committee 13, May 1953.
H. A. Peterson, Transients in Power Systens. New York:
Wiley, 1951; also, New York: Dover, 1966, pp. 17-25.

Optimum Number, Location, and Size of Shunt


Capacitors in Radial Distribution Feeders
A

Dynamic Programming

Approach

HERNANDO DURAN, MEMBER, IEEE

Abstract-A new method is described which determines the


optimum number, location, and size of shunt capacitors in a radial
distribution feeder with discrete lumped loads so as to maximize
overall savings, including the cost of capacitors. The method also
determines when capacitors are not economically justified. Dynamic
programming techniques are used and several algorithms developed
to obtain the optimal solution by regarding the optimization process
as a multistage decision process with the desired Markovian property. Special cases are studied and solutions obtained when no
capacitor cost, cost proportional to installed capacity, and cost
proportional to installed capacity plus a fixed cost per installed bank
are considered. The methods are suitable for efficient solution in a
digital computer.

INTRODUCTION

ECONOMIC benefits resulting from energy and peak loss


reduction when applying shunt capacitors in a radial
distribution feeder can be optimized by properly selecting the
location and size of the capacitor installations along the feeder.
Schmill's work [1] is a step in this direction. Using calculus,
Schmill develops a method of finding the optimum sizes of n
capacitor banks when their locations are specified. A search
routine is then suggested in which loss savings are compared for
different locations of the n capacitor banks until optimum savings
are obtained.
This paper presents a method for finding the optimum number,
location, and size of shunt capacitor banks in a radial distribution
Paper 31 TP 67-93, recommended and approved by the Trans- feeder using the techniques of dynamic programming. Some
mission and Distribution Committee of the IEEE Power Group for advantages over the previous work are:
presentation at the IEEE Winter Power Meeting, New York, N. Y.,
1) The cost of the capacitor banks is considered a function of
January 29-February 3, 1967. Manuscript submitted October 31,
1966; made available for printing April 4, 1968.
the kilovar capacity of the banks.
The author is with the University of the Andes, Bogota, Colombia.
2) The number of shunt capacitor banks is optimized. As will
He is presently on leave of absence with Stanford University, Stanbe shown, this number depends on the cost of the banks.
ford, Calif.

1770

IEEE

N-2

N-l

TRANSACTIONS

K-1

ON POWER

APPARATUS

SOURCE
NODE

Is~~ ~ ~ -I- r

1968

-L
Fig. 1. Single-ended radial feeder.

The formula was developed using Cook's work [2].


The cost of installing a shunt capacitor bank at node k is
taken to be a function C (irk) of the reactive current supplied by
the capacitor at the node. In the general formulation of the

ILK-I

IL K 3

AND SYSTEMS, SEPTEMBER

ICK

method it is assumed that C (ik) is known for values of


greater than zero.

iCk

'I

'I

--

Return Function
The return function considers the total savings in losses
afforded by shunt capacitors minus the total cost of the capacitor banks installed, namely

LK

Fig. 2. Conventions for currents at node k.

E [SLm(Icm) -C(icm)J
m=1

(2)

3) The routines which search for the optimum are simple and
suitable for efficienit computer programming.
It is assumed that voltage control is no problem. The method
does not conisider the increase in revenue due to increased
voltage and does niot take into consideration the use of time and/
or voltage switched shunt ca)acitors.

The objective of the method developed is to maximize the


return function F, subject to the restrictions

PROBLEM

METHOD

To describe the feeder, Fig. 1, a number is assigned to every


node and branch. Nodes are labeled in increasing order toward
the source node. Branches are identified by their receiving end
node labels.
Basic Formulas
Associated with every node k, Fig. 2, there is a connected
reactive load specified by the maximum reactive current ilk.
There may also be a shunt capacitor hank installed at the node
characterized by its reactive current ilk. Associated with every
branch m there is a line resistance rm. Also, 'tin deniotes the total
maximum load reactive current and 'cm the total capacitive
current flowxing through the branch.
Reduction in losses in the branch m are calculated by

S fm(7Ctli)

7--=

r7m (2AJcinIjii

BIc

2)

(1)

where
A

KILIT + K2
K1T

K2

and
LF load factor of reactive current
T duration of load cycle
K1 factor reducing energy losses in kilowatt-hours per load
cycle to dollars per year
K2 factor redtucing peak power losses in kilowatts to dollars
per year.

imt >,

0,

1, 2, 3, * *,

n.

A multistage maximization process is used.


Stage: Associated with every node there is a stage in the
process of maximizing the return function.
State: The state of the process at the kth stage is defined as the
total capacitive current flowing (toward the source) into node k.
For the single-ended feeder, this is the current clk-l flowing
through branch k -1.
Decision: At any stage k a decision conisists in choosing a
value for the capacitive current Ick flowing through the kth
branch.
Constraints: The non-negativeness of the current ilk supplied
by the shunt capacitor bank at node k requires that the total
capacitor current Ilk flowing through the kth branch be at least
equal to the total capacitive current flowing into node k, namely

Ick

Cck-1.

(3)

Policy: A policy consists in choosing a set of decisions for the n


stages, namely, a set of values for the variables lck, k = 1, 2, 3,
* * , n. Any policy must be consistent with the constraints of the
problem.
Optimal policy: This policy maximizes the return function.
Markovian property of return function: Any return function is
called Markovian if it possesses the following property: After k
decisions, the effect of the remaining n k stages of the decision
process upon the return function depends only upon the state
of the process at the end of the kth decision and the subsequent
decisions [3].
-

1771

DURAN: SHUNT CAPACITORS IN RADIAL DISTRIBUTION FEEDERS

If Fk is the return function up to the kth stage, then


k

Fk = L> [SLm(Icm) - C(Icm

m==1

m-1) ]

C1* -IcI* (0)


I2* = IC2*(Ic*)

(4)

(9)

with the constraints


'cm

I fn*
= Icn*(f*n-1*),

Icm-i.

The state ck after the kth stage depends only upon decisions
taken at this and previous stages. The total return function is
F

F1

E [SLm(Icm) - C0(I -Icm-l)]


m=k+l

(5)

The second member of the right side of (5) involves the set
of decisions (Ick+ 1, Ick + 2, * * Jcn) and the knowledge of the state
ck at the end of the kth stage. Thus, the return function is
Markovian.
-

Recurrence Relation
Equation (5) can be written

Fn-I + [SLn(Ic6n) - C(Icn -Icn-l)]

F =

(6)

For a given value of 'cni, the choice of I,n has no influence on


Fn-l. Then the decision Ion which maximizes the return function

(optimal decision) must be chosen so as to maximize the value


of the term in brackets in (6), subject to the constraint I,. I
cn-l.- The optimum value Icn*(Ia-l) of the nth decision is thus
obtained as a function of the state ICn- at the nth stage. The
optimum value of the term in brackets is

G,(Icn-1)

Icn

max

I
'l.n-

[SLn(Icn)

C(Icn- Icn-1)

The following algorithm summarizes the procedure to obtain


the optimal solution using tables to represent the funietions
Gk(ICk1j) and lck*(Idck-1l.

Algorithm. 1:
1) A discrete set of capacitor current levels (states) is chosen
over an adequate range. These levels are lI for k = 1, 2, *, m.

The difference between consecutive levels may be taken as the


current corresponding to the smallest capacitor size considered
for any installation. A conservative maximum for the capacitor
current levels is the value of the total reactive load current.
The minimum current level is zero.
2) A value is assigned to GQ+l(I1k) for all k. These values
correspond to savings beyond the source node due to the capacitor currents Ik. If these savinigs are not taken into conisideration,
the assigned value will be zero.
3) Using the recurrence relation

[SLk (I) - C (1j


Gk.(I) = Imax
j >I i

- I,) +

Gk+1(I)1

tables of the 2n functionis Gk(Ii) and Ik*(Ii) are built for all
capacitor current levels and for values of k = n, 1.
4) The state corresponding to node 1 is zero. Best overall
savings for the feeder are then given by Gi(0).
The sizes of the capacitor banks are obtained using the
relations

Ik*

ick

Jk*(Ik-f*)
Ik* - k-1*F = Fn-2 + [SLn i(Icn-1)
C(Icn- Icn-2)] + Gn(Icn-1)
If i0k = 0, no capacitors should be installed at node k. It must be
subject to I'n-i ) Ien-.2 For a given value of Icn-2, then
noted that the only constraint considered in item 3 of algorithm I
is the non-negativeness of the capacitor current supplied at
max
Gn1(Icn-2) =
[SLn_i(Icn-1) C(Icn-1 Icn-2)
Icn-i
Icn- 2
any node. However, in many situations additioinal restrictions
be placed on the problem. One example is the consideration
may
+ Gn(Icnl4)]
of a maximum shunt capacitor bank size which would be inand Icn-1*(Icn-2) is the optimum value of Icn-ni Proceeding in stalled at any one node. This constraint can be incorporated in
item 3 by ignoring the cases in which Ij - 1i is larger than the
this manner the following general relations are obtained:
specified maximum bank size.
F = Fk-i + [SLk(Ick)
C(Ic Ict-4)] + G0+1(Ick) (7)
The preceding algorithm is valid for any capacitor cost mnodel.
For
sonme simple cost models, however, more efficient algorithms
and
Proceeding, the return function can be written
-

Gk(Ick-1)

max
Ick Ick-I

[SLk(Ick)

C(Ck

Ick-l)

can be developed. Three cases will be considered later.


It is convenient to introduce the following definitions:
k-1

+ G+1 I(Ick)]. (8)


Equation (8) gives the

recurrence

relation for the functions

Gk(Ick-1). Equation (7) for the last node (label 1) gives:


F

G0(0)

SLi(Icl)
max
Ic ) 0

C(Ici) + G2(Icl)

[SLI(Il)

Ri= mE=i rm (sum of resistances between nodes i anid k-)

C(Ici) + G2(ICI)].

The state Ico corresponding to the first stage is taken as zero


since no capacitor current is flowing into this node. The optimal
value of the return function is then F G1(0). The optimal
policy is obtained, using the relations
=

(10)

Mik

k-1

E Ilmrjm (sum of moments between nodes i aind k).

m=i

(11)

These are generalizations of the definitions used by Schmill


[1]. They are particularly convenient as they make it possible to
express the optimal loss savings and optimal capacitive current
in any section of the feeder in a very compact way. They are
also useful when implementinig algorithm 1.

1772

IEEE TRANSACTIONS ON POWER APPARATUS AND

The followin-g two theorems, proved in Appendix I, generally


hold. Thev are used in further developments in this paper.
Theorem. 1: When no capacitors are installed between nodes
i and j, i < j, the value of capacitor current I,i? flowing into node
j which mfaximizes loss savings LSij(I,ij) between these nodes is

I1cj

AM.jIBR1j

(12)

and the corresponding value of maximum loss savings is

SLij

3AM1J20iIIBRtj.

(13)
Theoren. 2: Letting a capacitor bank be iristalled at node k
between nodes i and j (i < k < j), and assuming that all reactive
loads are iniductive, the following relations hold:
a)
Icik Ickj
SLik* + SLkj* ; ZSI
b)
where I,ik* and I,?* are the values of capacitor current which
maximize loss savings between nodes i - k and k - j, respectively; SLik* and SLkj* are the corresponding loss savings with
capacitors installed at node k; and SL1j* is the value of maximum
loss savings between nodes i and ] with no capacitors installed at
node k.
In other words, this theorem states that loss savings can be
improved by installing capacitors between any two nodes of the
feeder. The equal sign holds when the reactive load connected
at node k is zero.
SPECIAL CASES
In considering the following cases it is assumed that any
amounit of shunt capacitive current may be iupplied at any node
of the feeder. This may be a realistic assumption if the optimum
capacitive current needed at a node is large compared with the
current supplied by the smallest increment available. It is also
assumed that the loads are inductive.
=

Case I-Capacitor Cost Not Considered


From theorem 2 it follows:
Theoremn 3:
1) Maximum loss savings are obtained when capacitors are
installed at all reactive loads.
2) The optimum capacitor installation at node k is Ailk/B,
where ilk is the reactive load connected at the node.
(16)
3) Best total loss savings in the feeder are given by
3A2 n
best loss savings = - E rMIlM2
B m=1
A2

(original reactive

current

losses). (17)

Case II-Capacitor Cost Proportional to Installed Capacity


If C, is the annual cost in dollars per reactive ampere of
installed capacity, then the cost of any installation is C(1,) =
C,l1, assuming a linear cost model. Maximum overall savings for
Case II are obtained when capacitors are installed according to
the following algorithm.

Algorithm 2:
1) The largest i is found for which the following inequality
holds:

(18)
Min- Ri,i?-_ > C,16A.
If there is no such node, then it is not economical to install
capacitors in the feeder. The optimum value of the shunt ca-

SYSTEMS, SEPTEMBER 1968

pacitor bank at node i is

AMin/BRin - C1/6BRin- AI1i1/B.

(19)

2) Shunt capacitors are installed at all nodes k < i, of capacity


Aizk/B where ilk is the reactive load connected at node k.
Case III-Capacitor Cost Proportional to Installed Capacity
Plus Fixed Cost per Installed Bank
C, is as defined for Case II and Co is the annual cost in dollars
per bank of a fixed charge for installing a capacitor bank at a
node of the feeder. Best overall savings for Case III are obtained
when capacitors are installed in the manner indicated by the
following algorithm.

Algorithm 3:
1) The largest i is found for which the inequality (18) holds.
If no such node exists, it is not economical to install capacitors.
2) Starting with node i and going down to node 1, the quantities Gk' (k = i, i - 1, * 1) are calculated using the following
recurrence relation:
Gk' = max (3A I(Mke- C1/6A)'/BRkfl
t) j>kC

CO, Gj' + 3A2

Jll'j/BRk

C0). (20)
With every k a number j*(k) is associated which gives the node j
-

for which (20) is maximum. If the maximum corresponds to the


term in the left, let j*(k) = n.
3) Best overall savings for the feeder are found from
best savings = max (Gk').
(21)
i. k ) I

If the maximum value of (21) is negative, then it is not economical


to apply shunt capacitors. The number k for which (21) is
maximum is the number of the node with smallest label at which
capacitors should be installed.
The value of the capacitor installation at this node is

(22)
ick = AMkj*(k)/BRkj*(,) = Ic.
If j*(k) = n, part 6) applies, otherwise;
4) The number i = j*(k) gives the next node number in
sequence to node k where the next capacitor bank should be
installed, the optimum size of the installation being Ic, -I1,

1,ci

AMjj*(j)1BRvj*(@v)-

If j*(i) n, see part 6); otherwise;


5) The new value of Ic is ICV and the new value of k is i. Step
4) is repeated until the condition j*(i) = n is reached; then;
6) The last capacitor bank is installed at the node i for which
j*(i) = n, with capacity given by
(23)
ice = A(Min - Cl/6A)/BRn- Ic
A proof of algorithms 2 and 3 is given in Appendix II. Some
observations should be made concerning algorithm 3:
1) Fixed costs may be made to depend on the capacitor location. It is only necessary to replace Co in (20) by the fixed cost
Cok at node k.
2) Schmill's problem of optimizing the location and size of a
given number of capacitor banks in a feeder can be solved in a
simple manner using algorithm 3. In this case the value of C,
is zero. To solve this problem it is necessary to keep track of the
number of capacitor banks installed beyond node k when finding the value of Gk'. Then, in (20) the fixed cost is made zero if
this number is less than the specified number of capacitors or a
very large value otherwise.
=

1773

DURXN: SHUNT CAPACITORS IN RADIAL DISTRIBUTION FEEDERS

COMPARISON OF METHODS
The main advantages of algorithm 1 are:
1) It is more general in that it does not make any assumptions
about the cost of the capacitors.
2) It is more realistic because it takes into account the fact
that capacitive current may be supplied onily in discrete amounts.
On the other hand, algorithm 3 is much simpler and more
efficient. It may be used with advantage when
1) the smallest increment available in capacitive current is
small compared with the amount needed at the nodes.
2) the actual cost of the capacitor banks suits the cost model of
Case III.
FURTHER WORK
Further work on the application of dynamic programming to
distribution feeder planning may include the addition of voltage
control or other constraints to the problem, the consideration of
different types of voltage and/or loss control devices (voltage
regulators, switched capacitors, etc.), the extension of the
problem to multi-ended radial feeders, and the evaluation of
additional technical or economical features of the feeder (reactive
losses, load level, etc.)
Also, consideration of present conditions of the system may
include already existing capacitor banks or other devices that
can be relocated to improve overall benefits.

Then

SLi/c* + SLk] = A2_jlfik2/BRik + A2tIfAc2/BRkj

A2(M1s, + Mkj)2/B(Rik + RB/)


= (AMIj)2/BRij = SLij*.

APPENDIX II

Proof of Algorithm 3
The decisions are defined as the choice of:
1) the next node in sequence j > k at which shunt capacitors
are installed; the node k may be the last capacitor installation,
too;
2) the value of the capacitive current clk flowing between
nodes k and j, assuming that capacitors are installed at node k.
Letting Fk/c be the return function up to the (k - 1)th stage
and GCi(ck) be the best returns beyond nodej as a function of the
capacitive current c/k = I'j i flowing into node j, the total return
function may be written
F

Fk/l + [SLki(l/ek)

C(Ick -Ick-l) I + C(Ick) (24)

where

SLkj(Ick)

BRkcI ck/)

3 (2A ikjlIck

and
APPENDIX I

Proof of Theorems 1 and 2


Expressions (12) and (13) of theorem 1 follow immediately
after equating to zero the derivative of the loss savings with
respect to the capacitive current.
To prove theorem 2, from definitions (10) and (11) the following relation holds:
RfkIjj K M (k< Rikllk

and
A
B

A
B

C(Ick

GkC(Ick-l)

A
SI
B
--

and finally
Iclk*

Ic/j*.
To prove part 2): the following relation is used:
a2/b + c2/d ) (a +c)2/(b + d)(equal sign if a/b = c/d).
Using the result of theorem 1,
(

SLi/* = A2Mj/2/BRij
SLik *= A 2Mik21BRik
SLkj* = A2,111kj2/BRk/g
Also
M = Mik +

Mi/kc

Co + C'Ic/

ClI7Ck-l-

max

Ic/k) cIck-I

[3(2AM/jIck - BRk/lckI)

C'Ick + C,Ictc, + GC(Ick)]. (25)


Since the only dependence of GC with Ic/-l is through the
term ClIck-l, the new functions are defined:
- C0 -

Gk'

GkC(Ick-1)

ClIck-l.

Then (24) and (25) become


F

ICU^

lIk-1)

Then

Similarly for k < j


A
-Ilk <,
B

Fk-I + (SLk/i(Ick)

CO + Cljck-1) + Gj'

(26)

and

Gk/'

max [3(2AMkjIlk-

Ick >)Ick- 1

BRikICk/2)

Co + GC']. (27)

Now the maximum of (27) occurs when

Ick/ = AMkj/BRckJ
and the constraint Ick* > fCk-,* is seen

theorem 2.
The maximum value of (27) is

GC'

to be satisfied from

GC? + 3A2M/C12/BR/J - CO.

(28)

The possibility of not installing capacitors between nodes


k and n must be taken into consideration; for this case
=
max [3(2AMknlck -CO -

GkCI

ick ) Ick-

BRcnIck/2)

Clck/]

giving

and

Rik + RkJ

Rij.

GCk'

3A2(Mcn - Cl/6A)2/BRkn

Co.

(29)

1774

IEEE

The value of Ice which maximizes (29) is

(30)
Ik-l*, a sufficient condition is

Ic).* ) AMk-,,/BRkl,k ) AMil/BRik

= Ick-l*

where i is the preceding node where capacitors have been installed. But

AM7_l,k/BIRk-,k

AIlk-IIB.

Then
I&*= A(Mkn -

CI/6A)BRkn

AlIx-/B

or

M.n- RicI_-

C1/6A.

ON

POWER APPARATUS AND SYSTEMS, SEPTEMBER 1968

Discussion

Ick* = A (Mkn- C1/6A)/BRRt..


To satisfy the constraint Ilk
that

TRANSACTIONS

(31)

Only nodes k for which (31) holds need to be considered.


Otherwise the constraint will not be satisfied. When maximizing
(27), then, it is sufficient to consider nodes j such that i ) j > k
where i is the largest number satisfying (18).
This proves algorithm 3. Algorithm 2 is encompassed as a
particular case of algorithm 3 for which C0 = 0. Consideration
of (20) and theorem 2 leads to the observation that the maximum is obtained for j = k + 1 and subsequently to algorithm 2.
REFERENCES
[1] J. V. Schmill, "Optimum size and location of shunt capacitors on
distribution feeders," IEEE Trans. Power Apparatus and
Systems, vol. PAS-84, pp. 825-832, September 1965.
[21 R. F. Cook, "Calculating loss reduction afforded by shunt
capacitor application," IEEE Trans. Power Apparatus and
Systems, vol. 83, pp. 1227-1230, December 1964.
[3] R. Bellman, Adaptive Control Processes, A Guided Tour. Princeton, N. J.: Princeton University Press., 1961, ch. III.

D. Purdy (New York State Electric & Gas Corporation, Binghamton,


N. Y.): I wish to compliment the writer on his paper which should
be very useful in the economic location of capacitors.
From a practical standpoint, it is sometimes infeasible to locate
capacitors in banks large enough to take full advantage of the
economic locations indicated without resorting to special current
limiting fuses. The expense of the current-limiting fuses then destroys
the economics of the particular installation involved. This occurs
because the case rupture withstanding capability is limited to a
maximum of 5000 amperes. The standard fuse which must be used to
protect the capacitors against rupture limits the size of the capacitor
bank to 300 kVA on 4800- or 4160-volt distribution lines.
Even at the 12 470- and 13 200-volt level, the large new units now
offered encounter the same difficulty, imposed by their 5000-ampere
case rupture with standard rating. Locations which appear justifiable
on other bases lose the economic advantage because they require
protection with expensive current-limiting fuses. I would like to see
stronger capacitor units become available.

Manuscript received February 21, 1967.

H. Duran: I wish to thank Mr. Purdy for his discussion. He makes


reference to a very important fact which has often been neglected
when considering the economics of capacitor installations. The cost
of a capacitor installation must include the cost of all accessories,
like current-limiting fuses if necessary. It seems that the high cost of
this protective equipment may increase the relative weight of the
fixed cost of the capacitor installation. In any case, from the standpoint of the paper it must be noticed that the algorithms described
do take into consideration the cost of capacitor installations. In
particular, algorithm 1 considers the actual size and cost of available
capacitor installations, and algorithm 3 considers a fixed cost plus a
cost proportional to installed capacity. Therefore, if the cost of
protective equipment is included in the cost of overall installation,
then the locations specified by algorithms 1 and 3 will in fact be
optimum from an overall econiomy standpoint.

Manuscript received July 10, 1967.

You might also like