IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO.

2, MAY 2000 879

Optimum Voltage Regulator Placement in a Radial

Power Distribution Network
Anastasia S. Safigianni, Member, IEEE, and George J. Salis

AbstractA computer algorithm for optimal voltage control propose VRs for a network completely compensated with
with voltage regulators, suitable for large radial distribution net- capacitors.
works is given in this paper. An objective function concerning the The network voltage is the criterion for the selection of the
total cost of the voltage regulators (investment and maintenance
cost) as well as the cost of losses of the examined networks is de- optimal VR number, locations and tap positions in [3], [4], [5].
veloped and constitutes the base of the algorithm. This algorithm An economic function, which estimates the power and energy
makes the initial selection, installation and tap setting of the voltage losses, is given in retrospect but this function is not utilized
regulators, which provide a smooth voltage profile along the net- during the main problem-solving process concerning the op-
work, utilizing former algorithms suitably modified and optimized. timum VR placement and installation. The power flow at the
Then it attempts to minimize the number of the initially selected
voltage regulators as much as possible, by moving them in such network segments is calculated by an approximate method (V-P
a way as to control the network voltage at the minimum possible model, [11]).
cost (maximization of the objective function). The algorithm is fast, In [6], the voltage regulation is initially attempted by
efficient and reliable as its application to practical distribution net- changing the tap positions at the substation and solving again
works shows. the capacitor problem. If the desirable voltage regulation is
Index TermsRadial primary distribution networks, Voltage not achieved in this way, a VR is placed at the main feeder,
control, Voltage regulators, Cost minimization. next to the node where the subfeeder with the heaviest load is
connected and then the proper tap position of this VR is de-
I. INTRODUCTION termined. Sometimes VRs are placed at very long subfeeders
to improve their voltage profile. Consequently in this paper

I F THE reinforcement of a network were required because

of excessive voltage drop, any such reinforcement could be
deferred if the voltage drop could be sufficiently reduced by
the VR positions are in a way predetermined and they are not
indicated by the method.
The present paper utilizes the basic philosophy of the algo-
some means, subject to economic as well as technical consid-
rithm of [3], [4], [5] only with regard to the initial VR selection,
erations. Various devices such as capacitors and voltage regula-
placement and tap setting. But even in this stage it differenti-
tors (VRs) can be installed to reduce the voltage drop, experi-
ates the method of the above papers by calculating the currents
enced at critical points of medium voltage networks. Conductor
at the network segments by load flow analysis instead of an ap-
replacements at network segments can also be used to maintain
proximate method as these papers do. After the determination of
the voltage along the entire network.
the initial VR number and locations the optimization procedure
The papers [1], [2] propose reconductoring of currently
does not finish in opposition to [3], [4], [5], because it is true
operating primary distribution networks in order to optimize
that the resultant solution solves the problem of the excessive
them technically as much as possible defraying the minimum
voltage drop technically but it is not surely the most econom-
cost. Representative papers dealing with the optimization of
ical one. For this reason an objective function is developed in
medium voltage networks operation by selecting (kind and
the context of this paper, which includes the VR total cost and
size), installing and controlling the appropriate number of ca-
the cost of losses of the examined network (evaluating the power
pacitors, are [3][10]. Fewer papers deal with the determination
losses for the peak load and the energy losses according to an ap-
of the optimal locations and real-time control (tap positions)
proximate annual load curve consisting of twelve distinct mean
of a minimum VR number, in order to minimize the peak
monthly load values). Based on this function a procedure in-
power and energy losses and provide a smooth voltage profile
vestigating the possible reduction of the number of the initially
along a distribution network with lateral branches, under time
selected VRs is evolved so as the finally proposed solution will
varying conditions. Recent papers of this kind are [3], [4], [5],
be the most economical one.
[6]. Specifically an integrated method for the optimal reactive
The application of the proposed method to a great number of
power and voltage control of radial distribution networks by
practical radial primary distribution networks has proved that
using capacitors and VRs is given in [3], [4], [5], which are
there are no restrictions according to the size of the examined
parts of a complete study, as well as in [6]. All these papers
networks as the method is easy to use, very fast and efficient.
decouple the capacitor problem from the VR problem and

Manuscript received September 28, 1998. II. INITIAL VOLTAGE REGULATOR PLACEMENT AND CONTROL
The authors are with the Department of Electrical and Computer Engineering,
Democritus University of Thrace, Xanthi (67 100), Greece. A power flow analysis method precedes the procedure of the
Publisher Item Identifier S 0885-8950(00)03833-5. VR selection and installation in a radial distribution network, in
08858950/00$10.00 2000 IEEE
880 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 2, MAY 2000

order to evaluate the exact node voltages and segment currents. where is the cost of losses before the VR installation
Thus it is possible to avoid the fallacious assumption that a ra-
dial distribution network is optimally compensated using par-
ticular capacitors, because it is impossible to have a capacitive (4)
current absolutely equal to the inductive current at each lateral
or sublateral branch due to the tree structure of the said network.
The power flow method used in this paper, which is based on where:
the direct application of Kirchhoffs voltage and current laws, : the network losses at peak load without VRs [kW]
is analytically described in [12] and has excellent convergence : the network losses during the time period T without
characteristics when applied to large scale weakly meshed or VRs [kW]
radial networks. According to this method, the voltage at the The VR problem consists of two subproblems, that of the
source as well as the active power and the reactive power optimum selection and that of the optimum control. The first
at all the other nodes are given. The network is swept back- subproblem concerns the determination of the VR necessary
ward and forward and the segment currents as well as the number and locations on the network, to achieve the desirable
node voltages are calculated. This power flow technique pre- voltage profile, while the second subproblem concerns the se-
supposes that the examined networks are recorded as tree data lection of their tap positions. The procedure for the first VR se-
structures, so that they can be processed without special node lection and installation is relative to that described in [3], [4],
numbering. This simplifies the node sweep and increases the [5]. The following steps are executed by sweeping the network
flexibility of the method according to the addition or subtrac- forward (from the source to the ends):
tion of branches. The voltage at each network node i must be Localization of the nearest node to the source where the
limited between specific limits and affected by the voltage drop is out of the predetermined limits.
size and the character of the load, in order that the total voltage A VR installation at this node and tap setting of this VR so
drop along the network and for each one of the network ends is that the voltage of the node is as close as possible to the max-
less than a definite percentage % of the rated voltage. imum permissible voltage, without exceeding it.
If this requirement is not satisfied a voltage regulation method Calculation of the node voltages by load flow analysis, which
must be applied. takes into account the already existing VRs. The initial node
If the maximum possible voltage level were to be achieved, voltages, at the beginning of the repetitive load flow proce-
as the power losses are proportional to the current squared and dure, do not necessarily need to be identical to the source
the current is inversely proportional to the voltage, it would be voltage . They may be equal to the node voltages before
theoretically desirable to put one VR at each network node. This the VR installation at the step II.
would lead to the minimum cost of losses . In fact VRs Examination of the node voltages to see if the desirable limits
are installed only at some network nodes and they maintain the are violated. If not the VR installation procedure comes to an
voltage every moment according to the load. This means re- end or else all the steps are repeated.
spective time variation of the network current and relative losses A VR tap position at the node i can be generally determined by
saving. The cost of losses after the VR installation is given the relation (5).
by the relation:
Int (5)
(1)
where:
Int : the function which returns the integer part of its argu-
where : ment
: the annual demand cost [dr/(kW a)] : the voltage at the node i before a VR installation at
: the energy cost [dr/kWh] this node under peak load conditions
: the number of the network segments : the VR voltage step (0.006 25 p.u.)
: the network losses at peak load with VRs on the
During the VR installation procedure the voltage refers
network [kW]
to peak load conditions, which means that this voltage takes its
: the network losses during the time period with
minimum value given by the relation:
VRs on the network [kW]
: time period during which the network losses are
(6)
constant
and
where:
: the percentage voltage drop from the source to the
h (2)
node i resulting from power flow analysis.
When the method proposes for the VR a tap position greater than
The cost lies between two limits its possibilities, a location closer to the source must be searched
out for this VR, which will lead to a smaller tap position, even
(3) if the voltage drop at this location is not unacceptable.
SAFIGIANNI AND SALIS: OPTIMUM VOLTAGE REGULATOR PLACEMENT 881

The profit because of the losses saving, after the VR installa-

tion is:

(7)

This profit increases as the voltage at the network nodes ap-

proaches its maximum permissible value. That means the tap
positions must be adjusted during the year, so that the VR output
voltage approaches its maximum value without exceeding
it. The necessary variation of the tap positions , in order to
have the maximum output voltage, is given by the relation:

int

(8)

where:
: the previous VR tap position

(9)

The above described method for the VR placement is defective

because it does not take into account the annual investment and
maintenance cost for all the installed VRs which is rela-
tively high. An objective function taking into account this cost
can be written as:
Fig. 1. Block diagram of the Pull_Back_Regulators procedure.
(10)

with: When the method proposes more than one VR it must be deter-
mined if the voltage can be limited between the predetermined
limits using a smaller VR number in a way that maximizes the
objective function (10).

(11) III. ALGORITHM FOR THE RESTRICTION OF THE VOLTAGE

REGULATORS
where: A recursive procedure named Reg_Recursive( ,Reg) is
: the number of the network nodes used to examine whether fewer VRs contribute an acceptable
: the VR annual investment and maintenance cost voltage profile to the network whilst simultaneously increasing
[dr/a] the value of the objective function (10). The arguments of this
when a VR exists at the node i procedure, which constitute the state variables of the problem,
when there is no VR at the node i are the initially proposed VR number and relative information
: the VR investment cost [dr] Reg [ ] for each one of these VRs such as the cor-
: the VR annual maintenance cost [dr/a] responding network node and tap positions. The first time the
: the annual capital recovery factor given by the above recursive procedure is executed, the VR number and tap
relation: positions are those resulting from the algorithm previously de-
scribed which are referred to as Scheme 1 in the block dia-
gram of Fig. 1.
(12) The first stage of the Reg_Recursive( ,Reg) procedure
is the Find_path( ,Reg) procedure as in Fig. 1 is shown.
where : During this procedure each one of the VRs is successively
: the annual interest rate (no inflation) moved to nodes closer to the source, which without fail belong
: the expected VR life. to the route joining directly the source and the VR location. The
After the VR installation according to the above mentioned steps total movement route for one VR depends on the current stage
I-IV the initial value of the objective function (10) is calculated. of the recursive procedure and whether or not VRs are present
882 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 2, MAY 2000

at the nodes which are further away from the source and stem
from the initial position of this VR. The contribution of these
VRs to the network voltage profile depends directly on the lo-
cation and the tap position of the examined VR. Every time the
VR is moved, its tap positions are adjusted since its rated input
voltage tends to increase as the VR approaches the source and
the voltage at the new VR location must approach the maximum
permissible value without exceeding it, according to the relation
(8).
The next step is the voltage computation at each node
belonging to network branches, which stem from the new
VR location and an investigation of whether some voltage
drop problem exists. If no problem exists this new location
is acceptable and the VR is moved to the next node toward
the source within a procedure trying to find the route of the Fig. 2. Radial feeder having four voltage regulators.
permissible VR movements. If a voltage drop problem exists
the Find_path( ,Reg) procedure for the examined VR
terminates. In this case the end of the route of the permis-
sible movements is the closest node to the source, where
the VR installation keeps between the acceptable limits the
voltage of the nodes which stem from the VR position. The
examined VR is moved back to the node it was at before the
beginning of the above described procedure, acquiring again
the tap position which corresponds to this location and the
Find_path( ,Reg) procedure goes on examining the next
VR. When this procedure finishes, each VR is accompanied
by a path of its permissible movements. This path may be
regarded as fictitious due to the fact that both the origin and the
end of the path are situated at the initial VR location, making
actual movements redundant.
The next step of the Reg_Recursive( ,Reg) procedure
is the formulation of all the possible pairs (reg ,reg ) derived
Fig. 3. The radial feeder of the Fig. 2 after the reduction of the number of the
from the set of all the installed VRs at the network. voltage regulators.
Supposing that set=[reg , reg , , reg ] then each pair can
be described as (reg ,reg ), with
newReg[ ], provided the VR number is decreased
by one. The value of the objective function (10) is calculated
afterwards (Calc_SF procedure) for the new situation and if
and this situation is more beneficial economically than the previous
situations, it is kept as the best one resulting from the inves-
The total number of the pairs is: tigation procedure up to this moment. The Reg_Recursive
procedure is executed with the form Reg_Recursive( ,
newReg) despite what the result of the objective function is
(better or worse).
(13) The Reg_Recursive( ,Reg) procedure comes to an
end when all the possible VR pairs, (reg ,reg ) have been
For each one of these pairs the following are executed. It is examined. If this procedure has not been called inside a
checked if there are common nodes between the routes path of previous Reg_Recursive( ,Reg) procedure, the total
the VRs corresponding to each pair (Check_paths(reg ,reg ) Pull_Back_Regulators procedure, concerning the VR move-
procedure). If there are not, the Reg_Recursive( ,Reg) ment toward the source in order to reduce their number,
procedure goes on dealing with the next pair (reg ,reg ). In finishes.
the opposite case, the common node furthest from the source, The Pull_Back_Regulators procedure can be made quite
named com is chosen as the node where one VR is installed clear with the following example. Lets suppose that four
instead of the two VRs of the pair. All the VR tap positions are VRs have been installed during some stage of the recursive
adjusted according to (8) and a new power flow analysis takes procedure at the nodes 7, 9, 10 and 16 of the feeder given
place. The network acquires a new form and the Reg_Re- in Fig. 2, which is a quite hypothetical feeder (a practical
cursive procedure acquires new values of the state variables application of the optimum VR placement method to a realistic
( ,newReg), which are , and feeder is given at the next section). The first VR set examined
SAFIGIANNI AND SALIS: OPTIMUM VOLTAGE REGULATOR PLACEMENT 883

Fig. 4. Feeder examined as a practical application of the optimum VR placement method.

by the Reg_Recursive( ,Reg) procedure is set=[reg , execution levels of this procedure. One way to avoid this repeti-
reg , reg , reg ] with . Lets also suppose that the VR tion is to store the network forms at each level of the procedure
paths for the first execution of the above procedure are: and to compare the current examined form with the stored forms
to decide if it is possible to reduce the proceedings. However,
path reg node 7-5-3 this procedure is at least as time consuming as the repetitive
path reg node 9-6 procedure described earlier, so there is no call for its implemen-
path reg node 10-6 tation.
path reg node 16-14
and the six pairs to be relatively examined are: IV. CONCLUSIONS AND RESULTS

reg reg reg reg reg reg A computer algorithm for the voltage control of large radial
reg reg reg reg distribution networks is presented in this paper. The algorithm
reg reg takes into account the actual network data such as conductor
sizes, lateral and sublateral branches as well as load distribution
For each one of the above pairs it is checked if a common node and time variation and handles fast large distribution systems as
com exists. The node 6 is such a node for the pair (reg ,reg ). a total and not in parts making relative approximations. It aims at
So a VR is installed at the node 6 and the VRs of the nodes 9 an acceptable voltage profile along the network at the minimum
and 10 are subtracted. The feeder acquires a new form given in possible cost. That is obtained by moving toward the source and
Fig. 3, which is likely to constitute an improvement in the value readjusting the initially selected VRs, which probably leads to
of the objective function (10). At this point the recursive proce- a reduction of their number and a positive economical result.
dure Reg_Recursive ( ,newReg) with A feeder of the primary power distribution network of the
is executed again for the network form given in Fig. 3. New VR area of Xanthi, Greece, which belongs to the Public Power
paths are defined and there are now three pairs to be examined. Corporation (P.P.C.) and has 229 nodes (load locations and
Every time the Reg_Recursive( ,newReg) procedure branch points), has been examined as a practical application
is completed the feeder acquires again its previous form. of the optimum VR placement method described above. The
During the detection of the most economical way to have an furthest feeder end is 45 km away from the source and the
acceptable voltage profile along the examined network, using feeder segments have mainly conductors of the minimum size
the above described recursive procedure, it is possible for a spe- ( mm ACSR), as Fig. 4 shows. The segment lengths are
cific network form to be examined more than once at different given in Table I. The rated voltage of the feeder is 20 kV and
884 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 2, MAY 2000

TABLE I
DATA AND RESULTS OF THE APPLICATION OF THE OPTIMUM VR PLACEMENT METHOD ON THE MEDIUM-VOLTAGE FEEDER OF FIG. 4

% of the rated voltage is taken as the acceptable voltage nodes. A capacitor of 450 kVAr has already been installed at
variation at each distinguished node as well as at all the terminal the node 167 of the feeder in order to minimize the losses. The
SAFIGIANNI AND SALIS: OPTIMUM VOLTAGE REGULATOR PLACEMENT 885

TABLE I (Continued)
DATA AND RESULTS OF THE APPLICATION OF THE OPTIMUM VR PLACEMENT METHOD ON THE MEDIUM-VOLTAGE FEEDER OF FIG. 4

total peak load of the feeder at the moment the optimization nodes of the feeder given in Table I are coincident with the
procedure begins is equal to 80 A. The installed loads at the above peak load before the optimization procedure begins. The
886 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 2, MAY 2000

standardized VRs used by P.P.C. have nominal values 5 MVA [4] J. J. Grainger and S. Civanlar, Volt/Var Control on Distribution System
and 10 MVA, tap range of % and tap increments 0.625 % with Lateral Branches Using Shunt Capacitors and Voltage Regulators,
Part 11: The Solution Method, IEEE Trans. on PAS, vol. 104, no. 11,
( tap positions). Given in Table I is also the percentage pp. 32843290, November 1985.
voltage drop before the VR installation. It is obvious that the [5] S. Civanlar and J. J. Grainger, Volt/Var Control on Distribution System
voltage drop exceeds the permitted limits in many feeder nodes. with Lateral Branches Using Shunt Capacitors and Voltage Regulators,
Part III: The Numerical Results, IEEE Trans. on PAS, vol. 104, no. 11,
The application of the VR optimization method given in [3], pp. 32913297, November 1985.
[4], [5] after its differentiation, in order to include load flow [6] M. M. A Salama, N. Manojlovic, V. H. Quintana, and A. Y. Chikhani,
analysis, indicates three VRs with nominal values 5 MVA at Real-Time Optimal Reactive Power Control for Distribution Net-
works, International Journal of Electrical Power & Energy Systems,
the nodes 100, 112 and 139 and tap positions for the peak load vol. 18, no. 3, pp. 185193, 1996.
16, 16 and 15 correspondingly. This solves the voltage drop [7] M. E. Baran and F. F. Wu, Optimal Sizing of Capacitors Placed on a
problem as the relative column of Table I shows. At this point Radial Distribution System, IEEE Trans. on Power Delivery, vol. 4,
no. 1, pp. 735743, January 1989.
the Pull_Back_Regulators procedure, proposed in this paper, [8] M. E. Baran and F. F. Wu, Optimal Capacitor Placement on Radial
is used aiming at the restriction of the necessary VR number Distribution System, IEEE Trans. on Power Delivery, vol. 4, no. 1, pp.
and consequently of the requisite cost. Specifically through the 725734, January 1989.
[9] H. D. Chiang, I. C. Wang, and G. Darling, Optimal Capacitor Place-
Reg_Recursive procedure all the VR pairs are examined in ment, Replacement and Control in Large-Scale Unbalanced Distribu-
order to replace them by one VR closer to the source. The final tion Systems: System Solution Algorithms A Numerical Studies, IEEE
result is that the usage of only one instead of three VRs is pro- Trans. on PWRS, vol. 10, no. 1, pp. 363369, February 1995.
[10] H. D. Chiang, I. C. Wang, and G. Darling, Optimal Capacitor Place-
posed which is obviously a more economical solution. This VR ment, Replacement and Control in Large-Scale Unbalanced Distribution
has nominal value 5 MVA and tap position for the peak load 13, Systems: System Modeling and A New Formulation, IEEE Trans. on
it is placed at the node 37 and it solves the voltage drop problem. PWRS, vol. 10, no. 1, pp. 356362, February 1995.
[11] J. J. Grainger, S. Civanlar, K. N. Clinard, and L. I. Gale, Optimal
It is also offering to the feeder more room to increase its induc- Voltage Dependent Continuous Time Control of Reactive Power on
tive load without voltage problem, compared to the case of the Primary Feeders, IEEE Trans. on PAS, vol. 103, no. 9, pp. 27142722,
three VRs, as the relative columns of Table I show. September 1984.
[12] D. Shirmohammadi, H. W. Hong, A. Semlyen, and G. X. Luo, A Com-
The algorithm described in this paper was coded in the Turbo- pensation-Based Power Flow Method For Weakly Meshed Distribution
Pascal language. The time needed for the solution of the prac- and Transmission Networks, IEEE Trans. on Power Delivery, vol. 3,
tical problem given above is 33s using a Pentium-233 MHz. no. 2, pp. 753762, May 1988.
The optimization problem examined in this paper is part of
a more general problem, which has already been examined by
the authors of the paper and aims at giving the more econom-
ical combined optimization solution for currently operating pri- Anastasia S. Safigianni (M97) was born in Greece in 1957. She received her
Dipl.-Eng. degree and Ph.D. degree from the Electrical Engineering Department
mary power distribution networks, in an annual as well as in a of the Democritus University of Thrace, Greece, in 1981 and 1988 respectively.
long-term basis, by using all the available means (capacitors, She is currently Assistant Professor in the Department of Electrical and Com-
VRs, network reconductoring). puter Engineering of the Democritus University of Thrace, Greece. Her teaching
interests include power systems and electrical installations. Her research inter-
ests include power system planning and optimization and short-circuit losses
REFERENCES and forces in metal enclosed arrangements.
[1] G. Salis and A. Safigianni, Economically Justified Modification of a
Radial Network to an Optimal Form, Advances in Engineering Soft-
ware, vol. 21, no. 1, pp. 4965, 1994.
[2] G. Salis and A. Safigianni, Economical Optimization of a Radial Power
Network Based on Voltage Drop Criterion, Advances in Engineering George J. Salis was born in Zurich, Switzerland, in 1970. He received
Software, vol. 22, no. 1, pp. 120, 1995. his Dipl.-Eng. degree from the Electrical Engineering Department of the
[3] J. J. Grainger and S. Civanlar, Volt/Var Control on Distribution System Democritus University of Thrace, Greece, in 1992 and his Ph.D. degree from
with Lateral Branches Using Shunt Capacitors and Voltage Regulators, the Electrical and Computer Engineering Department of the Democritus
Part 1: The Overall Problem, IEEE Trans. on PAS, vol. 104, no. 11, pp. University of Thrace, Greece, in 1998. His research interests include computer
32783283, November 1985. application to distribution system planning and optimization.