Electrical Power & Energy Systems, Vol. 18, No. 7, pp.

469-474, 1996
Copyright © 1996 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
ELSEVIER P I I : S0142-0615(96)00005-1 0142-0615/96/$15.00 + 0.00

A modified linear programming

method for distribution system
reconfiguration
Ali A b u r
Department of Electrical Engineering,
Texas A&M University, College Station, TX 77843,
USA

large number of permissable ones among which one

This paper presents a reconfiguration method for radial
needs to pick the optimal configuration that results in
distribution systems. The reconfiguration problem is
minimum loss. Efficiency of this search is the key to the
formulated and solved using a modified linear programming
success of any proposed technique.
approach. The modifications are introduced mainly to
The reconfiguration problem has been proposed to be
maintain a radial network configuration while enforcing
formulated and solved in various different ways in the
the lineflow capacity limits. The implementation details are
explained using a smrrll tutorial example. Simulation past. One approach is to start from a meshed con-
figuration where all switches are closed, and to find the
results for sample distribution systems are also given.
Copyright © 1996 Elsevier Science Ltd optimal radial configuration by opening one switch at a
time 1. An approximate formula for estimating the change
Keywords: distribution systems, mathematical program- in losses each time a switching takes place, is developed in
ming, utility application Reference 2. Operational constraints are taken into
account while balancing the feeder loads by using
different search techniques in References 3 and 4.
Reference 5 studies the optimality conditions for
I. I n t r o d u c t i o n minimum loss reconfiguration problem when using
Efficient operation of power distribution systems can be distributed and lumped load models. An efficient power
achieved by reconfiguring the system to minimize the flow algorithm for distribution systems is developed and
losses as the operating conditions change. Reconfigura- incorporated into the reconfiguration problem solution
tion has to be carried out by taking into account various in Reference 6. Heuristic search techniques are employed
operational constraint,; such as line and transformer in Reference 7. Loss reduction and load balancing
ampacity limits, voltage limits, radiality of the resulting problems are formulated as a multi-objective, con-
system configuration, etc. strained and non-differentiable optimization problem
Reconfiguration is done by opening and closing and solved using a modified simulated annealing
switches in order to modify the amount of power flow algorithm in References 8 and 9. Reference 10 formulated
through the feeder sections. Given a network containing n the problem as a trans-shipment problem with quadratic
switches, there will be 2n possible configurations costs. An efficient heuristic method for loss reduction is
corresponding to the open and closed positions of all described in Reference 11. Reference 12 applies a genetic
the switches. Some ef these configurations are not algorithm to solve the minimum loss reconfiguration
permissable because tl~ey yield either a disconnected problem.
system with several islands, or a non-radial configuration. In this paper, the reconfiguration problem is
Distribution systems are configured radially in order to formulated as a minimum cost network flow problem.
comply with the protective relaying requirements. Such problems can be solved using linear program-
Certain other configurations may not be feasible due to ming as long as the line capacity limits are ignored. In
the violation of operational constraints. Despite all these order to enforce these limits, the revised Simplex
non-permissable configurations, there still remains a algorithm is modified and the details of its implementa-
tion are presented along with examples. The resulting
Received11 May 1995; revised13 August 1995; accepted4 January algorithm yields a radial configuration which does not
1996 violate any line capacity limits, while minimizing the losses.

469
470 Modified linear programming method. A, Abur

II. P r o b l e m f o r m u l a t i o n GENERATION LOAD

Consider the power balance equations where the system is Xl
assumed to be lossless: ~ 6
[A][Pf]=[P] (1) x2
where [A] = node-to-branch reduced incidence matrix,
[Pf] = vector of all branch flows, and [P] = vector of
Figure 1. Two bus example system
nodal injections, excluding the reference node.
If the system is fully connected and has N nodes, then
the column rank of the coefficient matrix A in equation x2
(1) will be N - 1. Hence, for a system with L branches,
i.~"-~ CAPACITY LIMIT [
only N - 1 branches will carry non-zero flows, the rest of
them will represent open circuits. Depending on the 6
choice of these N - 1 closed branches, the solution of ' • • . ~ . . . - - - - I OPTIMALSOLUTION
5.
equation (1) will change, yielding a different operating "'••• [ J* = 11.0
condition. The radial configuration of the system will be 4_
maintained as long as each time an open branch is closed,
an already closed branch is opened. 3_ •• POWER BALANCE
A new objective function J(Pf) will be considered EQUATION
where, J(Pf) is a weighted sum of the absolute power 2.
flows through all the network branches. The weights 1.
represent the penalties (or cost) of transporting a unit
of power over the respective branches. While other I I f I •'t I v

choices are also possible, in this study this cost is set as 2 3 4 5 6

equal to the resistances of the associated branches. x1
In order to determine the optimal radial configuration
while enforcing the line capacity limits, the following J= Xl+2X 2
optimization problem will have to be solved:
Figure 2. Graphical illustration of LP solution
L I1.1 Example 1
Minimize J = ~ 0dilef(i)[ (2)
Consider the two node, two branch system shown in
i=1
Figure 1. Denoting the branch flows by xl and x2, and
Subject to [A] [Pf] = [P] (3) assuming that the resistances of the branches are 1 and 2
per unit, respectively, and the flow through branch 1 is
[E] [[Pfl] ~<[pu] (4) limited to 1.0 per unit, the corresponding LP problem can
be written as below:
were ~Mi is the cost of transmitting per unit power along
branch i. This cost is set equal to the resistance of the Minimize J = Xl + 2x2 (8)
corresponding branch, thus penalizing high flows Subject to Xl + x2 = 6 (9)
through branches with higher resistance, pu is the
vector of branch power capacity limits, and E is a x I ~<1 (10)
binary matrix relating the branch flows and their limits.
Let Pf(i) = X i - Yi, such that both Xi and Yi are X1,X2 >~0. (11)
non-negative variables which cannot both be non-zero The optimal solution is shown graphically in Figure 2.
simultaneously. The above optimization problem can The objective function has a value of 11.0 at the optimum
then be written as an equivalent linear programming and both branch flows are non-zero, xl = 1.0, x2 = 5.0,
(LP) problem as below: i.e. the system configuration is not radial. The non-radial
solution is obtained due to the fact that one of the branch
L flows has to be constrained at its capacity limit.
Minimize Z ~oi(Xi+ Yi) (5)
i=1
III. Proposed m e t h o d
Subjectto [ ; -A][;I ~ [:u ] (6) A new problem formulation and its solution method will
now be described. The formulation will incorporate the
line capacity limits and the proposed solution method will
[x],[v] 0. (7) ensure a radial system configuration within these limits if
one exists.
The main difficulty in this formulation is how to Let us rewrite the LP problem given by equations (5)-
maintain the radiality of the system while enforcing (7) by introducing slack variables:
the line capacity limits. Attempting to solve the above
LP problem using the well-known Simplex method Minimize WzT x Z (12)
will not yield a radial system if at least one of the
line flows hits its capacity limit. In this case, the Subjectto [ H ; ] [ Z ] = [:u] (13)
regular LP solution will correspond to a system contain-
ing loops. The following simple example will be used to
illustrate this point. Z,s ~>0 (14)
 Modified linear programming method: A. Abur 471

where U = the identity matrix, W f = [wl ... WLWl ... w/,], After solving the LP problem given by equations (15)-
Z T = [xTyT], S = vector of slack variables, and (17) using the modified Simplex method:
• If the objective function equation (15) is still non-zero,
this will imply that a radial system configuration
cannot be found within the line capacity constraints.
In order to ensure that the corresponding system Hence, the problem will be declared to be infeasible.
remains radially configured, only (N - 1) columns of It • If the problem is feasible, then the optimal solution will
in equation (13) must remain in the basis. This will be the yield a zero objective function and the initial basis for
case if and only if all columns of U (corresponding to the the subsequent Phase II solution. This initial basis will
slack variables s) are forced to remain in the basis as well. contain the columns of all the slack variables and will
Hence, if the LP problem of equations (12)-(14) can be look like:
initialized by choosing a basis which contains all the
columns of the slack wtriables s, and during the subse-
quent Simplex iterations if these slack variables are not
allowed to leave the basis, then the converged solution
,:[.l 1
Here, the columns [Da/0 ] of the starting basis are all
will yield the best possible radial configuration.
replaced by a subset of the columns of It, denoted here
Therefore, the overal.1 algorithm will be composed of
by the submatrix a 1.
two phases:
(1) Phase I: Initialization; 111.2 Phase I1: optimization
(2) Phase II: Optimization. Let us assume that an initial basis that includes the
columns of all the slack variables, has already been
The two phases of the proposed algorithm will be obtained in Phase I, as described in the above subsection.
described in detail below. In Phase II, the LP problem of equations (12)-(14) will
Ilia Phase I: Initialization and feasibility check then be solved, again using the modified Simplex method.
The objective function is the only change needed to switch
The LP problem of equations (12)-(14) can be initialized from Phase I to Phase II. The objective function of Phase
by introducing artificial[ variables Xa. The reason behind I given by equation (15) will be replaced by equation (12).
the introduction of these artificial variables x a, is to create The solution will be found when the modified Simplex
an artificial starting basis which contains all the slack method converges.
variables s. The columns corresponding to the artificial
variables will be driven out of the starting basis one by 111.3 Modified Simplex algorithm
one and will be replaced by a subset of the columns of H
The modified Simplex method will be described for the
during Phase I. In order to accomplish this, the following Phase II problem, with the understanding that by a
LP problem will be set up and then solved:
simple change of objective function and initial basis, it
N-I can be directly applied to Phase I as well.
Minimize ~ MXa(i) (15) Assuming that Phase I initialization is successful, the
i=l constraint equations can be written in terms of the basic
x a and non-basic x o variables as shown below:

I
Subject to t t O a 1U
° li =['] pu
(16) [. 0] iz, ,18,
B xB b
Z, Xa, s 3 0 (17)
Here, Zl and Z 2 represent the basic and non-basic
where M is a large, positive weight given to the artificial branch flow variables, respectively. Note that as long as
variables, while all the other variables Z, s are assigned all the slack variables remain in the basis during the
zero weights. Simplex iterations, i.e. the columns leaving B are always
The starting basis will be chosen as: picked from those corresponding to Z1 variables, then the

[':a 0][':] [;. 1 resulting system configuration will remain always radial.
In order to ensure this, the well-known Simplex algorithm
for solving linear programming problems 13, will have
to be slightly modified. The proposed modifications
where Da is a diagonal matrix whose entries are given by:
in the Simplex iterations will be described using the
1.0 if P(i)>~0 ] . notation of equation (18). The steps of the algorithm
Da(i, i)--- are given below:
k -1.0 otherwiseft=l'""N-l"
(1) Compute the basic solution xB = B-lb.
The LP problem of equations (15)-07) is solved using (2) Set the counter k = 1.
a modified Simplex method. The modification of the (3) Calculate the relative costs of non-basic variables, rD:
Simplex logic is rather simple and involves a restriction
on the selection of the columns to leave the basis at each r T = cT _ c~B-1a
Simplex iteration. The columns that correspond to the
slack variables will not be allowed to leave the basis. A Sort the entries of r D in ascending order as:
detailed account of how this can be implemented in the
form of an algorithm is given in Section III.3 below. rD(il) ~rD(i2) ~ ' ' " ~rD(iL)
472 Modified linear programming method." A. Abur

(4) IfrD(ik)/>0, stop, XB is the final solution• Else, go to .~.~ CAPACITYLIMIT [

the next step. x2
(5) Solve By = Dik, where Di~ is the ikth column of D. i I OPTIMALRADIALSOLUTION
(6) If y(j)~<0 for all j, then the solution is unbounded• J* = 12.0
Else, find
• xB(j)
mln-- fory(j) > 0.
J y(j)'
(7) I f j t h variable of the basis is a slack variable, then POWERBALANCE
increment the counter k = k + 1 and go to step (4).
Otherwise, go to the next step.
EQUATION
(8) Update the basis B, by replacing thejth column of B
by Dik. Go to step (1).

3.4 Example 2 I I I I •I
Let us apply the above described algorithm to the 2 3 4 5 6 x!
solution of Example 1, presented earlier. Introducing
the slack variable, s, the problem can be rewritten as: [J=Xl+2X2 [
Minimize Xl + 2x2
Figure 3 Graphical illustration of the radial solution
Subject to Xl + x2 -- 6
XI+S=I If the objective function is not zero, then declare the
problem infeasible and stop. There is no radial
Xl~X2~S >jO configuration possible that will satisfy all the line
Assume that Phase I is executed and the initial basic capacity constraints•
feasible solution is obtained as (2) Starting with the initial basic feasible solution
obtained at the end of Phase I, solve the LP problem
Xl = 0,x 2 = 6,s-- 1 given by equations (12)-(14) again using the modified
i.e. the basis columns are chosen as the ones correspond- Simplex algorithm described in Section 111.3. The
ing to variables x2 and s. branches whose corresponding flow variables are
Then the steps of the algorithm for the Phase II basic at the solution will be the closed branches•
solution will be executed as follows: The non-basic branch flow variables will indicate
the open branches with zero flows through them.
ek=l.
• There is only one non-basic variable Xl and its relative
cost i S r o ( 1 ) = l - - 2 = - - l . V. Test results
• Relative cost is negative, so continue. A computer program which executes the proposed
• Solving for y = [1 1]r. algorithm is developed. A 16 bus test system, whose
• min(XB(l)
\ y, ' xB(2)"
y2 )I = ( T6' T 1
)=I's°j=2" one-line diagram is shown in Figure 4, is used for the
simulations. The branch data and bus loads for this
• But the second variable Xa (2) is a slack variable, so k is system are given in the Appendix. The developed
incremented by 1, k = 2 . Go to step (4) of the program only uses the real power injections and the
algorithm. branch resistances in the optimization procedure. Once
Since D has only one column, the kth (second) column the optimal configuration is determined by the program,
does not exist. Thus, the algorithm terminates at the a power flow for the corresponding configuration is
fourth step with the desired radial solution: executed to determine the actual losses for that

x;=0.0, x~=6.0
A graphical illustration of this is given in Figure 3. In
this case, the modified algorithm settles at a solution
which not only satisfies the line capacity constraint but
also the radiality of the network. Note that the optimal
objective function value is 12.0 which is higher than the
one corresponding to the non-radial solution.

IV. Summary of the proposed method ,j

The proposed two phase solution method can now be
summarized as below:
14
(1) Solve the LP problem given by equations (15)-(17)
using the modified Simplex method described in
Section Ill.3.
If the objective function is zero at the solution, then
continue to step (2). Figure 4. 1 6 Bus Test System
 Modified linear programming method."A. Abut 473

Table 1. Unconstrained ease results

Configuration Base case Optimal configuration
Open branches 3-8, 5-12, 6-13, 9-15, 10-16, 10-12 2-3, 8-9, 9-10, 11-12, 12-13, 10-12
Losses 14.2 MW 10.0 M W
Minimum bus voltage 0.918 0.952

Table 2. Constrained ease results

Configuration Case 1 Case 2

Open branches 2-3, 9-10, 9-15, 11-12, 12-13, 10-12 7-8, 8-9, 9-10, 11-12, 12-13, 10-12
Constraints P1,2 ~<2.0 Pl,7 ~ 2.0
P8,9 ~<1.0 P2,3 ~<3.0
P14,15 ~ 1.0 P3,11 ~<3.0
Pll,12 ~<3.0 P7,8 ~<1.0
P10,12 ~<1.0 Pl0,16 ~ 1.0
P10,12 ~ 1.0
PlI,12 ~<3.0
Losses 10.1 MW 19.3 MW
Minimum bus voltage 0.951 0.913

configuration. Voltage constraints are not taken into where voltage constraints are also accounted for in the
account in this work. optimization process, is currently under study.
The base configuration is given in Table 1. Only the
open branches are indicated in the table in order to
describe the system configuration. Initially, no capacity VII. A c k n o w l e d g e m e n t s
constraints are introduced and the program is executed to The author would like to thank Drs Dariush Shirmo-
find the optimal reconfiguration. This unconstrained case hammadi, Carol Cheng, Qin Zhou and Edwin Liu of the
resulted in the configuration and losses given in Table 1. Pacific Gas and Electric Company for the valuable
Note that even though voltage constraints are not used, discussions during this study.
minimum loss reconfiguration improved the bus voltage
profile as can be observed from the minimum bus voltage
given for the two cases. VIII. References
Two constrained ca:ses are presented in Table 2. The 1 Merlin, A and Back, H 'Search for minimum loss operating
constraints are satisfied in both cases, while losses are spanning tree configuration in urban power distribution
attempted to be miniraized. In the first case, the losses systems' Proceedings of 5th Power Systems Computations
remained almost at their optimal value as in the Conference, UK (1975)
unconstrained case. The second case which contains
2 Civanlar, S, Grainger, J J, Yin, H and Lee, S S H 'Distribu-
more stringent limits resulted in a configuration yielding tion feeder reconfiguration for loss reduction' IEEE Trans.
a much higher loss. The voltage profile is accordingly Power Delivery Vol 3 No 3 (1988) 1217-1223
deteriorated as can be seen from Table 2. In order to test
the program's ability to detect infeasible cases, the capa- 3 Aoki, K, Kuwabara, H, Satoh, T and Kanezashi, M 'Outage
city limit on branches 1-7 is reduced down to 0.5 p.u. in state optimal load allocation by automatic sectionalizing
switches operation in distribution systems' IEEE Trans.
the second case. The program was unable to initialize the Power Delivery Vol 2 No 4 (1987) 1177-1185
LP problem returning an infeasibility flag instead.
The program is altso tested using a realistic size 4 Aoki, K, Kuwabara, H, Satoh, T and Kanezashi, M 'An
distribution system having 474 buses and 496 branches. efficient algorithm for load balancing of transformers and
The tests are executed using a Sun Spare 1000 using the feeders by switch operation in large scale distribution
systems' IEEE Trans. on Power Delivery Vol 3 No 4
Solaris 2 . 3 0 / S and the recorded cpu time was 49.6 s for (1988) 1865-1872
this test system.
5 Liu, C C, Lee, S J and Vu, K 'Loss minimization of
distribution feeders: optimality and algorithms' IEEE
VI. C o n c l u s i o n s Trans. Power Delivery Vol 4 No 2 (1989) 1281-1289
A linear programming based algorithm is presented for 6 Baran, M E and Wu, F F 'Network reconfiguration in
distribution system reconfiguration with minimum loss. distribution systems for loss reduction and load balancing'
The proposed method does not take into account the IEEE Trans. Power Defivery Vol 4 No 2 (1989) 1401-1407
voltage constraints, but accounts for the line capacity 7 Morelato, A L and Monticelli, A 'Heuristic search approach
limits while maintaining a radial system configuration. It to distribution system restoration' IEEE Trans. Power
can also determine the infeasibility of a given problem Delivery Vol 4 No 4 (1989) 2235-2241
during the initialization of the linear programming solu- 8 Chiang, H D and Jumean, R J 'Optimal network recon-
tion. The suggested modifications can be implemented in figurations in distribution systems: Part 1: A New Formula-
an existing LP program easily, as described in this paper. tion and a solution methodology' IEEE Trans. Power
The natural extension of the presented method to the case Delivery Vol 5 No 4 (1990) 1902-1909
474 Modified linear programming method." A. Abur

9 Chiang, H D and Jumeau, R J 'Optimal network recon- Load data

figurations in distribution systems: Part 2: Solution
algorithms and numerical results' IEEE Trans. Power
Delivery Vol 5 No 3 (1990) 1568 1574
D0 Glamocanin, V 'Optimal loss reduction of distribution net-
works' IEEE Trans. Power Syst. Vol 5 No 3 (1990) 774-782 Load (pu)
11 Shirmohammadi, D 'Service restoration in Distribution Bus number P O
networks via network reconfiguration' IEEE Trans.
Power Defivery Vol 7 No 2 (1992) 952-958 1 0.000 0.000
12 Nara, K, Shiose, A, Kitagawa, M and Ishihara, T 2 0.330 0.180
'Implementation of genetic algorithm for distribution 3 0.410 0.210
systems loss minimum reconfiguration' IEEE Trans. 4 0.200 0.100
Power Syst. Vol 7 No 3 (1992) 1044-1051 5 0.130 0.040
13 Luenberger, D G 'Linear and nonlinear programming' 6 O.05O 0.020
Addison-Wesley (1984) 7 0.400 0.200
8 0.500 0.220
9 0.210 0.160
Appendix 10 0.050 0.020
11 0.250 0.150
Branch data 12 0.770 0.240
13 0.180 0.023
14 0.105 0.053
15 0.220 0.050
16 0.430 0.030
From To R (p.u.) X (p.u.) B (p.u.)*
1 2 0.0083 0.0080 0.0645
2 3 0.0098 0.0150 0.0409
3 4 0.0012 0.0066 0.0190
4 5 0.0025 0.0040 0.0129
5 6 0.0030 0.0080 0.0174
1 7 0.0020 0.0040 0.0138
7 8 0.0039 0.0050 0.0235
8 9 0.0099 0.0175 0.0274
9 10 0.0069 0.0099 0.0220
9 15 0.0058 0.0098 0.0109
7 14 0.0048 0.0080 0.0386
14 15 0.008l 0.0180 0.0203
15 16 0.0032 0.0074 0.0055
3 11 0.0069 0.0169 0.0115
11 12 0.0078 0.0110 0.0494
12 13 0.0054 0.0160 0.0273
5 12 0.0038 0.0080 0.0143
6 13 0.0062 0.0130 0.0272
10 16 0.0002 0.0041 0.0062
10 12 0.007l 0.0147 0.0074
3 8 0.0024 0.0051 0.0134
*½line charging susceptance.