Professional Documents
Culture Documents
bound is greater than the actual cost of the best feasible solution
N = IUJ so far found, then it is not necessary to actually search the decision
a, = supply of power at node i for i e N alternative given by that branch or any decision subsequent to it.
b = demand for power at node j for j e N
unit "cost" of sending power from node i to node j For a branch and bound method to be successful in solving
cii =
large problems, it is necessary that the bounding method be
where i, j e N sufficiently good to eliminate most of the potential branches on
x,1 = amount of power actually sent from i to j, where the search tree. Otherwise, the tree will grow to be so large that
i, j eN excessive computer time will be required. Fortunately, the two
U = upper bound power flow from node i to node j, bounding methods to be described next have proved in practice
to be quite effective in reducing the amount of search to a
where i, j eN computationally feasible level for the model considered here.
F,, = fixed charge if substation i is opened
0 if substation i is closed, i I c 4.1 Minimum Incremental Cost Bound
yis = To calculate the minimum incremental cost (cost associated
1 if substation i is open, i e I with the feeder network), we set the capacity of all substations
(actual or potential) to maximum capacity and solve the transporta-
There are two special conditions the c 's satisfy. First, tion problem (1 ) - (4) with all Fs, = 0. The answer gives the smallest
variable cost way of meeting power demands.
C11 = 0 for all i e N,
Given any set of potential substations which it is proposed
that is, the "cost" of retaining excess power at node i is zero to open, a lower bound on the cost of providing power demands
for all i. Correspondingly, the variable x11 should be interpreted as using the existing substations together with this set of potential
a slack variable. Second, if there is no arc connecting nodes i and j substations can quickly be calculated as the sum of the fixed costs
then ci, = 00. In fact, such infinite costs are never entered as data. of the potential substations plus the minimum incremental cost.
The mathematical form of the fixed charge transshipment 4.2 Shortest Path Customer Assignment
model can now be stated as: To calculate a lower bound on the cost of serving a customer
if only one of the potential substations, say substation s,, is open
Minimize N 2; c x ++ IYisFfs (1) we solve the transshipment problem (1) -(4) with (a) all existing
lEN leN J
stations open at infinite capacity, (b)- substation s, open with
infinite capacity, (c) all fixed charges set to 0, and (d) all feeder
Subjectto E x =a1+Kfor6rieN (2) segment upper bounds set to + oo. Then the linear programming
dual variables for each demand location are obtained from the
corresponding optimal linear programming solution. These dual
X =bj+KforjeN (3) variables represent the marginal cost of supplying one more unit of
power to the corresponding demand locations in this station
configuration; they are lower bounds on the marginal cost of
0 xU S U1j for is j c N, j 0 s (4) supplying an extra unit of power in any more constrained situa-
tion. Suppose now we have potential substations s1, s 2 . e 'Sk all
.
go to (10).
(4) (Branching step.) Let s* be the first element of Fk (the
one having the smallest weak lower bound); set
Sk Sk
= U {s* and Fk FFSk 1
(5) (Bounding step.) If Sk has enough station capacity
for a feasible solution, and WLB(Sk) < BC, go to (7).
Else go to (6).
(6) k = k+ l.Goto (l).
(7) Solve the transshipment problem defined by opening
all substations in Sk .
(8) If the new solution has lower cost than the best
solution so far found, save it and update BC.
(9) If the new solution variable cost minus the minimum
incremental cost is less than the smallest fixed cost
of any station in Fk, go to (3). Else go to (6).
(10) Stop. The lowest cost feasible solution found is
optimal.
266ASCR 2/OC
266A 2C
5.0 SAMPLE PROBLEM 3/OA 4C
Figure 1
In order to clarify the branch and bound process, suppose we
consider a small sample problem. The sample problem is taken
from 1 ] and consists of 41 demand points and 3 potential station Table I Substation Data
locations from which to choose. The diagram in Figure illustrates
the layout of the' feeder network and ea'ch potential'station loca- Station No. Capacity Cost
tion. The capacities and costs of the potential'stations are indicated (KVA) ($1000)
in Table 1. 1 Maryvale 4 41,900 1,205
2 Christy 34 20,970 475
The branch and bound process begins by first calculating the 3 Cartwright 22 39,700 1,200
minimum variable cost of the feeder network configuration
assuming all possible stations are in service. This network variable The number above each vertex is the value of the WLB for
cost is a lower bound on the variable cost associated with any
combination of station locations. Of course, if this first problem system configuration represented by the vertex. A solid branch
has no feasible solution then there is no feasible solution for the indicates that the station has been selected from the set of free
location problem and the algorithm is terminated. stations and an arrow pointing to the right along the tree indicates
a station is opened in the solution while an arrow pointing to the
For the sample problem, the lower bound on the network left indicates that the station is closed in the solution. A dotted
variable cost is 515. The algorithm next calculates a shortest path line indicates that the station is a free station which will have to be
table used in calculating weak lower bounds (WLB). considered later.
1
1
4%1
_I, 2356 [32621
The system configuration at vertex 3 has adequate capacity
to meet the system load and so the full capacitated linear program
formulation of the system is solved and the actual total cost is
enclosed with the box in the above diagram (note that the cost of
the configuration is quite close to the WLB estimate). This cost At vertex 7 there is insufficient capacity with no more
(3395) now forms the current minimum CB for the location free stations available and thus we backtrack to vertex 0 where
problems. the branch and bound search process terminates because the set of
free stations is exhausted.
At this point station 3 is taken out of the solution by "back-
tracking" to vertex 2 which has no free stations remaining, and so The optimal solution is represented by vertex 6 and the
again we backtrack to vertex 1. The one remaining free station at minimum cost is 3262 for the sample problem, which is obtained
vertex 1 has inadequate capacity to meet the system load and so by opening substation 1 and 3.
we again backtrack to vertex 0. At vertex 0 we select from the
remaining free stations to bring into the solution. Station 3 is 6.0 OUTPUT AND COMPUTATIONAL RESULTS
selected and the branch and bound search tree appears below. The results obtained from solving this model are the optimal
new substation locations together with essentially the same
3378 information as that is obtained from solving our previous model
2324 [1]. Tabular output summaries by demand location, station and
total system are given of power flows, costs of losses, capital cost
1387 of feeders and stations, and finally voltage drop in percent. All
new stations used are indicated with an asterisk in the output.
A printer plot of the existing station and new station service areas
is provided along with a graphical display of the feeder network
associated with the optimal solution. The printer plot indicates
the center of load of each station service area which is useful in
deciding on new potential station locations.
The computational results for the problems attempted so far
4%
%.
2356 are summarized in Table II and are quite promising.