Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

MICROPR(X:ESS(~SAND

MICROSYSTEMS
ELSEVIER Microprocessors and Microsystems 20 (1996) 251-258

Effective algorithms for designing power distribution networks


Y.K. Wong, K.L. Shi, T.F. Chan
Department of Electrical Engineering, Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kong
Received 15 September 1995; revised 19 February 1996

Abstract

In transmission planning, we need a good network design to minimize the construction cost and to determine the optimal location
of substations. This paper presents a feasible design method, using two mathematical techniques (Dijkstra's algorithm and Trans-
portation algorithm) to optimize substation sizes and service boundaries, given alternative locations for the substations under
reliability constraints. A reliability index, the Average Annual Customer Interruption Rate (AACIR) is used to determine the
reliability of loads. Appropriate measures can then be taken to increase the reliability of the system'.

Keywords. Distribution network; Optimization; Dijkstra's algorithm; Transportation algorithm

1. Introduction method [3] is used to determine the reliability index and


the configuration of the system. The network design
In a power system, the distribution network is very therefore meets reliability condition while satisfying the
important and the selection of the routes will greatly minimum cost criterion.
affect the construction cost and reliability. In a large To implement those algorithms, a program,
system, some effective algorithms have to be employed R O U T E . E X E has been developed for designing optimal
in order to find the optimal routes. Moreover, as the load source to load distribution networks. The power system,
demands increase, new substations should be added. We China Light & Power Company of Hong Kong is used
need to decide where to locate the substations in order to as a case study. In designing a network, some other con-
have maximum cost effectiveness. Although computer straints have to be considered by planner at the same
algorithms cannot give out a detailed configuration of time, such as the environmental, geo-technical, engin-
the distribution network, they can present quickly an eering, construction and land resumption. As a result
optimal solution based certain simplifying assumptions. of calculation, the optimal network design has been
This optimal solution is a very significant reference for obtained.
the initial stage of designing a power distribution net-
work. So, the planner can spend more time to process
informal tasks, e.g., obtaining the geographic data and 2. Optimization algorithms
load forecasting. In this paper, the problem encountered
is the planning of power distribution substation, sizes, The optimization algorithms include (1) Dijkstra's
and service boundaries. For optimization, two techni- algorithm, (2) the Transportation algorithm and (3) the
ques, Dijkstra's algorithm [1] and Transportation algo- reliability constraints (Average Annual Customer Inter-
rithm [2] are employed because of their computational ruption Rate method).
efficiency. The analysis leads to a preliminary view of
substation configurations with minimum construction 2.1. Dijkstra's algorithm
cost as well as transmission and distribution expenses.
Reliability is another important factor in distribution Power distribution network design can be formulated
system design. In order to provide good service to con- as a shortest-route problem that considers the determi-
sumers, a distribution system must have a high level of nation of the minimum cost (distance) from an origin to
reliability. In this paper, the Average Interruption Rate a destination through some connecting graph [4]. One
0141-9331/96/$15.00 © 1996 Elsevier Science B.V. All rights reserved
PH S0141-9331 (96)01088-5
252 Y.K. Wong et al./Microprocessors and Microsystems 20 (1996) 251 258

powerful solution technique of the shortest-route prob-


lem is Dijkstra's algorithm. Referring to an appraisal of
algorithms by Dreyfus [5], the Dijkstra's method is
superior because it depends more on the number of
arcs than nodes. Especially with 25 percent or more of
the N ( N - 1) possible arcs present, Dijkstra's algorithm
out-performs other algorithms (N is the number of
nodes), hence it is chosen for solving the present design Fig. 1. Network example for illustrating Dijkstra's algorithm.

problem of distribution network. node. Note that if the old temporary label is still mini-
Dijkstra's algorithm determines the shortest chain mal, it will remain unchanged.
from the source to all other nodes in a network. Each Step 2. Select the minimum temporary label and
arc (i, j ) of the network has a positive cost (distance) cij declare it permanent. If node k is given a permanent
associated with it. If there is no arc from node i to k, the label, the algorithm is terminated. Otherwise, return to
cost cik is infinite. The algorithm does not require that step 1.
costs between a pair of nodes are symmetric (i.e., As an example, let's consider the network shown in
cjk :/: ckj is possible). If there is an edge between nodes Fig. 1. The distances associated with the arcs and
i a n d j in a network, it is replaced by two arcs (i,j) and nodes are indicated. By applying Dijkstra's algorithm
(j, i) with costs cij and cji equal to the costs associated to the network, it can be shown that the shortest path
with the edge. between nodes 1_.and 8_is the route 1 - 2 - 5 - 7 - 8 .
Let us assume that the cost or distance between any
two nodes is a non-negative number represented by c~;. 2.2. Transportation algorithm
The algorithm assigns to every node a label which is
either temporary or permanent. Initially, each node The objective of the transportation model is to select a
except the source node receives a temporary label minimum cost routing plan for transporting a single
which represents the direct distance between the source commodity from a number of sources to a number of
node and that node. The source node is assigned a per- destinations [6].
manent label of zero. Any node that cannot be reached There are m source points in a network, where source i
directly from the source node is assigned a temporary has a supply of a i units of a particular item (commodity).
label of c~, while all others receive temporary labels of In addition, there are located n destination points, where
c~j, j :/: s. When it is determined that a node must belong destination j requires bj units of the commodity. We
to the minimal path, its label becomes permanent. The assume that a i > 0 and bj > 0. Associated with each
algorithm operates on the simple logic that if a shortest link (i, j), from source i to destination j, there is a unit
path from node s to n o d e j is known and node k belongs cost c,7 for transportation. The problem is to determine
to this path, then the minimal path from s to k is the the feasible shipping pattern from sources to destinations
portion of the original path ending at node k. The algo- that minimizes the total transportation cost.
rithm starts withj = s and successively resetsj untilj = k Let xz; be the number of units shipped along link (i, j )
and then the process is stopped. from source i to destination j. Further assume that the
For a given node j, the cost Lsk is used to represent an total supply equals the total demand, that is
estimate of the length of the shortest path from the ~"~m=1 a i = ~7=1 bj. If the total supply exceeds the total
source node s to node k. When this estimate cannot be demand, then a dummy destination can be created with
further improved, we call it a 'permanent label' and the demand b,+l = ~m=l a i - - ~--~7=1b j , and Ci, n+ 1 = 0 for
cost is represented by LIsk. Otherwise, it is referred to as a i -- 1 , . . . , m. Assuming that the total supply equals the
'temporary label'. Initially, only the source node is per- total demand, the linear programming model for the
manently labeled. Every other label is temporary and transportation problem may be formulated [2] as:
contains the direct distance from the source to the corre-
sponding node. To identify the node that is closest to the Minimize Z = ~ ~ £ijXij subject to
source node, we select the minimum of the temporary i=1 j=l
labels and declare it permanent, from this point on the
algorithm cycles in two phases, until the sink node is ~'~ Xi/ ~ ai, i= 1,2,...,m (1)
permanently labeled. j=l
Step 1. Inspect the remaining temporary labeled
nodes. Compare each temporary label already assigned ~ xij = bj, j = 1 , 2 , . . . , n x~j > 0 , integers
with the sum of the last permanent label and the direct j=l
distance from the node with the permanent label to the m n
node under consideration. The minimum of these two Z;ai:Zbj
distances is defined as the new temporary label for that i=l j=l
Y.K. Wong et al./Microprocessors and Microsystems 20 (1996) 251-258 253

Supplies i ClI
Demands As an illustration of applying to the transportation
problem, an initial solution by the northwest corner
rule is shown in Table 2,
Using formula 1, the total cost Z1 = 35 100.

• Optimal assignment
Because there are different total costs of assignment
for various arrays of supplies and demands by the North-
Fig. 2 Graphical structure of the transportation problem. west corner method, the optimal assignment is needed to
find the least total cost. Changing the permutation of the
A graphical structure of the model to be formulated is array {ai, bjli=l,...,m;j=l,...,n} , a n d repeating the northwest
shown in Fig. 2. Input information required for the for- corner method and the transportation formulation (for-
mulation of the model is displayed in a rectangular array mula 1), we can get a set of total c o s t {Zk[k=l,Z,...n!xm!/2 }.
which contains the cost coefficients for all possible When the total supply equals the total demand, the
routes, and the supply and demand figure for the produc- northwest corner method has a property that the
tion and demand contain, respectively. As an illustra- assigning results are the same for any permutation
tion, such an arrangement is shown in Table 1 for the [al,...amlbl...bnl and its reversal symmetric permu-
case corresponding to Fig. 2, where x q is the number of tation [ a m , . . . a l l b n . . . b l ] . In a pair of symmetric per-
units to be shipped from source i to destination j. mutation only one should be selected, so the total
Because demands are usually more than supplies number of permutations is equal to n!~m!. Hence, by
(n > m) in power systems, the northwest corner method taking Zmin = m i n { Z k } , we can get the minimum total
is used in the paper to minimize the total cost of power cost. For this example, Zmin = 28 220 and the assignment
assignment, according to formula 1. is shown in Table 3. An optimal transportation result has
been thus obtained.
• Northwest corner method
Begin by assigning as much flow as possible to the 2.3. Reliability-average interruption rate method
route linking the first source with the first terminal. Let
Xll be the number of units assigned. If al = bl, delete We know that a reliable power system is very impor-
both sources 1 and terminal 1, and select the route cor- tant in providing an uninterrupted supply to the con-
responding to x2> Otherwise, delete the source of term- sumers. Therefore, in this paper, the reliability is
inal whose supply or demand is satisfied, and select either considered as a constrained condition of the optimiza-
x12 or x21, depending on which one corresponds to a tion algorithms.
route that is feasible. The Average Interruption Rate method [4] is quite
Assign as much flow as possible to the selected route, straightforward and can be applied to a relatively com-
and remove the source or terminal whose supply or plicated network. It provides a measure of continuity
demand is also satisfied. If the supply and the demand rather than service quality by examining the simul-
conditions are simultaneously satisfied, both the source taneous conditions that must exist for power flow in
and the terminal are removed. Continuing in this fash- series and parallel combinations of system components.
ion, ifxij is the last basic variable selected, the next one to The application is based upon four principles.
be considered is xi4+l if the source has any supply avail-
a) A component operates in only two states, available
able, is Xi+l,j if t h e j t h terminal has any unfilled demand,
and unavailable. Maintenance is not considered and
o r Xi+l,j+ 1 otherwise.
the probability of a component being unavailable is
Table 1 given by its forced outage rate p. If q is the availability
Input information for the transportation problem rate, then p 4- q = 1.0.
b) Component failures are assumed to be independent
bl b2 b3 bn and therefore the probability of simultaneous failures
is given by the product of the respective probabilities.
¢11 C12 Cl3 C1n
X12 X13
c) In a series system all components must be available
al XI1 Xln
I for power flow into a receiving point. The probability
a2 ¢21 [ ¢22 C23 C2n of success is the product of the availability probabil-
X21 X22 X23 X2n ities. For a two-unit system with outage rates Pl and
P2 and availability rates ql and q2:

am Cml L Cm2 Cm3 Cmn qs = qlq2


Xml Xm2 Xm3 Xmtt
Ps = 1 - qlq2 = P l +P2 - P I P 2
254 Y.K. Wong et al./ Microprocessors and Microsystems 20 (1996) 251-258

Table 2
Initial solution by the Northwest Corner Rule

bl b2 b3 b4 b5 b6

al L 19 49 [ 43 55 47 53 250

02 [ 2o 37 28 42 30 38 200

a3 19 33 32 40 70 35 400

150 80 100 200 190 130 850

a) Row 1 assignment

bj b2 b3 b4 b5 b6

a1 150 80 20 250 Row 1 satisfied

b) Row 2 assignment

bl b2 b3 b4 b5 b6

al 150 80 20 250
a2 80 120 200 Row 2 satisfied

c) Starting basic feasible solution by the northwest corner rule

bl b2 b3 b4 b5 b6

al 150 80 20 250
a2 80 120 200
a3 80 190 130 400 Row 3 satisfied

If pl and P2 are much less than unity, the product Pl P2 outage time divided by the total component exposure
can be neglected. The failure probability of a series time and is the probability of component outage exis-
system in this case is the sum of the element failure tence. To provide an indication of both outage frequency
probabilities. and outage duration, the definition can be modified to
d) In a parallel system, all paths must fail, if no power is indicate the probability of outage occurrence rather than
to flow into the receiving point. For a two-element outage existence. The forced outage rate is defined as:
system, the failure probability is the product of the
sum of the days upon which an outage "~
two component failure values.
of specified minimum duration occurred J
p=
Forced outage rate is defined as the total component sum of unit days

Table 3
Optimal solution by the Northwest Corner Rule (Zmi n = 28 220)

b6 b2 b4 hi b5 b3

al 53 49 55 [ 19 47 43
130 80 190 400

a2 38 [ 37 [ 42 [ 20 [ 30 28
10 150 90 250

a3 33 40 [ 19 l 70 _ _32
100 I00 200

130 80 200 150 190 100 850


Y.K. Wong et al./Microprocessors and Microsystems 20 (1996) 251-258 255
B C Table 5
Line 1 Line3 Average Annual Customer Interruption Rate
Supply LoadC
Line2
LoadB
Load B. Lines 1 and 2 6.85 x 10 4
Load C. Lines 1, 2, and 3 0.1006
Fig. 3. A small systemto illustrate reliabilitycalculations.

If component forced outage rates for different minimum rates for each line section can be obtained by correctly
specified duration is compiled, it is possible to predict the combining the failure's rates of the series and parallel
frequency of occurrence of this condition at any par- equipment configurations within each section.
ticular point in the system. It should be realized that Reliability index is constructed by the AACIR of each
outages can occur on the same day but not simulta- load and security index is constructed by the maximum
neously. This method assumes that all outages occurring failure allowed for each load. If the AACIR of load is
during one day are simultaneous outages. This can be larger than the security index, then one more transmis-
partially reconciled by considering that if two compo- sion line will be added in parallel, or the type of lines will
nents within an area are forced out of service during be changed. The cost of the new lines has to be deter-
the same day, the probability of simultaneous occurrence mined again, with repeated use of Dijkstra's and Trans-
is somewhat higher than that implied by absolute outage portation algorithms, until the AACIR of load is less
independence. This method is restricted to the evaluation than the security index.
of continuity at a particular point.
As an example, let's consider the system shown in
Fig. 3. The failure rates for each line section are shown 3. Program and case study
in the Table 4.
Considering the failures per year as the number of The above algorithms and methods are incorporated
days upon which failures occur within the year, the prob- in a microcomputer program, ROUTE.EXE, for the
ability of an outage occurring on lines 1 or 2 is given by: optimal design of distribution networks. The C + + lan-
guage is employed for the program development. Pro-
Pl = P2 = 0.5/365 = 1.37 x 10 -3 gram runs provide the user a preliminary view of
Similarly: power transmission routing and the distribution of
loads. Using the data of China Light and Power Com-
P3 = 0.274 x 10 -3 pany (CLP), a major power company in Hong Kong, a
The Average Annual Customer Interruption Rate case study is carried out and the calculated results are
(AACIR) is defined as the expected number of days in presented.
a year that the specified outage condition for the load bus Fig. 4 shows the main flowchart of the program. Using
will occur. For the system, the AACIR of load B and the set-up program, the user can construct a map of
load C are:
Pb = (1.37 X 10-3)(1.37 X 10 -3) = 1.878 X 10 -6 JStart /

A A C I R b = (1.878 x 10 -6) x 365 = 6.85 x 10 .4


Set-up Program
Pc = (1.878 x 10 -6) + (0.274 x 10 -3) = 0.2759 x 10 .3

AACIRc = (0.2759 x 10 -3) x 365 = 0.1006


Dijkstra's Algorithm
These calculating results are shown in Table 5.
This method can be easily applied to a relatively com- 1
plicated system. The reliability algorithm is based Transportation Algorithm [ changesthe lines type ]
t with new line cost. )
entirely on the continuity of supply to the respective
load points. Therefore the system is assumed as a com- 1
pletely redundant system. In an actual system the failure Reliability Calculation, i l

Table 4
System parameters
T m e ~ False
1
Line Section Failures/Year

1 0.5 C,o,,,p,,, Ros,It i


2 0.5
3 0.1 Fig. 4. The main flowchartof the ROUTE.EXE.
256 Y.K. Wong et al./Microprocessors and Microsystems 20 (1996) 251--258

Land Type RelativeLine Cost Land lypc Cost

(Altitude)
m 0-100m ,0 m ............w a ~ e r ; ; a .......;iO . . . . . . . . . . . . . . . . . . " ~ ".or,
II (Altitude)
100-300m 14 Urban Area 80
Each type of icon
assigns different
cost directional
m (Altitude)
300-600m 20 Military Area 100
lines with a node
at c e n t r e
(Altitude)
above 600m 200 Restricted Area 999

Fig. 5. The transportation cost.

*The data is delivered by China hight and Po~er ('on~pan~ •

Fig. 6. A screen of construction result by user.

Fig. 7. Calculated network by the Transportation and Dijkstra's algorithms.


service area with the transportation costs and the failure transportation cost, as shown in Fig. 5. The cost o f land
probability o f lines. The service area is divided into a grid depends on the type o f transmission link used (overhead
o f squares called 'sectors', each characterized by a line, u n d e r g r o u n d cable or submarine cable) and the
specific icon. Based on the information f r o m CLP, altitude (the higher the altitude, the higher the cost).
eight types o f area icon are used, each having a different Fig. 6 shows the distribution area o f the power
Y.K. Wong et al./Microprocessors and Microsysterns 20 (1996) 251 258 257

company. Each sector represents an area of Castle Peak power station (Source l) supplies most of the
1.5kin x 1.5km. There are two source nodes (which substations. The Tsing Yi power station is near to some
correspond the two power stations in Castle Peak and substations, but it only supplies the Lai Chi Kok station
Tsing Yi) and six sink nodes (which correspond to the (Load 3) due to the relatively small power generation.
six 400/11 kV substations). In the figure, the circles repre- The output data (given in the Appendix) shows the
sent the source nodes while the dots represent the sink optimal routes from sources to each load distribution
nodes. and the reliability index Average Annual Customer
Program run of ROUTE.EXE invokes Dijkstra's Interruption Rate (AACIR). The AACIR indicates the
algorithm and the Transportation algorithm subpro- reliability of individual loading points, i.e. the expected
grams, yielding an optimization of the routing of the number of days in a year that the specified outage con-
CLP 400 kV network. Subsequently, the reliability calcu- dition for the load will occur.
lation subprogram computes the reliability indices of The network model used in this paper is assumed to
loads. If the failure rate is larger than the security have a single voltage level and transmission losses are not
index, then one more transmission line will be added in an important factor so that they are neglected during
parallel, or the type of line will be changed. Since the calculation. Also, we have assumed that the costs of
network configuration is modified, the computation pro- other equipment are not directly proportional to the
cess is repeated. The program run, when successfully length. The costs of other equipment are not included,
completed, gives the optimal source to load distribution e.g. transformers, circuit breakers. Furthermore, the
network configuration subject to reliability constraints. algorithms only give all individual transmission paths
Fig. 7 shows the computed results. It can be seen that the from the source to the load.

Table 6
Output table

Load 1 2 3 4 5 6 7 8 9
Demand 518 800 411 557 419 250 -

Source 1 3785 518 800 355 557 419 250 -


Source 2 56 56 - -

Source Load Line Loading AACI R

1 1 1 518 4.84039%-02
1 2 2 1318 4.039891e-04
l 3 2 335 4.039891e-04
1 4 2 557 2.272438e-04
1 5 2 419 1.783233e-04
1 6 3 669 1.009973e-04
2 3 1 56 4.800000e-02

Source Load Cost Sector(Route)


(Node) (Node)

1 (222) 1 (338) 956 222 223 224 205 186 187 188 189
190 191 192 213 214 235 256 2'77
298 318 338
I (222) 2 (298) 636 222 223 224 205 186 187 188 1'89
190 191 192 213 214 235 256 2'77
298
1 (222) 3 (295) 610 222 223 224 205 186 187 188 189
190 191 192 213 214 235 255 276
295
1 (222) 4 (234) 442 222 223 224 205 186 187 188 189
190 191 212 233 234
1 (222) 5 (137) 503 222 223 224 205 186 187 188 169
150 151 132 113 134 135 136 137
1 (222) 6 (150) 237 222 223 224 205 186 187 188 169
t50

2 (293) 3 (295) 448 293 274 295


258 Y.K. Wong et al./Microprocessors and Microsystems 20 (1996) 251-258

4. Conclusion [2] D.M. Crawford and S.B. Holt, A mathematical optimization tech-
nique for locating and sizing distribution substations, and deriving
their optimal service areas, IEEE Trans. Power Appar. Syst. Pas-
The Dijkstra's and Transportation algorithms are 94 (March/April 1975) 230-235.
found to be effective in the design of power distribution [3] R. Billinton, Power System Reliability Evaluation, Gordon and
networks. These algorithms yield the most economic net- Breach, Reading, 1970, pp. 163 203.
work routing, while the Average Interruption Rate [4] B.D. Sivazlian and L.E. Stanfel, Optimization Techniques in
Method ensures that the design meets the reliability spe- Operations Research, Prentice-Hall, Englewood Cliffs, NJ, 1975,
pp. 249-297.
cifications. The effect of new loads can be reviewed [5] S.E. Dreryfus, An appraisal of some shortest-path algorithms, in
quickly and more thoroughly than before, and the loca- Perspectives on Optimization: A Collection of Expository Articles,
tion of new substations can be determined easily. Very Addison-Wesley, Reading, MA, 1972.
encouraging results have been obtained from program [6] D.T. Phillips and G.D. Alberto, Fundamentals of Network
runs with particular reference to a local power system. Analysis, Prentice-Hall, Englewood Cliffs, N J, 1981, pp. 39-68,
pp. 88-89, pp. 111 129.
[7] F.E. Montmeat and A.D. Patton, Power System Reliability II
Applications and a computer program, IEEE Trans. Power
Appendix Appar. Syst. (July 1965) 636-643.
[8] M.S. Bazaraa and J.J. Jarvis, Linear Programming and Network
This optimization result (refer to Section 3) is a part of Flows, John Wiley, New York, 1990, pp. 353-383, pp. 481-492.
[9] B. Kolman and R.E. Beck, Elementary Linear Programming With
distribution network in Hong Kong. Applications, Academic Press, New York, 1980, pp. 271 300.
[10] S.I. Gass, Linear Programming: Methods and Applications,
OUTPUT DATA: McGraw-Hill, New York, 1975, pp. 245 269.

Number of Source: 2 power stations Y.K. Wong received the BSc and MSc degrees
Source 1: Sector 222, Supply 3785 MW from the University of London and the MPhil
degree .from the Chinese University of Hong
Source 2: Sector 293, Supply 56 MW Kong. He joined the Hong Kong Polytechnic"
in 1980. His current research interests include
Number of Load: 6 substations optimization theory', distribution network, con-
trol and identification theory, cognition science
Load 1: Sector 338, Demand 518 MW and artificial intelligence.
Load 2: Sector 298, Demand 800 MW
Load 3: Sector 295, Demand 411 MW
Load 4: Sector 234, Demand 557 MW K.L. Shi received the BSc degree from the
Load 5: Sector 137, Demand 419 MW Chengdu University of Science and Technology
Load 6: Sector 150, Demand 250 MW and the MSc degree from The Harbin Institute
Technology. He is now a PhD student at Hong
Power Distribution Kong Polytechnic University. His current
research interests include induction motor con-
trol, network theory, pattern theory and cogni-
Total Cost Product: 1 761 847 tion science.

Output tables are shown in Table 6.


T.F. Chan received BSc and MPhil degrees
from the University of Hong Kong. He joined
the Hong Kong Polytechnic in 1978. His cur-
References rent research interests include artificial intelli-
gence application, electrical machines analysis,
design and control.
[1] T.C. Hu, Integer Programming and Network Flows, Addison-
Wesley, Reading, MA, 1969, pp. 151 161.

You might also like