Вы находитесь на странице: 1из 6

Application of the Ant Colony System for

Optimum Switch Adjustment


Jen-Hao Teng, IEEE Member and Yi-Hwa Liu

Abstract: In this paper, a co-operative agent algorithm, the counterbalanced by the fact that, for some applications,
ant coloriy system, for optimum switch adjustments is the ACS can outperformother heuristics [4-81. ACS has
proposed. Switch adjustment is one of the most important been applied to many different optimization problems
tools for distribution automation, since it can improve the
such as quadratic assignment, graph sequential ordering
system reliability and reduce the interruption costs without
additional capital investments. The formulation of switch and single machine tardiness and in the field of control
adjustment is a combinatorial optimization problem with systems, communication systems and power systems etc.
non-linear and non-differential objective function. In this For most of those applications, the results show that the
paper we choose the ant colony system to solve the problem ACS based approach can outperform other heuristic
since the ant colony system has the characteristics o f methods. In power systems, ACS has hccn applied to
positive feedback, distributed computation and the use of a solve the optimum generation scheduling problems [7, 81.
constructive greedy heuristic. One of the main goals o f the [7] used the ACS to find the best results of the short-term
paper is to investigale the applicability of the ant-colony- generation scheduling problem. [8] employed ACS to
system-bwed algorithm in the power system optimizations. solve the hydroelectric generation scheduling. In this
Test results show that the proposed ant-colony-system-
based algorithm can offer an optimum solution for switch paper, authors will integrated the ACS into distribution
adjustment: system planning.

Index Terms: Ant Colony System, Customer Interruption The distribution system is an important part of the
Cost, Reliability Improvement, Switch Adjustment, total electrical supply system, as it provides the final link
Distribution Automation between a utility's bulk transmission system and its
customers. It has been reported that 80% of all customer
1. INTRODUCTION interruptions occur due to failures in the distribution
system. In order to improve customer service reliability,
Many optimization problems in power systcms are Distribution Automation (DA) is applied on the
combinatorial optimization problems; they are difficult distribution feeders to achieve significant and immediate
to he solved by traditional linear or nonlinear improvement in reliability and service to the utility
programming methods due to its non-linear and non- customers. Switches are utilized comprehensively in
differentiable characteristics. Therefore, many heuristics primary distribution systems for various applications,
have been proposed that attempt to find a near-optimum such as, to improve reliability, isolate a fault, reconfigure
solution for large power systems within reasonable time. and restore the network and hence reduce the
Such heuristic algorithms include Ant Colony System interruption cost after a fault. The selection of an
(ACS), evolution strategies, genetic algorithms, adequate number of switches and their locations is a
simulated annealing, neural networks, tabu search, and difficult task in distribution system planning [9-14].
threshold accepting etc. Moreover, after the switches were installed in the
distribution networks and operated for few years, the
Among these methods, ACS was proposed in the existent switches need to he adjusted according to the
early nineties as a novel heuristic approach for the load growth and network expansion. By adjusting the
solution of combinatorial optimization problems [1-5]. existent switches, the distribution system reliability can
The ACS, originally introduced in p, 21, is a new be improved and the feeder interruption cost can be
cooperative search algorithm inspired by the behavior of reduced. The formulation of switch adjustment is a
real ants, since ants are able to find good solutions to combinatorial optimization problem with non-linear and
shortest path problems between a food source and their non-differential objective function. Teng and Lu have
home colony. The characteristics of ACS algorithm proposed some useful strategic formulations for switch
include positive feedback, distributed computation and adjustment in [IS]. In this paper, the proposed method
the use of a constructive greedy heuristic. Positive integrates the ACS algorithm into the optimum switch
feedback makes sure for rapid search of global solution; adjustment problem. Test results demonstrated the
distributed computation avoids prematurc convergence performance ofthe proposed method.
and constructive greedy heuristic helps find acceptable
solutions as soon as possible. These properties are 2.BASlC CONCEPTS OF ACS ALGORITHM

The main concepts and procedures o f ACS are


This work was sponsored by National Science Council, Taiwan. under introduced in this section. ACS simulates the behavior of
research grant NSC 90-2213-E-214-052-. real ants [I-61. Real ants are ablc to find the shortest path
fen-Haa Teng is with Department of Electrical Engineering, I-Shou between a food source and their home colony. They also
Un bersity. Kaahsiung, Taiwan.
Yi-Hwa Liu is with Power Electronic. Lab., Electra-rechnoloa have the ability to adapt the environment changcs, for
Division, Energy Resource Labs., ITRl. example, finding a new shortest path once the old one is

0-7803-7525-4/02/$17.00 0 2002 IEEE. 75 1


no longer feasible due to a new obstacle. The studies of update the pheromone trail.
real ants reveal that they communicate via pheromones. 4. Define the Convergence criteria of the problem.
Ant deposits a certain amount of pheromone when it This process is iterated until the tour counter
walks. An ant tends to choose a path positively reaches the maximum pre-defined number of
correlated to the pheromone intensity of foundtsiak. The iterations or all ants make the same tours.
pheromone trail evaporates over time, i.e., it looses
intensity if other ants lay down no more pheromone. If The full procedure is problem-dependent; it is
many ants choose a certain path and lay down difficult to find a single searching space, configuration,
pheromones, the intensity of the trail increases and thus consideration and parameters of ACS algorithm that can
this trail attracts more and more ants. The process can satisfy every optimization problem. It needs to take the
easily be illustrated by Fig. I . Fig. I(a) shows that ants specific factors of each optimization problem into
are moving in a straight line, which connects a food account to deign a good ACS algorithm.
source and their home colony. Once an obstacle appears
in the straight line as shown in Fig. l(h), the environment 3. FORMULATION DEVELOPMENTS
changes and the ants that are just in the front cannot
continue following the pheromone trail. Therefore, those Problem Formulation
ants will choose the paths randomly; it means that the
ants have same probability to choose the path to point C Reliability worth assessment is currently receiving
or D.Later on, those ants choose the shorter path around considerable attention as it provides an opportunity to
the obstacle will move faster than the ants choose the incorporate the costs or losses incurred by utility
longer path. The pheromones on the shorter path will be customers as results of power failure [13-151. In this
reconstructed more rapidly and this will cause more ants paper, we try to integrate the reliability worth to obtain
to choose the shorter path. After all, all ants will choose the optimum locations for the switch adjustment. The
the same path due to the positive feedback. The full objective function of the optimum switch adjustment
procedures of the ACS algorithm can be summarized as: problem is to minimize the interruption costs for
distribution feeders by rearranging the existent switches.
B E E
In a distribution feeder the group of line segments
between adjacent switches or circuit breakers (CB) is
defined as a Section. Each Section will have several load
points and the equivalent load of a Section is obtained by
summing individual loads within the Section. The
Customer Interruption Cost (CIC) of the feeder can be
expressed as

tA
A A where n is the number of switches. li , Li and 2, are
(a) (b) (C) the length, load and outage rate (failurekm-year) of
Fig. I : An Example of ACS Section i . C, is the per kilowatt interruption cost ($/kW)

Initialize the ACS-based optimization problem. of load at Section j due to an outage at Section i. C , can
Construct searching space including the states and be expressed as
stages of the optimization problem and set the ant
number and the parameters of the ACS algorithm.
Find the paths for the ant dispatch. Each ant
chooses the states to complete a tour according to a
probabilistic state transition tule. Ants prefer to
move to states, which are connected by shorter
edges with a high amount of pheromone. Once all f C ( z j )f ,i ( x j ) are the load percentage and the
ants have finished their tour, some fitness functions
of the optimization problem can be used to interruption cost functions of residence, commercial and
evaluate the performance of the ants. industry customers respectively. y, is the duration of
Update the pheromones o f edges between each service interruption ofSection j for a fault at Section i . In
stage. The pheromone trail of each edge will computing y, s , the long-term historical data including
evaporate over time, i.e., it looses intensity if no
more pheromone is laid down by other ants. For average time required for feeder fault detection, isolation
those edges that ants traveled in this iteration, their and restoration needs to be collected. Besides, the feeder
pheromone intensity can be updated by the topology, switch location, failure rate, switch operation
pheromone-updating rule. Global and local (manual or automatic) time and load transfer capability
pheromone updating rules are generally used to all need to be taken into account.

152
Solution Technique
If the location of the switch between Section i and
Section i+l was adjusted, then the CIC before and after n
adjustments can he expressed as (3) and (4), respectively.

,It1 "+I
c I c ( y )= C I C , +AI,(( ccU!,)+ciiLi +CiitlLj+])
m=l j=l
~+i.i+l jti,i+l
ntl
(3)
+ A+lk+l(( cCitlj Lj )+C,+I,L,+
j=l
Ci+li+l
&+I)

j+i,i+l
"+I "+I
CIC(b)= C I C , +AI,:((CcijLj)+c,:iL;+Ci;+]L;.+])
m=l j=l
",*<i+l jti,i+l
"+I (4)
+ A + I L(( cc.,
j =I
j Lj )+ C;.+,,L:+C,:+li+lL;+l)
jiijtl

where li , 1,; and i,,, , Ii', are the length of Section i


4 b
and i+l before and after the switch adjustment. Li , Li Number of Candidate Locations
Fig. 2: The Searching Space for the Switch Adjustment
and Li+], ii+]
are the load of Section i and i+l before
and after the switch adjustment. C , and C, are per Fig. 2 is the searching space of the switch
kilowatt interruption cost of load at Section j due to an adjustment problem. All possible candidate locations for
outage at Section i before and after the switch switch rearrangements are the states in the searching
adjustment. space for one stage. The number of stages is equal to the
number of switches. Each ant will start its tour at the
The CIC variation before and after switch adjustments home colony and stop at the destination. Ant chooses the
can be written as state to go to with a probability that is a function of the
amount of pheromones present on the connecting edges
of each state. When an ant moves from one stage to the
ACIC=CIC(*)-CIC("
next stage, the state of each stage will be recorded in the
= A[l;(c;iL; +c;i+l
L)+J -li(ciiLi +Cii+lLi+l)]
(5)
location list. After the tour is completed, its location list
is used to compute the ant's current solution, i.e., the CIC
+ A+l~li'+,(ci'+liL;
+ ci'+,i+lL;.+,) based on this location list The location list is then
-1i+l(ci+liLi + Ci+li+lLi+I)l emptied and the ant is free again to choose a new trail

From (S), it can be seen that we can improve the Let T Y ( N ) be the pheromone intensity of trail on
service reliability and reduce customer outage cost
edge (i. J) between stage x and y at iteration N. Note that
without additional capital investments if the adjustment
stage x a n d y are the consecutive stages. Each ant at state
is done properly. Especially, after the switches were
installed in the distribution networks and operated for
i of stage x chooses the next stage according to state
few years, the load distribution and network topology has transition rules. Different optimization problems can use
been changed, the switches need to be adjusted and the the different state transition rules. In this paper, we use
proposed method will be very helpful. This paper the following transition rule to give probability for an ant
to select its edge.
investigates the benefits of rearranging the existing
switches. For this purpose, the objective function of the
problem is

where ?J,yis the CIC reduction between stage x and y .


where R is the location list of the switches.
a and p are parameters that control the relative
Due to its non-linear and non-differentiable importance of trail versus the CIC reduction between
characteristics, the problem is difficult to be solved by stage x and y. Therefore the transition probability is a
traditional linear o r nonlinear programming methods. In trade-offbetween the CIC reduction and trail intensity. It
this paper, an ACS-based algorithm is proposed to solve means that if the path has been a lot of traffic and larger
the problem. CIC reduction then it will have more probability to he
selected.

753
~

After this tour is finished a n t i will have its location list, relative importance ofthe CIC reduction between stage x
RI"', and fitting evaluation, CIC(R!"). At this point and y in our optimization problem. p is the trail
all trail intensity can be updated according to the persistence and (1- p ) can be interpreted as trail
following evaporation formula evaporation. Q is a constant value related to the quantity
of trail deposited by ants. Different parameters will
q ( N + 1) = p . q (N)+s,;y,'O' (8)
dcfinitely affect the performance of an ACS algorithm.
In this paper, we choose the following parameters: ant
n u m b e ~ 3 0 ; oc=l.0; p =1.0; p=0.7; and Q=5000000.
where p is a coefficient such that ( I - p ) represents the
evaporation of trail after iteration N . ~j?'"' is a small 4. NUMERICAL RESULTS
positive constant. The coefficient p must be set to a
The proposed algorithm was implemented with
value less than I to avoid unlimited accumulation of Borland C++ language on a Pentium-111 PC. Several
trail. feeders were tested, however, only the results of one
distribution feeder are repotted here due to limited space.
The ant with best solution at this iteration can provide a The test feeder is shown in Fig. 4. Each bus on the test
greater amount of pheromones to the edges of its location feeder is a candidate location for switch relocation.
list. For example, if ant k has the best solution in this tour, Outage rate of the feeder is 4 failures per 100 miles per
then the pheromones of those edges in its location list year. Average repair time is 1.5 hrs and the average
can be modified by duration to perform the necessary fault isolation,
switching and load transfer activities from substation and
neighboring feeders is 0.1 hr. Maximum feeder load
transfer capability from substation B, C and D are
800kW. Costs of energy not supplied are as those used in
[13, 151. The maximumiteration number is set as 100.

I:Switch

1 :Canadiate Location for Switch

to
the number of switches and candidate
....................

I I I I I I
M...
.................. i

........

...............
Fig. 4: One-LineDiagram of Test System

The best solution of each iteration and the CIC of


each ant at the final iteration are shown in Fig. 5 and 6,
respectively. From Fig. 5 , it can be seen that the ACS
based algorithm can be used to find the optimum location
list for switch adjustment. Fig. 6 shows that about 60%
.. of ants at the final iteration choose the same location list,
the other ants still try to find the best locations for the
switch adjustment. If the maximum iteration number is
Fig. 3: The Computation Procedure of the Proposed Method set 500, then about 98% ants are choosing the same
location list, only 2% ants still try to find the global
The computation procedure of the ACS-based optimum solution. The switch locations after
switch adjustment algorithm is shown in Fig 3. The rearrangement are shown in Fig. 7 . Thc yearly CICs
parameters that affect the computation of ACS algorithm before and after adjustments equal 20708 and 16336
directly or indirectly including ant number, a, p , p USD, respectively. Test results also show that only
and Q. CL can affect the relative importance of the trail, rearranging the existent switches can reduce customer
its value should be larger than zero. p affects the outage cost and improve service reliability.

754
reasonable computation time. The parameters of ACS
algorithm will definitely affect the performance of the
..... II proposed method. However, the hasic ideas proposed in
..... L
...... I ?! II
I this paper need no modification. The relative strengths
and weaknesses o f the proposed method with respect to
those parameters including ant number, (x, p , p and
!...... V[
I Q, and the comparisons of the ACS based method with

..... L
c
..... . . . . . . . . . . . .
other heuristics will be discussed in the future works.

6. REFERENCES

[ I ] Dorigo, M.; Maniezzo, V.; Colorni, A., Ant system:


.-..-.-.. optimization by a colony of cooperating agents,
lEEE Transactions on Man and Cybernetics Systcms
Fig. 5: The Best Solution of Each Iteration Part B, Vol. 26 Issue: 1, Feb. 1996 pp. 29 -41
[21 Dongo, M.; Gamhardella, L.M., Ant colony system: a
cooperative learning approach to the traveling salesman
problem: IEEE Transactions on Evolutionary
Computation, Volume: 1 Issue: I , April 1997, pp. 53 -66.
..... [3] Smtzle, T.; Hoos, H., MAX-MIN Ant System and local
......
!
search for the traveling salesman problem, IEEE
International Conference on Evolutionary Computation,
...... 1997, pp. 309-314.
[4] Dorigo, M.; Di Caro, G., Ant colony optimization: a
: .... new meta-heuristic: Proceedings of the 1999 Congress
on Evolutionary Computation, Vol. 2, pp. 1470.1477
. . . . . . . . . . [5] Merz, P.; Freisleben, B., A comparison of memetic
I ~~
..-..
Fig. 6: The CIC of Each Ant at the Final Iteration
I algorithms, tabu search, and ant colonies for the quadratic
assignment problem, Proceedings of the 1999 Congress
on Evolutionary Computation, Vol. 3, pp. 1999 -2070.
[6] Yong-Hua Song; Irving, M.R. Optimization techniques
for electrical power systems 11: Heuristic optimization
I:Switch methods: Power Engineering Journal, Volume: 15 Issue:
\ :Canadiate Location far Switch 3, June2001, pp. 151 -160
[7] In-Keun Yu; Chou, C.S.; Song, Y.H.,Application ofthe
ant colony search algorithm to short-term generation
scheduling problem of thermal units: lntemational
Conference on Power System Technology, Vol. 1. 1998,
pp. 552 -556.
[XI Shyh-her Huang, Enhancement of hydroelectric
generation scheduling using ant colony system based
optimization approaches, IEEE Transaction on Energy
Conversion, Vol. 16 Issue: 3, Sept. 2001, pp. 296-301.
[9] Luth J., Four Rules to Help Locate Protective Devices,
Electrical World, Aug., 1991, pp. 36-37.
[IO] Levitin, G , Mazal-Tov S., and Elmakis D., Optimum
Switch Allocation in Electric Distribution System by
Genetic Algorithm, Electric Power Systems Research,
No. 31, 19<4, pp. 97.102.
L
I I I 1 Billinton. R.. Jonnavithula.. S... ODtimum
1 . Switching
Device Piacement in Radial Distribution Systems: IEEE
Transactions on Power Delivery, Vol. I I , No. 3, July
Fig. 7: Switch Locations ARer Adjustment 1996,pp. 1646-1651.
[I21 Celli, C., Pilo, F., Optimum Sectionalizing Switches
5. DISCUSSION AND CONCLUSION Allocation in Distribution Networks, IEEE
Transactions on Power Delivery, Vol. 14, No. 3, July
The formulation of switch adjustment problem is 1999, pp. 1167-1172.
a combinatorial optimization problem with non-linear [I31 Chowdhury, A A , Koval, D.O., Value-Based
and non-diffcrential objective function. In this paper, we Distribution System Reliability Planning; IEEE
Transactions on Industry Applications, Vol. 34, No. I ,
used the ACS algorithm to solve the optimum switch Jan.-Feb. 1998, pp. 23 -29.
adjustment problem. It is difficult to find out a single [I41 Chen R. L., Allen, K., Billinton, R., Value-Based
searching space, configuration, consideration and Distribution Reliability Assessment and Planning, IEEE
parameters of ACS algorithm that can satisfy every Transactions on Power Delivery, Vol. I O , No. I , Jan.
optimization problem. In this paper, we integrated the 1995, pp. 421 4 2 9
ACS algorithm into the optimum switch adjustment [I51 Teng, J. H., Lu, C. N, Feedcr Switch Relocation for
problem successfully. Test results shown that the ACS Customer Interruption Costs Minimization: IEEE
algorithm can be used to find a good solution within

755
T ~on power
~ ~system,
~ vol.
, 17, N ~ 1,. pp, 254- University, Taiwan, since 1998. His current research interests
259,2002. are Energy Management System, Distribution Automation and
Power Quality.
7.BIOGRAPHIES Yi-Hwa Liu received his Ph.D. degree in electrical engineering
from National Taiwan University in 1998. Presently, he is a
Teng was born in 1969 in Tainan2 Taiwan. He research scientist in Industrial Technology Research Instihlte
received his BS, MS and Ph'D' degrees in (ITRI) Taiwan. His current research interest are in the areas of
engineering from the National Sun Yat-Sen University in 1991, Power Conversion and System Quality,
1993 and 1996 respectively. He has been with the I-Shou

756

Вам также может понравиться