You are on page 1of 6

A Unified Scheme for Testing Alternative

Techniques for Distribution System Minimum Loss Reconfiguration


Florentin Batrinu, Enrico Carpaneto and Gianfranco Chicco
global optimisation methods are needed. This problem has been addressed by using deterministic and heuristic methods. Early developments and results are presented in [7,14,15]. More recently, other algorithms have been developed by adopting the deterministic methods as iterative improvement (IT) and Tabu Search (TS) [9-11], or heuristics like Simulated Annealing (SA) [4,6,9,12,13], Ant Colony Search (ACS) [3] and Particle Swarm Optimization (PSO) [5,16]. Several literature papers address the implementation of optimal reconfiguration algorithms, but only a few papers contain extended information on comparing different algorithms [1,4,16]. However, none of these methods is by definition able to guarantee that the global optimum can be found in finite time or number of iterations. Hence, searching for efficient alternative methods or refining the existing ones is still a challenging task. This paper deals with developing a unified framework able to embed the characteristics of the various principles. methods for minimum loss reconfiguration. This framework is Indexing Terms - distribution systems, optimal reconfiguration, built by first using a conceptual decomposition of each minimum losses, deterministic and heuristic methods, iterative method, in order to identify the underlying principles that improvement, simulated annealing, particle swarm optimization. characterise every method. Then, each method is reformulated by structuring its solution in terms of the underlying I. INTRODUCTION principles. Finally, by comparing the structures of the solution Electricity distribution systems are typically structured as processes, it is possible to build the unified framework, meshed networks, with the system nodes interconnected by a consisting of a unique external cycle, with a few components number of branches higher than the number of nodes, but are depending on the individual methods, and a dedicated internal operated with radial configurations in order to simplify the cycle for each method, that includes its specific features. This protection schemes. The system nodes can be partitioned by unified structure makes it possible to simplify the comparisons identifying the supply nodes and considering all the other among the performance of the different methods, highlighting nodes as load nodes (also including interconnection nodes the differences between specific parts of the solution process. with null load). In a radial configuration, each load node has a In addition, it simplifies the writing of the program codes by unique path that connects it to one of the supply nodes. Then, increasing their modularity. Finally, the underlying principles each supply point becomes the root of a tree formed by can be combined to develop suitable hybrid or new opening a number of redundant branches of the meshed formulations of the optimisation techniques. The remainder of the paper is structured as follows. Section structure. There are several criteria for deciding which II sets up the notation and the mathematical formulation of the branches have to remain open, in order to optimise suitable objective functions. A typical objective function is given by constrained objective function. Section III presents the overall the total distribution system losses. Other possibilities may set of underlying principles. Section IV develops the unified include functions based on economic analysis terms or framework and applies it to interpret the existing methods, and to extend the application to the specific formulation of the reliability indicators, or multiobjective functions. This paper considers the distribution system reconfiguration Evolutionary Particle Swarm Optimisation, derived from the for power loss minimisation. The variety of the possible concepts introduced in [17], adapted to the minimum loss configurations of the distribution systems leads to a discrete reconfiguration by suitably mixing the underlying principles. combinatorial problem with several local minima, so that Section V shows the results obtained by the comparison of the methods on a large real distribution system.
The authors are with the Dipartimento di Ingegneria Elettrica, Politecnico di Torino, corso Duca degli Abruzzi 24, 1-10129 Torino, Italy (E-mail { florentin.batrinu,enrico. carpaneto,gianfranco.chicco } polito. it).

Abstract- The optimal reconfiguration of a large distribution system is a global optimisation problem typically solved by using deterministic or heuristic methods. Comparing the effectiveness of the various methods can be assisted by formulating a unified framework able to identify the common characteristics and the conceptual differences among the methods. This paper illustrates the development of such a framework, interpreting the solution process of a number of methods (Iterative Improvement, Tabu Search, Simulated Annealing, Ant Colony Search and Particle Swarm Optimisation) on the basis of a set of underlying principles, and applies this framework to the reconfiguration of a large real urban distribution system. The paper also shows how the proposed framework allows for developing additional solution algorithms, and presents effective results obtained by using a specific formulation of the Evolutionary Particle Swarm Optimisation derived from suitably mixing the underlying

II. SYSTEM STRUCTURE AND OBJECTIVE FUNCTION

III. CHARACTERISTICS OF THE OPTIMISATION METHODS

In order to illustrate the general framework for the optimal reconfiguration of a distribution system with radial operation, the set of load nodes K and the set of branches B are introduced to describe the status of a distribution systems. Let's then consider a distribution system with K nodes and B branches. Each configuration of the system, defined by the states of the branches, is represented by using a vector x containing B binary values (0 = open, 1 = closed). Let's call X the set of the radial system configurations. The reference configuration x() E X corresponds to the standard of operation of the distribution system. Each branch be B of the system is characterised by its electrical parameters (resistance Rb and reactance), and the electrical currents lb flowing through the branches generate the total power losses

Each optimisation method includes the application of specific mechanisms in the solution procedure. The solution procedures have been revisited with the aim of identifying a set of underlying principles either forming a common basis for the various methods, or embedding the structural differences among the methods. The set identified contains the following principles: acceptance decay elitism immunity selection self-adaptation parallelism topology

A. Acceptance The principle of acceptance drives the selection of the configurations to be further evaluated or discarded. It may be applied in two different ways: a) accepting only significant improvements: a threshold may be introduced for accepting only the configurations (1) f(x)= ZRbI leading to significant improvements in the objective function; beB b) accepting configurations leading to an objective function The system operation is characterised by a number of worsening: this is the key concept used for updating the constraints. The basic constraint refers to keeping the system configuration radial, i.e., xeX. In addition, there are the configurations by avoiding trapping into local minima; the various methods differ in the formulation of the acceptance thermal current rating limit lb - Ib M <0 for bE B, and the principle, but the implementation is similar, depending on a threshold (1-3) applied to the best objective function found so constraints for for ke K: far before comparing it to the current objective function. - VIMAX <0 * max node voltage magnitude VI
* * *

the min node voltage magnitude VkMIN


max max

Vk < 0

B. Decay

phase-earth fault current Ief -kef MAX <0 3-phase short-circuit current 1cC - 1sc'MAX <o0
MAX

Other constraints could be introduced to account for the maximum number of operations n to reach the final configuration starting from the reference configuration, implemented as n - n MAX < , or protection coordination, or specific contract agreements between supplier and user, as detailed in [4]. By denoting the constrained variables as oC j, forj 1,...,
OP

J,

each constraint is

subject

to the

respective

limit

o7i

LIM
.

The principle of decay is applied to suitable parameters of the methods, in order to allow larger initial flexibility in the application of the method, followed by more restricted possibilities in its development. The SA method adopts the decay principle as primary concept, so that it affects the value of the control parameter in the main or external cycle of the algorithm. The other methods typically apply the decay principle to one or more of their parameters. C. Elitism The principle of elitism is applied to avoid losing the configuration corresponding to the best objective functions found so far. The best configuration is taken as reference for further progress of the procedure.

In

the power loss minimisation problem, violation of any limit is associated to the presence of the corresponding penalty factor pj, for j =1,..., J. Each penalty factor is null if the corresponding constraint is not violated and positive if it is violated. The equality constraints are split into two inequality constraints, each of which is associated to the corresponding penalty factor. The penalised objective function is then formulated as
1(X)

D. Immunity The principle of immunity is based on recognising local properties of the solution that could drive the evolution of the algorithm towards more satisfactory configurations, and on applying them with higher priority with respect to other characteristic of the method. This principle is typically applied with success to algorithms able to provide effective global evolution, without considering in details the local solutions. E. Parallelism

ff (x)

pj(%

(2)

so

that the optimisation problem becomes


(3)

min f (x)
xeX

The parallelism principle consists of sending multiple entities in parallel to perform the same task. The entity of this "taskforce" that obtains the best result drives the successive evolution of the method. F. Selection The principle of selection refers to the possible choice of different configurations that have been already evaluated and

are simultaneously available. Several methods adopt random selection by means of the biased roulette wheel mechanism. G. Self-adaptation The self-adaptation principle consists of changing the parameters characterising the method in an automatic way according to the evolution of the procedure. The basic concept is to avoid the proliferation of user-defined parameters when some directions in the evolution of the parameters can be identified and associated to the behaviour of the method. H. Topology In order to ensure that only radial configurations are generated during the execution of the solution procedure, the constructive ACS method is intrinsically able to generate only radial structures. For the other procedures, the branchexchange mechanism [2] is adopted: once closing one of the open branches of the system, a closed loop is formed in the system configuration; in order to restore the radial configuration, it is necessary to open a branch inside that loop; the selection of the open branch to close (and successively of the closed branch to open) may be performed by using a specific rule, or at random; the importance of the branchexchange mechanism is due to the fact that the neighbourhood of the radial configuration currently under evaluation is composed of all the radial configurations that can be reached from the current one by performing a single logical step, i.e., by using the branch-exchange mechanism.
IV. UNIFIED FRAMEWORK FOR THE OPTIMISATION METHODS

directives given in [17], adapted in terms of the underlying principles. More specifically, the EPSO concept is based on the PSO method and requires at each iteration to replicate each particle; each copy of the particle is then subject to mutation of its weights and produces new candidate configurations according to the so-called "movement rule", leading to a modified version of the velocity equation of the PSO method; then, the objective function is evaluated for the initial particle and for all its modified copies, and the comparison among the results obtained (e.g., with tournament selection) gives the surviving particle. The blocks making the difference between the basic PSO and the EPSO are illustrated with dashed borders in Fig. 3, with their interpretation in terms of the underlying principles also indicated in Table I.
START

w sdep

algorithm dependent

computeJx(x )); xbest-x(0)

.~
control block
NO

m =, mo=0

internal procedure

top conditi satisfied?

YES OUTPUT

On the basis of the underlying principles, a common scheme has been developed to represent various optimisation methods under a unified framework. The basic methods tested include the deterministic IT and TS, and the metaheuristics SA, ACS and PSO. The various algorithms have been interpreted in the light of an external cycle (Fig. 1), structurally similar for all the algorithms and with a few parts depending on the type of algorithm, and of an internal cycle, where the specific features of the methods. Comparing the structure of the methods in the internal cycle is a good basis for understanding the application of the various principles and to create the field for formulating possible new procedures. The common part of the external cycle sets up the reference configuration x(), the iteration counter m and the counter mo that will contain the number of iterations of the external cycle performed without improvements in the best objective function. The other blocks depend on the algorithm. In particular, for the SA the control block contains the update (decay) of the control parameter cm, whereas for all the other methods it is empty (direct feedback). The stop condition is put in the same framework for all the algorithms, but the deterministic methods (IT, TS) stop when mo 1 and the other methods stop when mo reaches the limit specified by the user. The internal cycles are shown in Fig. 2 and Fig.3, from which it is easy to recognise the similarities among some blocks. The framework developed has the advantage of easily allowing for extensions of the existing methods. A successful example has been the implementation of the Evolutionary Particle Swarm Optimisation (EPSO) by following the

Fig. 1. The external cycle.

V. APPLICATION TO A LARGE REAL DISTRIBUTION SYSTEM A comparison among the various methods on a real 22-kV real urban distribution system is presented. The system has 4 supply points, 531 nodes, 554 branches (23 redundant branches), total power supplied 86.7 MW, and total losses in the initial configuration 933 kW (i.e., 0.93 p.u., corresponding to 1.08 % of the total power) [16]. A statistical evaluations of the methods has been performed, representing the results by means of the Cumulative Distribution Functions (CDFs) of the objective function, as in Fig. 4. The methods have been run several times with different values of their parameters. Fig. 4 shows the CDFs of the objective functions obtained by the heuristic methods. It can be seen that almost all the methods have reached the best solution (losses 0.7925 p.u.), but EPSO is the most robust (high slope of the CDF curve). For most methods, the correlation among parameters and objective function is relatively poor, with no clear indication on how to choose the set of parameters to obtain the better solutions. The differences in the CDFs are clearly due to the extension of the ranges of the parameters chosen. Fig. 5 shows an example of evolution of the solutions (the initial objective function is not represented), by stopping each algorithm when mo reaches 100. It results that EPSO and ACS are able to reach acceptably good solutions in a limited number of iterations.

(A): acceptance (B): decay (C): elitism (D): immunity (E): parallelism (F): selection (G): self-adaptation (H): topology TS method IT method| _
m01

mo-E]~~ranch (H) Exchange


4r

~~~~~~~~~~~~m0= 1

(F)

compute information to assist neighbourhood

select x,e x

Meha'creation and local selection


(F)
select xex-=

Br c

compute the objective

functionx,)C)

(H)

Mechanism

Exchange

AtX) (A) |fi f(xbet)( 1 5)|(


YES

crite~~~~~~~~~~~~~crtrion satisfie
D
\

NO as

1az

nYES

I 0

NO
functionx,,x)

(C)l

bs-

(C) x

(A) reject x,.

(A)

ftx

best

)(X|
NO

YES
NO

List open branches explored? /

cycle

exit

(C)X

(A) reject x

ts

Fig. 2. Internal cycles for the IT, TS, SA and ACS methods.

I EPSO method]
(E)

,~~~~h
compute the l

send particle h
distance
vector as in PSO

(B) (G)

Compute the distance vector:


-

memory

(G)

inertia component component w.r.t. local best Xinh cooperation component w.r.t. global best xbest

(B) (G)
(E)

replicate particle r,. R times

(G

(G)

compute the distance vectors of the replicated


particles with mutated quantities
ex as in
--

(F)

select x(

PSO and computeJ(Xnh)

Branch

-F

Mechanism
-_

Exchange

(E)

(E)

select x(r eX for the replicated particles and


m

computej(

x() )

for

......R(H)

ExcnR

Branch

F)

tournament among the x(r) configurations for ..,R I r,. for choosing the surviving particle x"
nAf e J(Xbest) xnw
)

Mechanism

YES

(C)

Ix

=Xh

YES

NO

NO

Fig. 3. Internal cycles for the PSO and EPSO methods.

principle
acceptance

method IT PSO * accept worst solutions * worst solutions not accepted * accept worst solutions on a threshold basis on a threshold basis * possible restrictions to best solutions on a threshold basis

TABLE I APPLICATION OF THE UNDERLYING PRINCIPLES To THE OPTIMIZATION METHODS


ACS * accept worst solutions on a threshold basis SA * accept worst solutions on a threshold basis

TS

decay
elitism

* permanence into the tabu list L

* none

* conservation of the best * solution * aspiration condition * immunity * tabu list * none parallelism * random or locallyselection * driven * * branch-exchange mechanism self-adaptation * tabu tenure variation *

implicit, accepts only improvements

* in the external cycle * applied to the memory * pheromone decay control block component of the variation vector * conservation of the best * conservation of the best * conservation of the best solution solution solution

topology

* radial neighbourhood

* none none * number of swarms H none random or locally-driven * random (biased roulette branch-exchange mechanism wheel) * branch-exchange mechanism none * implicit in the cooperation terms * radial neighbourhood * radial neighbourhood

* none * number of ants H * random (biased roulette wheel) * branch-exchange mechanism * pheromone reinforcement * radial neighbourhood

* * * *

none none random branch-exchange mechanism

* none
* radial neighbourhood

Since at each iteration of the SA method the number of configurations analysed may be different, the representation of the evolution of the iterative processes is shown in Fig. 6 w.r.t. the number of configurations analysed. In the case shown, the ACS, PSO and EPSO methods analyse 100 configurations per iteration, and the SA method analyses 400 configurations at the first iteration. Table II shows indicative values (average value and per cent standard deviation) of the relative computation time index (RCTI), normalised w.r.t. the computation time of the IT method, with all calculations performed on the same PC. The fast IT and TS algorithms stop at the loss value 0.7927 p.u., slightly higher than the best solution, but clearly acceptable. For the heuristics, all the results converge in showing that the EPSO method exhibits excellent overall characteristics.

TABLE II AVERAGE VALUE AND STANDARD DEVIATION OF THE RCTI

standard deviation

average value

method

IT
1

TS 1.1

SA 189 124%

ACS 2.6 54%

PSO 8.1 45%

EPSO 1.9

510%

VI. CONCLUSIONS

The solution process of various deterministic and heuristic methods has been reformulated in terms of underlying principles. In the application shown, the EPSO method has emerged as a promising technique. Further studies will detail the properties of the optimisation techniques under the proposed unified framework.
VII. REFERENCES [1]

0 0.8
C.,

SA

[2] [3]

0.6
4.1

ACS

(U

0.4

0.2

[4]

o k0.79

0.8

0.81

0.82

0.83

0.84

objective function

[5]
[6]

Fig. 4. CDFs of the objective function for the different h euristics.


0.93 -h 0.91
-

[7]

0.89 0.87

[8]
,,S

0.85 ,0.83 -L-

[9] [10]

0.79-_=
0.770.75 0
EPSO ACS
500

PSO
1000

1500

2000

[ 11]

number of iterations

Fig. 5. Examples of evolution of the solutions for the heuiristics.


0.95 0.93
-

[12]
[13]

0.91-"_ I

0.89

PSO

,-

[14]
0.89
-

0.877-"'ES
0.75
-

[15] [16]

100

1000

10000

10)0'000

1000000

number of configurations analyEsed


-- - --Fig. 6. Evolution of the solutions w.r.t. number of config,urations analysed. - - ---- --

[17]

A.Augugliaro, L.Dusonchet and E.Riva Sanseverino, Genetic, Simulated Annealing and Tabu Search Algorithms: Three Heuristic Methods for Optimal Reconfiguration and Compensation of Distribution Networks, ETEP .9, 1 (1999) 35-4 1. M.E.Baran and F.F.Wu, Network reconfiguration in distribution systems for loss reduction and load balancing, IEEE Trans. Power Delivery 4 (1989) 1401-1407. E.Carpaneto and G.Chicco, Ant-Colony Search-based minimum losses reconfiguration of distribution systems, Proc. IEEE Melecon 2004, Dubrovnik, Croatia (2004) 3, 971-974. E.Carpaneto, G.Chicco and E.Roggero, Comparing deterministic and simulated annealing-based algorithms for minimum losses reconfiguration of large distribution systems, Proc. IEEE Porto Power Tech 2001, September 10-13, 2001, paper PST3-237. R.F.Chang and C.N.Lu, Feeder reconfiguration for load factor improvement, Proc. IEEEIPES Winter Meeting (2002) 2, 980-984. H.-C.Chang and C.-C.Kuo, Network reconfiguration in distribution systems using simulated annealing, EPSR 29 (1994) 227-238. R.Cherkaoui, A.Bart and A.J.Germond, Optimal configuration of electrical distribution networks using heuristic methods, Proc. 11th PSCC, Avignon, France, Aug. 1993, pp.147-154. M.Dorigo, V.Maniezzo and A.Colorni, The Ant System: An Autocatalytic Optimizing Process. Technical Report No. 91-016 Revised, Politecnico di Milano, Italy (1991) Y.-J.Jeon and J.-C.Kim, Network Reconfiguration in Radial Distribution System Using Simulated Annealing and Tabu Search, Proc. IEEEIPES Winter Meeting (2000) 4, .2329-233. H.Mori and Y.Ogita, A Parallel Tabu Search Based Method for Reconfigurations of Distribution Systems, Proc. IEEEIPES Summer Meeting, Seattle, WA (July 2000), 1, 73-78. K.Nara, Y.Mishima, A.Gjvo, T.Ito and H.Kaneda, Loss Minimum ReConfiguration of Distribution System by Tabu Search, Proc. Transm. and Distrib. Conference (2002) 1, 232-236. K.Nara and M.Kitagawa, Distribution system loss minimum reconfiguration by simulated annealing method, IEE International Conference on Advances in Power System Control, Operation and Management, Hong Kong (November 1991) 461-466 V.Parada, J.A.Ferland, M.Arias and K.Daniels, Optimization of electrical distribution feeders using simulated annealing, IEEE Transactions on Power Delivery 19, 3 (July 2004) 1135-114 1. G.Peponis and M.Papadopoulos, Reconfiguration of radial distribution networks: application of heuristic methods on large-scale networks, IEE Proc.Gen.Trasm.Distr. 142, 6 (1995) 631-638. R.J.Sarfi, M.M.A.Salama and A.Y.Chikhani, A survey of the state of the art in distribution system reconfiguration for system loss reduction, Electric Power Systems Research 31 (1994) 61-70. F.Batrinu, E.Carpaneto and G.Chicco, A novel Particle Swarm method for distribution system optimal reconfiguration, Proc. IEEE Power Tech 2005, St. Petersburg, Russia, June 27-30, 2005. V.Miranda and N.Fonseca, "EPSO - Best-of-two-worlds meta-heuristic applied to power system problems", Proc. CEC '02 Evolutionary Computation 2 (12-17 May 2002) 1080-1085.

You might also like