Electrical Power and Energy Systems 37 (2012) 86–94

Contents lists available at SciVerse ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier .com/locate / i jepes

Reconfiguration of large-scale distribution networks for planning studies

A. González a,⇑, F.M. Echavarren a, L. Rouco a, T. Gómez a, J. Cabetas b

a School of Engineering, Universidad Pontificia Comillas, C/Alberto Aguilera 23, 28015 Madrid, Spainb Iberdrola Distribución, S.A.U, Spain

a r t i c l e i n f o

Article history:Received 9 September 2010Received in revised form 5 October 2011Accepted 5 December 2011Available online 26 January 2012

Keywords:Heuristic AlgorithmPower distribution planningPower system restorationReconfiguration

0142-0615/$ - see front matter � 2011 Elsevier Ltd. Adoi:10.1016/j.ijepes.2011.12.009

⇑ Corresponding author. Tel.: +34 91 542 2800x450E-mail address: alezeia.gonzalez@iit.upcomillas.es

1 MV networks operate mainly between 1 kV and 311 kV in Europe and Asia). Typical values in Spain are

a b s t r a c t

Medium voltage (MV) distribution networks are usually operated radially. However, its structure may bedesigned meshed, so that in case of a failure of an element, the load downstream can be supplied by adja-cent feeders through reconfiguration of the network. Under this assumption, a reconfiguration tool is akey resource for the planning and the operation of MV distribution networks. This paper presents a heu-ristic reconfiguration algorithm of MV distribution networks. The algorithm determines the configurationthat minimizes the Non-Delivered Power not only in case of the failure of a single branch (line, trans-former) but also in case of a complete failure of a high voltage to medium voltage substation. The con-straints considered are that the radial structure of the network must be maintained and that linethermal limits, transformer capacities and bus voltages must be within their admissible ranges. Theresults of the algorithm show up weak regions of the network where planning actions should be carriedout. The main features of the algorithm are interpretability of the solution and low computation timeswhile handling complex and large-scale MV distribution networks. The algorithm has been tested on sev-eral actual distribution networks. Results of the application of the algorithm to real scenarios of Madrid’sdistribution network are provided.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

MV distribution networks1 are usually operated radially becauseof the low cost and simplicity of their protection schemes and oper-ation procedures, besides their lower short circuit currents. How-ever, MV distribution networks can be meshed to improve thesecurity of the supply. The design is made so that in case of failurethe load downstream from the failed element can be supplied byadjacent networks through its reconfiguration. Hence, a reconfigura-tion tool is a key resource to determine the support capability ofadjacent networks in case of contingencies.

The reconfiguration problem has been addressed from differentperspectives and with different objectives. These can be broadly cat-egorized into planning and operation problems. In operation analy-sis, the reconfiguration problem addresses issues such as servicerestoration [1–5], power losses minimization [2,6–12], solvingbranch overloads or bus undervoltages [4,9,12]. In planning studies,reconfiguration algorithms can be used to find optimal locations ofnew investments [13] and reliability assessments [10,14,15].Efficient planning of new investments must consider the supportprovided by the adjacent distribution networks. Otherwise, newreinforcements may have to be built to solve power system

ll rights reserved.

8.(A. González).0 kV (13.8 kV in America and15 kV and 20 kV.

constraints which could be avoided if efficient reconfiguration iscarried out. Reliability can be assessed by the minimization ofNon-Delivered Power (NDP) or Non-Delivered Energy (NDE). NDPis the amount of power which is impossible to supply in case of aparticular failure, whereas NDE corresponds to the total energy thathas not been supplied during the time a failure interrupts the sup-ply. The service restoration process may be analyzed by computingdifferent reliability indexes [16].

A wide range of methodologies are available for solving the net-work reconfiguration problem, namely Classical Optimization tech-niques (CO), Genetic Algorithms (GA), Parallel Simulated Annealing(PSA), Tabu Search (TS), Reactive Tabu Search (RTS), Data Base (DB),and Knowledge-based Heuristic Algorithms (KHA). CO techniquesformulate an optimization problem to obtain the optimal reconfig-uration of the distribution network. The evaluation of almost everypossible state of the network is needed when CO techniques areapplied to the reconfiguration problem [8,17,18]. Hence, computa-tion times increase dramatically when optimization algorithms dealwith large distribution networks. None of the references found inthe literature present CO techniques applied to large-scale distribu-tion networks. CO techniques are combined with heuristics in[1,6,19,20] to reduce the size of the optimization problem and thecomputation time. The solutions provided by the CO techniquescombined with heuristics are not easily interpreted, due to the factthat the solving process is not related to an actual state of the net-work. Thus, industrial users still do not trust the solutions found.GA [21], PSA [22], TS [23], and RTS [24] are metaheuristic algorithms

Fig. 1. Trip of substation S1.

A. González et al. / Electrical Power and Energy Systems 37 (2012) 86–94 87

widely used to solve combinatorial optimization problems. Thesealgorithms are based on the generation of neighboring states usingdifferent criteria. In each iteration, the best searching points are se-lected based on the evaluation of an objective function consistentwith the minimization of ENS. A comparative study for these fourtechniques is presented in [25]. The main disadvantage of GA, PSA,and TB is that search parameters must be adjusted to obtain an effi-cient search process. Although RTS solve this problem, the compu-tation times are still a drawback. None of the references found inthe literature that make use of Heuristic Algorithms present resultsfor large-scale real distribution networks. KHA are also well repre-sented within the literature to solve the reconfiguration problem[3,7,9,12,14,15,26]. More recent works develop KHA based on FuzzyReasoning [10,11,27–29] and Expert Systems [2,30–32]. KHA guar-antee interpretable solutions since they are obtained by implement-ing rules similar to real operation procedures. Besides, computationtimes are shorter than those achieved by CO techniques and meta-heuristics since a simpler and more flexible network model may bedesigned.

Regardless of both the applicability of the algorithm and theoptimization techniques, the starting point of the reconfigurationproblem may condition the solution and the computation time.Starting from an actual configuration usually leads to a simplesolution, but it sometimes prevents reaching a global optimum[4,21]. Not considering an actual initial state of the network usu-ally leads to higher computation times but less trustworthy solu-tions [2,7,8].

This paper presents a KHA for reconfiguration of MV distribu-tion networks. Although the main application of the proposed algo-rithm is within the planning environment, it can also be used foroperation studies. The algorithm provides the reconfiguration pro-cess that minimizes the NDP in case of failure of a line, a trans-former or a high voltage (HV) to MV substation. Emphasis isgiven on the resulting topology after reconfiguration and not onthe duration of the fault, and thus the NDP instead of the NDEhas been selected as the objective function. The constraints consid-ered are that the radial structure of the network must be main-tained and that line thermal limits, transformer capacities andbus voltages must be within their admissible ranges. The algorithmhas been designed using knowledge-based heuristic techniques asan iterative process. Actual rules followed on actual operation pro-cedures have been implemented in order to obtain a good estima-tion of reconfiguration processes and NDP.

As the algorithm aim at dealing with complex and large-scalereal distribution networks, features such as low computation timesand interpretability of the solution must be achieved. The algo-rithm reaches quasi-optimal solutions in short computation times.The algorithm uses a real configuration as the starting point in or-der to improve trustworthiness of the results. The performance ofthe algorithm will be shown with actual scenarios of MV distribu-tion network in Madrid.

Comparing the proposed algorithm with CO techniques, the maindrawback is that the proposal does not always achieve the optimalsolution. Compared with modern metaheuristics such as GA, PSA,TS and RTS, the proposed algorithm carries out a less exhaustivesearch. On the other hand, a huge reduction in computation costsis obtained and more complex and larger real networks are possibleto deal with while CO and modern metaheuristics present some lim-itations in handling very large and complex real network models.Furthermore, the proposal increases the interpretability of the solu-tion, which means a significant advantage when complex and largenetworks are analyzed. Nevertheless, good quasi-optimal results areobtained with the proposed approach.

The paper is organized as follows. In Section 2 some basic con-cepts about the reconfiguration problem and the proposed algo-rithm are explained. Section 3 describes in detail the algorithm.

Section 4 illustrates the performance of the proposed algorithmwhen applied to a real large scale distribution network. Finally,Section 5 contains the conclusions of the paper.

2. Basic concepts

This section reviews the basic concepts associated with the dis-tribution network reconfiguration problem. Basic concepts includemain support, margin of a support, virtual congestion, main sup-ports sub-network, secondary supports sub-network, secondarysupport, and basic switching.

2.1. Main support

MV distribution networks have a meshed structure, but they areoperated radially. Therefore, the distribution network model con-sists of a set of radial sub-networks which are interconnected bysome normally-open supporting lines, which are only closed incase of reconfiguration.

A main support is a bus of a particular sub-network which isconnected to a bus located on the failed sub-network through anopen line. They are used to transfer the load from the failed net-work to the adjacent network. The features of a main support arethe potentially recovered load and its margin. A main support isfeasible if its margin is greater than the potentially recovered load,otherwise the support would be unfeasible.

2.2. Margin

The margin of a support is defined as the maximum load of thefailed sub-network which can be supplied through the adjacentsubstation whilst maintaining branch power flows within capacitylimits. The margin of a support j is calculated as the minimumamong the spare capacity of every element i of the adjacent sub-network on the path from the support to the substation:

marginj ¼miniðfþi � fiÞ ð1Þ

where fþi maximum power flow of element i; fi power flow of ele-ment i.

Fig. 1 shows a common meshed topology of a distribution net-work. If substation S1 trips, substation S1 can no longer supply thedownstream load. NDP is the Non-Delivered Power when S1 trips.In this case, it is assumed that two main supports, A and B, areavailable. The margin for support A would be:

marginA ¼minðfþ1 � f1; fþ2 � f2; fþ3 � f3; fsþ2 � fs2Þ ð2Þ

As Fig. 2 displays, when a main support is selected, some of theunsupplied load nodes are disconnected from the failed networkand the service is restored through the main support A. Then,neglecting energy losses, the power flow supplied by S2 increasesfrom fs2 to fs2 + f1 and NDP is reduced to NDP � f1.

Fig. 2. Main support.

88 A. González et al. / Electrical Power and Energy Systems 37 (2012) 86–94

2.3. Virtual congestion

During the reconfiguration process, if a main support is usedand the corresponding switching is operated, the violation of thecapacity or thermal limit of a line or transformer belonging to asub-network adjacent to the failed one may occur. Thus, the mainsupport becomes unfeasible and the violation of the limits is calledvirtual congestion. For instance, in the example shown in Fig. 2, avirtual congestion would exist if fs2 + f1 exceeded the thermal limitof the transformer in substation S2. In that case, the main support Awould be unfeasible.

2.4. Main supports sub-network and secondary supports sub-network

The term main supports sub-network is used to refer to everysub-network connected to the failed sub-network through an openbranch. In other words, a main supports sub-network provides asupport for the failed sub-network. In Fig. 1, the network down-stream from the substation S2 is a main supports sub-network.The term secondary supports sub-network is used to refer to everysub-network connected to a main sub-network through an openbranch. In other words, a secondary supports sub-network pro-vides a support to a main supports sub-network. In Fig. 3, the net-work downstream from the substation S3 is a secondary supportssub-network.

2.5. Secondary supports

Secondary supports are buses which can be used as a support tosolve virtual congestions and, hence, make feasible an unfeasiblemain support. Secondary supports are a useful resource to increasethe number of feasible main supports.

In Fig. 3, it is assumed that all the main supports available (Aand B) are not feasible since a thermal limit is violated in S2 ifany of them is used. Therefore, a virtual congestion exists in S2.

Fig. 4 shows the network configuration after having made amain support feasible by resorting to a secondary support C andhaving transferred f2 from the failed to the main supports sub-network. Fig. 5 shows the final network configuration after havingused the main support A.

Fig. 3. Unfeasible main supports and virtual congestion in substation S2.

2.6. Basic switching

Basic switching is the strategy adopted in this approach. The pro-posal is an iterative algorithm. In each iteration, the service is re-stored in part of the network (see Fig. 6) by means of a basicswitching. Basic switching entails closing a branch correspondingto a main support to transfer a single load from the failed networkto a main network, and opening the suitable branch in order toguarantee the radial configuration of the network. Basic switchingis also applied to transfer some load from a main network to a sec-ondary network so as to make the main support feasible. Finalreconfiguration implies the aggregation of some basic switchingactions.

3. Algorithm for distribution networks reconfiguration

This section provides an overview of the proposed algorithmand describes the two steps of the algorithm.

3.1. Overview

The aim of the algorithm is to find the network reconfigurationwhich searches the network configuration which minimizes theNDP in case of failure of a line, a transformer or a HV to MV substa-tion. The NDP related to a failure is the load which is not possible tosupply when the aforementioned failure occurs requiring thereconfiguration process to start.

Fig. 7 depicts an overview of the proposed algorithm of thereconfiguration of distribution networks. The algorithm has beendesigned as an iterative process, using heuristic rules. In each iter-ation, a temporary reconfiguration is obtained by using a mainsupport and an undetermined number of secondary supports ifthey are required. The process finishes when NDP is removed orall main and secondary supports are exhausted.

The security constraints considered are line and transformerspower flows and bus voltages. Additionally, the network must re-main radial. Within every iteration, the margin of each main andsecondary support is evaluated to check its feasibility ensuring thatmaximum power flows remain within the limits. After a powerflow study, the voltage of each bus is checked for each tentativeswitching. If the switching violates the minimum voltage con-straint, it is rejected. Finally, the radial configuration of thenetwork is ensured by the way basic switching is defined, as out-lined in the previous section.

The algorithm is comprised of two main steps where the state ofthe network is changed. In the first step, named Main Supports Step(MSS), feasible main supports are searched for and sorted indecreasing margin order. The main support with the highest marginis selected in first place, thus ensuring that the final reconfigurationis obtained with a low number of switching actions [33]. If theselected main support is feasible, the corresponding basic switchingis carried out as it is explained in Section 3.2. Following this

Fig. 4. Secondary support.

Fig. 5. Final reconfiguration.

D

no supply

D

D

SWITCHING

Fig. 6. Basic switching strategy.

Fig. 7. Overview of the flowchart of the reconfiguration algorithm.

Fig. 8. Main supports step flowchart.

A. González et al. / Electrical Power and Energy Systems 37 (2012) 86–94 89

switching, a new iteration is run. The second step, called SecondarySupports Step (SSS), is only launched if, within an iteration of theMSS, no feasible main support is available and some loads of the sys-tem remain unsupplied. The objective of the SSS is to solve the vir-tual congestion corresponding with the best main support on thatiteration (the one which, although being infeasible, could providethe highest restored load). The virtual congestion is solved by carry-ing out an undetermined number of basic switching operationswhich make a main support feasible by transferring some load fromthe main supports sub-network to a secondary supports sub-net-work. SSS process is detailed in Section 3.3. If the selected main sup-port is not able to come feasible by means of secondary supports, anew iteration of the SSS is carried out in order to make feasible thefollowing best main support. The SSS ends when the selected mainsupport becomes feasible or when no unfeasible main support canbe made feasible. In the first case, a new MSS iteration starts andthe selected main support is used since it has already come feasible.Otherwise, the reconfiguration problem ends.

3.2. Main supports step

The main supports step flowchart is shown in Fig. 8. The MSS isan iterative routine. Firstly, all the feasible main supports are

searched. The support with the highest margin is selected for thereconfiguration. Then, the selected supporting line is closed. Inorder to maintain the network radial, the line which connects therecovered load and the failed sub-network is opened as shown inFig. 5. The highest margin is the criterion to select a specificswitching because it benefits the load balancing. Moreover, basicswitching strategy plus this criterion ensure that the final reconfig-uration is obtained with a low number of switching actions [33]. Asmentioned, final switching actions will aggregate a number ofbasic switching actions. The higher the margin of the selectedswitching is, more basic switching actions are aggregated, and lessfinal switching operations have to be done.

After performing a switching, a power flow is run and the volt-age and the current flow constraints are checked for every nodeand branch of the involved sub-networks. If a constraint is violated,the switching is reversed, and the next best switching is tested.When a switching respects the constraints, the new network con-figuration is analyzed, and the NDP is checked. If necessary, newunsupplied load nodes are defined and a new iteration is carriedout.

The MSS finishes when no more supports are found or when nomore NDP remain.

3.3. Secondary supports step

The secondary supports flowchart is shown in Fig. 9. SSS is aniterative routine like MSS. SSS is run if, within an iteration of theMSS, some loads of the system remain unsupplied and all feasiblemain supports are used up. The objective of the SSS is to alleviatevirtual congestions so as to make an unfeasible main supportfeasible.

The unfeasible main supports sub-network to make feasible inthe SSS is the one which could provide the feasible main supportwith the biggest expected restored load in the next MSS iteration.

The search of secondary supports is done similarly to the mainsupports search. The device where the virtual congestion takesplace is the one which needs to be supported. The stop criteria ishaving transferred enough loads to make the main supportfeasible.

When the virtual congestion is relieved, the MSS is run again. Ifall the secondary supports to a particular virtual congestion are

Fig. 9. Secondary supports step flowchart.

90 A. González et al. / Electrical Power and Energy Systems 37 (2012) 86–94

used up, and the virtual congestion is not solved, all the switchingoperations carried out in this iteration are reversed and anothermain support and virtual congestion are analyzed.

4. Applicability

The results provided by the algorithm are: switching operationsfor network reconfiguration after any of the simulated failures,places where a sectionalizing device must exist to obtain the pro-vided reconfiguration, reallocated load due to main supports andwith the possible contribution of secondary supports, and finalNDP. These results are useful for operation and planning analyses.The suitable interpretation is explained below.

� The final network configuration and the restored load after eachswitching provides useful information for operation program-ming since these results must be known to design an optimalservice restoration process.� The failure of every device in the network (transformer, substa-

tion or line) can be simulated; therefore the algorithm is usefulfor the security analysis in a real distribution network.� Improving quality of service may require new investments if

NDP cannot be removed with a reconfiguration process. Thereare two reasons for the existence of NDP after an optimal recon-figuration: the absence of main supports or the existence ofunfeasible main supports because of violations of voltage orthermal limits. The proposed algorithm provides informationabout the feasible support provided by each sub-network. Asa consequence, the weaknesses of each sub-network can bedetected and results may help in planning analysis.� Since provided results on the reconfiguration process (the

opened and closed branches) are presented based on the pro-cess followed by the algorithm, the solution process comesinterpretable and some results may be obtained from partialresults. Moreover, the interpretability of the reconfigurationprocess may highlight suitable locations for additional section-alizing devices.

� Because a successful re-establishment of the service for a givenfailure does not guaranty the success of a failure downstreamfrom the first one, the algorithm has been adapted to automateall possible failures on a meshed distribution network. Shortcomputation times make this enumeration possible for plan-ning operation and planning tasks.

The algorithm is currently being used in planning and operationof a MV distribution network in Madrid with more than 24.000buses.

5. Case study

The application of the algorithm will be illustrated for part20 kV distribution network in Madrid. This network has been se-lected because its size may prove that the algorithm is efficientwhen it runs an actual, complex and large-scale network. More-over, this network is partially composed of long branches; there-fore the case study shows that the algorithm is able to cope withvoltage limit violations.

Due to the network being a real one, the loads have been mod-ified for reasons of privacy. The main characteristics of the networkare:

� 2730 buses.� 3010 branches.� 11 132/20 kV transformers.� 1597 loads.

The remainder of this section is organized as follows. In Section5.1, the failure of a transformer has been simulated in order toshow the results of the reconfiguration process. Afterwards, in Sec-tion 5.2, the failure of each transformer of the network has beensimulated to obtain results about the NDP of the considered net-work. The results are analyzed in order to show their applicabilityfor operation and planning decisions.

5.1. Reconfiguration process

In this subsection, the reconfiguration process in case of a trans-former failure is simulated. The results obtained are helpful foroperation planning analysis since they provide the switching oper-ations suitable to obtain the network configuration which wouldallow the maximum load to be supplied after the failure of thetransformer.

The sub-network supplied by the transformer MECOT4 beforethe failure consists of 228 buses and was supplying 26.22 MW.For this application, the complete 20 kV level has been includedin the network model. The boundary buses are the 20 kV bars ofthe 132/20 kV substations. A simplified diagram of the pre-faultnetwork is shown in Fig. 10. Only the out of service zone, the mainsupports and the secondary supports which are used in the prob-lem have been represented.

During the process, 98 power flow analysis, 54 iterations for themain supports routine, and 55 iterations for the secondary sup-ports routine have been carried out. The partial reconfigurationof the network, if the secondary supports are not used, is obtainedby opening four branches and closing four branches. After thisstage, 22.3 MW are still unsupplied. After nine steps, the finalreconfiguration of the network is obtained by opening 11 branchesand closing 11 branches (see Table 1). At the end of the whole pro-cess, 17.9 unsupplied MW remain. Final network configuration isshown in Fig. 11.

As explained in Section 4, a sectionalizing device is modeled inevery branch of the network in order to obtain their optimal loca-tions to solve a particular failure. In this case study, in order to

Fig. 10. Network before the reconfiguration process.

A. González et al. / Electrical Power and Energy Systems 37 (2012) 86–94 91

obtain the computed solution, a sectionalizing device must exist inevery branch which changes its state at the final reconfigurationwith respect to the initial configuration. In the presented casestudy, branches which are opened during the reconfiguration pro-cess are: 1, 2, 3, 5, 6, 7, 8, 9, 10, 14, 15, 16, 18 and 19. If any of themare not possible to operate, new sectionalizing devices must be in-stalled to carry out the obtained reconfiguration process.

5.2. Non-Delivered Power

In this section, the trip of each transformer of the considerednetwork is simulated. As a result, the NDP is calculated for eachsimulation. The results obtained in this section are useful for plan-ning tasks since they are required to carry out the quality analysisand to plan new quality-driven reinforcements which provide therestoration of all loads.

For the simulation of each transformer failure, the service resto-ration process is carried out as explained in Section 5.1. Therestored load is calculated both when only main supports are con-sidered and when main and secondary supports are considered.This information allows knowing what transformers are well sup-ported and which ones should be reinforced to avoid NDP in case offailure. Fig. 12 shows the percent values of restored load if only themain supports are considered and when both main and secondarysupports are taken into account. This information is useful to eval-uate how secondary supports may be a solution to find a correctivereconfiguration. Moreover, considering secondary supports in-creases efficiency and may defer or avoid new investments.

Table 2 complements the aforementioned information. Foreach transformer trip, the unsupplied load is calculated when onlymain supports are considered and when secondary supports arealso take into account. For example, when the transformer ARDOZN. T1 trips, the entire load is restored just by considering mainsupports. Other transformer trips are covered only by main sup-ports since secondary supports are not required. After the trans-former TOROTE T2 failure is simulated, the entire load is alsorestored, but secondary supports are necessary in this case. How-ever, the simulation of other failures, as for example the cases ofTOROTE T1 or MECO T4 (see complete simulation results in Sec-tion 5.1), show NDP after the restoration process. Herein, somereinforcements should be done for the main or for the secondarynetworks.

5.3. Algorithm performance

The Heuristic Algorithm we propose provide solutions foractual, complex and large-scale networks in a few seconds for mostcritical cases (e.g., a transformer failure as in Section 5.2. may re-sult in transferring more than 150 loads to other main feeders,which also feed a high number of loads). Table 3 shows CPU timesfor transformer failures analyzed in 5.2. The longest computationtime obtained for the network studied in 5.2 is 24 s. Results havebeen obtained implementing the algorithm on Python within agraphical user interface program on an Intel Core 2 CPU 4300 @1.79 GHz with 1.99 GB of RAM.

Table 1Switching actions after a transformer failure.

Branches to open Branches to close Number of transferred load buses Transferred load (MW)

First stage (main supports)6 11 8 1.7

10 17 9 1.238 12 4 0.289 13 4 0.9

Second stage (secondary supports)22 23 4 21.9316 21 8 29.56

Third stage (main supports)1 6 16 1.747 9 12 1.04

Fourth stage (secondary supports)19 22 1 2.0718 20 1 1.47

Fifth stage (main supports)5 7 10 0.49

Sixth stage (secondary supports)15 18 6 13.52

Seventh stage (main supports)3 4 9 0.82

Eighth stage (secondary supports)14 15 3 2.55

Ninth stage (main supports)2 3 3 0.5

Fig. 11. Network after the reconfiguration process.

92 A. González et al. / Electrical Power and Energy Systems 37 (2012) 86–94

0% 50% 100%

ALCALA T2

ALCALA T3

ARDOZ N. T1

ARDOZ N. T2

ARDOZ N. T3

ARDOZ N. T4

MECO T4

TOROTE T1

TOROTE T2

VENTEROS T1

VENTEROS T2

Re-allocated load (%)

Tra

nsfo

rmer

Main supports Main + Secondary supports

Fig. 12. Re-allocation of the loads of each transformer considering and notconsidering secondary supports.

Table 2Results for each transformer trip.

Transformer CPU time (s)

ALCALA T2 13.5ALCALA T3 10.3ARDOZ N. T1 4.6ARDOZ N. T2 4.7ARDOZ N. T3 4.9ARDOZ N. T4 7.9MECO T4 11.2TOROTE T1 15.4TOROTE T2 24.4VENTEROS T1 11.1VENTEROS T2 9.2

Table 3CPU time for each transformer trip.

Transformer Initialunsuppliedload (MW)

Unsupplied load aftermain supports (MW)

Unsupplied load aftersecondarysupports (MW)

ALCALA T2 38.29 25.4 18.31ALCALA T3 37.14 23.44 21.41ARDOZ N. T1 16.06 – –ARDOZ N. T2 16.14 – –ARDOZ N. T3 32.73 – –ARDOZ N. T4 37.54 – –MECO T4 26.22 22.3 17.9TOROTE T1 38.69 8.51 8.51TOROTE T2 20.56 11.8 –VENTEROS T1 20.13 15.39 15.22VENTEROS T2 15.26 14.54 14.54

A. González et al. / Electrical Power and Energy Systems 37 (2012) 86–94 93

6. Conclusions

A heuristic optimization algorithm for reconfiguration of MVdistribution networks has been presented. The reconfigurationproblem consists in minimizing the Non-Delivered Power in caseof failure of a line, a transformer or a complete substation. The con-straints considered are that the radial structure of the networkmust be maintained and that line thermal limits, transformercapacities and bus voltages must be within their admissible ranges.The proposed algorithm obtains a new network configuration byapplying heuristic rules based on real operation procedures. As aresult, good estimations of NDP are obtained. The algorithmachieves quasi-optimal results for complex and large-scale distri-bution networks in short computation times. The reconfiguration

process and the results become understandable and reliable for po-tential users. The results provided by the reconfiguration analysisare useful for both planning and operation tasks.

The advantages of the proposed algorithm are admissible com-putation times for planning applications and the interpretability ofthe solution (since results can be presented based on the processfollowed by the algorithm). Moreover, it is remarkable the useful-ness to assess distribution companies for planning tasks and toevaluate the reliability performance of their networks, since actualrules followed by distribution operators during emergency statesof the network have been implemented.

The algorithm has been successfully tested in complex andlarge-scale real MV distribution networks. Two different analyseshave been described in the paper. The first one provides the servicerestoration process in case of a transformer failure. The resultsshow the switching operations needed to reconfigure the networkand the restored buses and load. This analysis is useful for opera-tion and operation programming tasks. Since a sectionalizing de-vice is modeled for every branch of the network, the first resultsare also useful in taking decisions about investments on new sec-tionalizing devices. The second case study analyzes the reliabilityof the network. The nonserved load is assessed for the single failureof every transformer. The NDP is computed both with and withoutconsidering the secondary supports. This analysis is useful forinvestment planning tasks.

Acknowledgements

The work reported in this paper has been fully supported byIberdrola Distribucion SAU. The project was started thanks toRaimundo Criado’s iniciative.

References

[1] Aoki K, Satoh T, Itoh M, Kuwabara H, Kanezashi M. Voltage drop constrainedrestoration of supply by switch operation in distribution systems. IEEE TransPower Delivery 1988;3(3):1267–74.

[2] Lin WM, Chin HC. A new approach for distribution feeder reconfiguration forloss reduction and service restoration. IEEE Trans Power Delivery1998;13(3):870–5.

[3] Popovic DS, Ciric RM. A multi-objective algorithm for distribution networksrestoration. IEEE Trans Power Delivery 1999;14(3):1134–40.

[4] Taylor T, Lubkeman D. Implementation oh heuristic search strategies fordistribution feeder reconfiguration. IEEE Trans Power Delivery1990;5(1):239–46.

[5] Zhou Q, Shirmohammadi D, Liu WHE. Distribution feeder reconfiguration forservice restoration and load balancing. IEEE Trans Power Syst1997;12(2):724–9.

[6] Glamocanin V. Optimal loss reduction of distribution Networks. IEEE TransPower Delivery 1990;5(3):774–80.

[7] Goswami SK, Basu SK. A new algorithm for the reconfiguration of distributionfeeders for loss minimization. IEEE Trans Power Delivery 1992;7(3):1484–91.

[8] Peponis GJ, Papadopoulos MP, Hatziargyriou ND. Distribution networkreconfiguration to minimize resistive line losses. IEEE Trans Power Delivery1995;10(3):1338–42.

[9] Savier JS, Das D. Loss allocation to consumers before and after reconfigurationof radial distribution networks. Int J Electr Power Energy Syst 2011;33(3):540.

[10] Bernardon DP, Garcia VJ, Ferreira ASQ, Canha LN. Electric distribution networkreconfiguration based on a fuzzy multi-criteria decision making algorithm.Electr Power Syst Res 2009;79(10):1400.

[11] Berredo RC, Ekel PY, Martini JSC, Palhares RM, Parreiras RO, Pereira Jr JG.Decision making in fuzzy environment and multicriteria power engineeringproblems. Int J Electr Power Amp Energy Syst 2011;33(3):623.

[12] Augugliaro A, Dusonchet L, Mangione S, Sanseverino ER. Fast solution of radialdistribution networks with automated compensation and reconfiguration.Electr Power Syst Res 2000;56(2):159.

[13] Brown RE, Hanson AP, Willis HL, Luedtke FA, Born MF. Assessing the reliabilityof distribution systems. IEEE Comput Appl Power 2001;14(1):44–9.

[14] Tsai LH. Network reconfiguration to enhance reliability of electric distributionsystems. Electr Power Syst Res 1993;27:135–40.

[15] González A, Echavarren FM, Rouco L, Gómez T, Cabetas J. Fault location andservice restoration for large-scale distribution networks, presented at powersystems computation conference (PSCC11), Stockholm; 2011.

[16] IEEE guide for electric power distribution reliability indices. IEEE Std 1366,2001 Edition.

94 A. González et al. / Electrical Power and Energy Systems 37 (2012) 86–94

[17] Merlin A, Back H. Search for a minimal-loss operating spanning treeconfiguration for an urban power distribution system, presented at powersystems computing conference, Cambridge; 1975.

[18] Nagata T, Sasaki H, Kitagawa M. A method of determining target configurationfor power systems restoration by means of mix integer programmingapproach. Trans IEE Jpn 1994;114-B(2):179–85.

[19] Shirmohammadi D. Service restoration in distribution networks via networkconfiguration. IEEE Trans Power Delivery 1992;7(2):952–8.

[20] Zhou Q, Shirmohammadi D, Liu WHE. Distribution feeder reconfiguration forservice restoration and load balancing. IEEE Trans Power Syst 1997;12(2):724–9.

[21] Fukuyama Y, Hsaio-Dong C. A parallel genetic algorithm for service restorationin electric power distribution systems, presented at IEEE internationalconference on fuzzy systems; 1995.

[22] Mori H, Takeda K. Parallel simulated annealing for power systemdecomposition. IEEE Trans Power Syst 1994;9(2):789–95.

[23] Mori H, Ogita Y. A parallel tabu search based approach to optimal networkreconfigurations for service restoration in distribution systems, presented at2002 IEEE international conference on control applications, Glasgow, Scotland,UK; September 18–20, 2002.

[24] Toune S, Fudo H, Genji T, Fukuyama Y, Nakanishi Y. A reactive tabu search forservice restoration in electric power distribution systems, presented at IEEEworld congress on computational intelligence; May 1998.

[25] Toune S, Fudo H, Genji T, Fukuyama Y, Nakanishi Y. Comparative study ofmodern heuristic algorithms to service restoration in distribution systems,presented at IEEE power engineering society winter meeting, November 2001.

[26] Lin WM, Huang CH, Lin CH, Chen CM, Wan LW. Service restoration forsecondary power system with grey relational analysis, presented at IEEETENCON, Hong Kong; November 2006.

[27] Hsu YY, Kuo HC. A heuristic based fuzzy reasoning approach for distributionsystem service restoration. IEEE Trans Power Delivery 1994;9(2):948–53.

[28] Huang CM. Multiobjective service restoration of distribution systems usingfuzzy cause-effect networks. IEEE Trans Power Syst 2003;18(2):867–75.

[29] Kuo HC, Hsu YY. Distribution system load estimation and service restorationusing a fuzzy set approach. IEEE Trans Power Delivery 1993;8(4):1950–7.

[30] Chen CS, Lin CH, Tsai HY. A rule-based expert system with colored petri netmodels for distribution system service restoration. IEEE Trans Power Syst2002;17(4):1073–81.

[31] Lim SI, Lee SJ, Choi MS, Lim DJ, Ha BN. Service restoration methodology formultiple fault case in distribution systems. IEEE Trans Power Syst2006;21(4):1638–44.

[32] Miu KN, Chiang HD, Yuan B, Darling G. Fast service restoration for large-scaledistribution systems with priority customers and constraints. IEEE TransPower Syst 1998;13(3):789–95.

[33] González A, Echavarren FM, Rouco L, Gómez T, Cabetas J. A tool forreconfiguration of large-scale distribution networks, presented at16th power system computation conference (PSCC08), Glasgow, Scotland;2008.