This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Research ArticleComparing Evolutionary Strategies ona Biobjective Cultural Algorithm
Ricardo Soto14 Joseacute-Miguel Rubio15 and Fernando Paredes6
1 Escuela de Ingenierıa Informatica Pontificia Universidad Catolica de Valparaıso 2362807 Valparaıso Chile2 Universidad Finis Terrae 7500000 Santiago Chile3 CIMFAV Facultad de Ingenierıa Universidad de Valparaıso 2362735 Valparaıso Chile4Universidad Autonoma de Chile 7500138 Santiago Chile5 Departamento de Computacion e Informatica Universidad de Playa Ancha 33449 Valparaıso Chile6 Escuela de Ingenierıa Industrial Universidad Diego Portales 8370109 Santiago Chile
Correspondence should be addressed to Carolina Lagos carolinalagoscmailpucvcl
Received 9 April 2014 Accepted 27 June 2014 Published 31 August 2014
Academic Editor Xin-She Yang
Copyright copy 2014 Carolina Lagos et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Evolutionary algorithms have been widely used to solve large and complex optimisation problems Cultural algorithms (CAs) areevolutionary algorithms that have been used to solve both single and to a less extentmultiobjective optimisation problems In orderto solve these optimisation problems CAs make use of different strategies such as normative knowledge historical knowledgecircumstantial knowledge and among others In this paper we present a comparison among CAs that make use of differentevolutionary strategies the first one implements a historical knowledge the second one considers a circumstantial knowledge andthe third one implements a normative knowledge These CAs are applied on a biobjective uncapacitated facility location problem(BOUFLP) the biobjective version of the well-known uncapacitated facility location problem To the best of our knowledge onlyfew articles have applied evolutionary multiobjective algorithms on the BOUFLP and none of those has focused on the impactof the evolutionary strategy on the algorithm performance Our biobjective cultural algorithm called BOCA obtains importantimprovements when compared to other well-known evolutionary biobjective optimisation algorithms such as PAES and NSGA-IIThe conflicting objective functions considered in this study are cost minimisation and coverage maximisation Solutions obtainedby each algorithm are compared using a hypervolume S metric
1 Introduction
Evolutionary algorithms (EAs) are an effective alternative to(approximately) solve several large and complex optimisationproblems as they are able to find good solutions for awide range of problems in acceptable computational timeAlthough less studied EAs for multiobjective optimisation(MO) problems called evolutionarymultiobjective optimisa-tion (EMO) have demonstrated to be very effective In factduring the last two decades several authors have focusedtheir efforts on the development of several EMO algorithmsto solve a wide range ofMO problems For instance Maravall
and de Lope [1] use a genetic algorithm (GA) to solve themultiobjective dynamic optimisation for automatic parkingsystem In [2] the authors propose an improvement to thewell-known NSGA algorithm (called NSGA-II) based on anelitist approach In [3] the author presents an EMO algorithmapplied on a specific variation of the well-studied capacitatedvehicle routing problem (CVRP) where the author includesin the EMO algorithm an explicit collectivememorymethodnamely the extended virtual loser (EVL) Other well-knownEMO algorithms developed during the last two decades arePAES [4] and MO particle swarm optimisation [5] Morerecently hybrid techniques have been also applied to a large
Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 745921 10 pageshttpdxdoiorg1011552014745921
2 The Scientific World Journal
number of optimisation problems (see [6]) A comprehensiveliterature review related to EMO algorithms can be found in[7]
EMO algorithms have some problems that must be takeninto account though For instance they tend to fall intopremature convergence with low evolution efficiency [8]This is because of implicit information embodied in theevolution process and domain knowledge corresponding tooptimisation problems which are not fully included in thesolution approach [9] To overcome these problems onecan make use of implicit evolution information Reynolds[10] proposes an EA called cultural algorithm (CA) whichis inspired from human culture evolution process and thatmakes use of the implicit evolution information generated ateach iterationTheCAs have a dual evolution structure whichconsists of two spaces population and belief space On theone hand the population space works as in any other EAthat is using evolutionary operators such as mutation andcrossover On the other hand in the belief space implicitknowledge is extracted from selected individuals in thepopulation and stored in a different way Then they are usedto guide the whole evolution process in the population spacesuch that they can induce the population to escape fromlocal optimal solutions It has been proved that CAs caneffectively improve the evolution performance [9] Althoughless studied CAs have been also used to solve MO problemsCoello et al [11] andCoello et al [12] two remarkable surveysonCAs onlymentionCoello and Landa [13] work as exampleof a CA application solving MO problems More recentlyZhang et al [14] present a CA which is enhanced by usinga particle swarm optimisation algorithm This enhanced CAis applied to fuel distribution MO problem Srinivasan andRamakrishnan [15] present a MO cultural algorithm that isapplied on data mining domain In [16] the authors applieda CA to a biobjective portfolio selection problem usingnormative knowledge in the belief space In [17] authorspresent a formal framework to implement MO culturalalgorithms To the best of our knowledge [18] is the onlyarticle that uses CAs to solve the biobjective uncapacitatedfacility location problem (BOUFLP) Furthermore we didnot find any article which compares the performance of CAsusing different evolutionary strategies at the belief space levelThus in this paper we present an extension of the biobjectivecultural algorithm (BOCA) developed in [18] We use twodifferent strategies at the belief space level and compare theperformance of our new algorithms with the performanceof the previous one We also compare its results with otherwell-known EMO algorithms such as PAES and NSGA-IIObtained solutions were compared using the hypervolume Smetric proposed in [19]
The remaining of this paper is organised as followsSection 2 shows an overview on MO focused on EMOalgorithms Section 21 presents briefly the BOUFLP andsome of its distinctive features In Section 3 we describe ourimplementation for the BOCA algorithm We describe themain differences between our implementation and the one in[18] Section 32 presents the BOCA algorithm applied to a setof well-known instances from the literature Finally Section 4presents the conclusions of this work
2 Overview
In this section we show an overview of topics relatedto this paper In Section 21 MO concepts are presentedemphasizing EMO algorithms and its state of art Morespecifically we focus on the development of EMO algorithmsfor MO combinatorial optimisation (MOCO) problems InSection 22 we present the BOUFLP formulation based on acost-coverage approach Finally in Section 23 we present anoverview of CAs and its multiobjective extension Details ofour CA implementation are also presented at the end of thissection
21 (Evolutionary) Multiobjective Optimisation In this sec-tionwe briefly introduce themain principles ofMOproblemsand particularly MOCO problems For a comprehensivereview of this topic see [20 21] In this paper we will makeuse of the following notation for the comparison of vectors(solutions) Let 119910 and 119910
1015840isin R119901 We say that 119910 ≦ 119910
1015840 if119910119896≦ 1199101015840
119896forall119896 = 1 119901 Similarly we will say that 119910 le 119910
1015840 if119910 ≦ 119910
1015840 but 119910 = 1199101015840 Finally we say that 119910 lt 119910
1015840 if 119910119896lt 1199101015840
119896forall119896 =
1 119901 A solution 119909 isin R119899 with 119899 being equal to the numberof decision variables is called an efficient solution and itsimage 119891(119909) with 119891 R119899 rarr R119901 a nondominated point ofthe MO problem if there is no 119909 isin R119899 with 119909 = 119909 such that119891(119909) le 119891(119909) In [22] the author describes several excellencerelations These relations establish strict partial orders in theset of all nondominated points related to different aspectsof their quality Previously in [23 24] the authors considerseveral outperformance relations to address the closenessof the set of nondominated points found by an algorithmto the actual set of nondominated points called ParetoFrontier (PF) In [25] a comprehensive explanation of thedesirable features of an approximation to the PF is presentedIn this paper we choose the S metric which is properlyexplained in [24] The S metric calculates the hypervolumeof a multidimensional region [19] and allows the integrationof aspects that are individually measured by other metricsAn advantage of the S metric is that each algorithm can beassessed independently of the other algorithms involved inthe study However the S values of two setsA andB namelySA and SB respectively cannot be used to derive whethereither set entirely dominates the other Figure 1(a) shows asituation where SA gt SB and setA completely dominates setB Figure 1(b) shows a situation where SB gt SA but neitherA dominatesB norB dominatesA
In this paper we use EAs to solve the BOUFLP An EAis a stochastic search procedure inspired by the evolutionprocess in nature In this process individuals evolve and thefitter ones have a better chance of reproduction and survivalThe reproduction mechanisms favour the characteristics ofthe stronger parents and hopefully produce better childrenguaranteeing the presence of those characteristics in futuregenerations [26] EAs have been successfully applied to alarge number of both single- andmultiobjective optimisationproblems Comprehensive reviews of EMO algorithms arepresented in [11 24] andmore recently in [12] Amore general
Figure 1 S metric represented as the area that is dominated by a setof (approximately) nondominated points
review of hybrid heuristics solving MOCO problems whereEMO algorithms are also included is presented in [27]
22 Biobjective Uncapacitated Facility Location ProblemFacility location problem (FLP) is one of the most importantproblems for companies with the aim of distributing productsto their customers The problem consists of selecting sites toinstall plants warehouses and distribution centres allocatingcustomers to serving facilities and interconnecting facilitiesby flow assignment decisions Comprehensive reviews andanalysis of the FLP are presented in [28ndash30]
In this paper we consider a two-level supply chain wherea single plant serves a set of warehouses which serve a set ofend customers or retailers Figure 2 shows this configuration
Two main (conflicting) objectives can be identified in theFLP
(i) minimise the total cost associated with the facilityinstallation and customer allocation and
(ii) maximise the customers rate coverage
Several works on single-objective optimisation have beencarried out considering these two objectives separately Onthe one hand uncapacitated FLP (UFLP) is one of moststudied FLPs in the literature In the UFLP the main goalis to minimise the location-allocation cost of the networkOn the other hand 119901 median FLPs are one of the mostcommon FLPs among those that are focused on coveragemaximisationMost important FLPmodels arewell describedand formalised in [29] MO FLPs have been also well studiedin the literature during the last two decades A survey on thistopic can be found in [31]
As we mentioned before in this paper we solve theBOUFLP BOUFLP has been modelled with minisum andmaxisumobjectives (cost and coverage)The followingmodelformulation is based on [32] Let 119868 = 1 119898 be the set ofpotential facilities and 119869 = 1 119899 the set of customersLet FC
119894be the fixed cost of opening facility 119894 and 119889
119895the
demand of customer 119895 Let 119888119894119895be the cost of assigning the
customer 119895 to facility 119894 and ℎ119894119895the distance between facility 119894
and customer 119895 Let119863MAX be the maximal covering distancethat is customers within this distance to an open facility areconsidered well served Let 119876
119895= 119894 isin 119868 ℎ
119894119895le 119863MAX
be the set of facilities that could serve customer 119895 within themaximal covering distance 119863MAX Let 119909119894 be 1 if facility 119894 isopen and 0 otherwise Let 119910
119894119895be 1 if the whole demand of
customer 119895 is served by facility 119894 and 0 otherwise Consider
with 119891 R119899 rarr R119901 and 119901 = 2 Objective functions 1198911and 119891
2
are as follows
1198911(119909 119910) = sum
119894isin119868
FC119894119909119894+ sum
119894isin119868
sum
119895isin119869
119888119894119895119910119894119895 (6)
1198912(119909 119910) = minussum
119895isin119869
119889119895sum
119894isin119876119895
119910119894119895 (7)
Equation (6) represents total operating cost the first termcorresponds to location cost that is the sum of the fixed costsof all the open facilities and the second term represents theallocation cost that is the cost of attending customer demandby an open facility Equation (7) measures coverage as thesum of the demand of customers attended by open facilitieswithin the maximal covering distance Equations (2) and (3)
4 The Scientific World Journal
Plant or centralwarehouse
Warehouse 1
Warehouse 2
Warehouse 3
Customers(clusters)
Figure 2 A two-level supply chain network configuration
ensure that each customer is attended by only one facilityEquation (3) also forces customer to be assigned to an openfacility Finally equations (4) and (5) set decision variables asbinary
23 Biobjective Cultural Algorithm The experience andbeliefs accepted by a community in a social system are themain motivations for the creation of the CAs Originally pro-posed by Reynolds [10] CAs model the evolution of culturalsystems based on the principles of human social evolutionIn this case evolution is seen as an optimisation process [10]The CAs guide the evolution of the population based on theknowledge Knowledge acquired during previous iterationsis provided to future generations allowing accelerating theconvergence of the algorithm to good solutions [33] Domainknowledge is modelled separately from the populationbecause there is certain independence between them whichallows us to work and model them separately in order toenhance the overall algorithm performance Figure 3 showsthis interaction
CAs are mainly characterised by presenting two inher-itance systems one at population level called populationspace and the other at knowledge level called belief spaceThis key feature is designed to increase the learning ratesand convergence of the algorithm and thus to do a moreresponsive system for a number of problems [34] Moreoverit allows us to identify two significant levels of knowledge
a microevolutionary level (represented by the populationspace) andmacroevolutionary level (represented by the spaceof beliefs) [35]
CAs have the following components population space(set of individuals who have independent features) [35] beliefspace (stored individuals acquired in previous generations)[34] computer protocol connecting the two spaces anddefining the rules on the type of information to be exchangedbetween them by using the acceptance and influence func-tions and finally knowledge sources which are describedin terms of their ability to coordinate the distribution ofindividuals depending on the nature of a problem instance[35] These knowledge sources can be of the followingtypes circumstantial normative domain topographic andhistorical
The most distinctive feature of CAs is the use of thebelief space which through an influence function affectsfuture generations For this reason in this paper we focuson the effect on the algorithm performance of changes insuch an influence function To do this we have consideredresults obtained previously in [18] where the authors usedan influence function based on historical knowledge andwe compare those results with our BOCA implementationwhich considers two influence functions the first one basedon circumstantial knowledge and the second one basedon normative knowledge Algorithm 1 shows the generalprocedure of our BOCA algorithm
The Scientific World Journal 5
Belief space
Population space
Update ()
Accept ()Influence ()
New generation ()
Best individuals ()
2
2
2
1
1
1
n minus 1 n
n minus 1 n
n minus 1 n
middot middot middot
middot middot middot
middot middot middot
middot middot middot
Figure 3 CA general diagram
begin119896 = 0initialise Popuation 119875
119896
initialise BeliefSpace 119861119896
while 119896 lt 119894119905119890119903119886119905119894119900119899119873119906119898119887119890119903 doEvaluate(119875
To initialise the population we use a semirandom func-tion In its first phase this function defines in a stochasticway the set of facilities that will be opened (selected facili-ties) Then we allocate each customer to a selected facilityminimising the cost function 119891
1while avoiding minimising
the coverage function 1198912 This strategy provides better results
than using completely random initial populations and itscomputational time additional cost is marginal
To obtain the next generation two parents are used in arecombination process To avoid local optimal values we donot overuse the culture Thus a parent is selected from thepopulation to obtain diversity and the other parent is selected
from the belief space to influence the next generation Thebelief space keeps a list of all the individuals which meetsome criteria These criteria depend on what knowledgethe algorithm implements In this paper the circumstantialknowledge selects the best individuals found so far foreach objective function Thus one individual will give usinformation on the best value found for 119891
1and the other will
do the same for 1198912 The historical knowledge stores a list of
individuals with the best fitness value found so farThe fitnessvalue is calculated as the hypervolume S that is covered by anindividual Finally normative knowledge considers a list ofindividuals which are pairwise nondominated with respect tothe other individuals of their generation
Let |119869| be the number of available facilities and let |119868|
be the number of customers of our BOUFLP In this paperdecisions variables 119909 and 119910 are represented by a binary |119869|-length vector and |119869| times |119868| matrix respectively Comparingtwo different solutions (individuals) needs an evaluationcriterion In this paper we use the same criterion explainedin [18]
3 Computational Experiments
In this section we present the set of instances that are usedin this study as well as results obtained by our BOCAimplementation
31 Instances Presentation The instances that were used inthis paper correspond to random instances using a problemgenerator that follows the methodology from UflLib [36]Previous works in the literature have also used this problemgenerator to create their test instances [18 26]
The BOCA algorithm has several parameters that need tobe set As in [18] the number of generations considered inthis paper is equal to 100 Population size 119871 is set equal to 100mutation probability in the population space 119875ps is equal to02 and probability of mutation in the belief space 119875bs is 004Both 119871 and 119875ps values are different from the values used in[18] These values are chosen as they all together yield to thebest performance of the algorithm given some test instancesThus although resulting values are different from that used in[18] the method we use to set them is the same as that usedin that work This is important in order to fairly compare thedifferent BOCA algorithms
32 Results and Discussion In this section we compare theresults obtained by the previous BOCAalgorithm (see [18] forfurther details) and our approach Moreover a comparisonbetween results obtained by well-known EMO algorithmssuch as NSGA-II and PAES and our BOCA algorithm is alsopresented in this section
Tables 1 and 2 show the results obtained by the BOCAimplementations using historical [18] circumstantial andnormative knowledge respectively In the same way Tables 3and 4 present the results obtained by the well-known NSGA-II and PAES algorithms For each algorithm 119878 value () time119905 (in seconds) and the number of efficient solutions 119909 isin 119883
have been included in these tables As we mentioned before
6 The Scientific World Journal
Table 1 Results obtained by the BOCA implementations for instance of class 119860
Instance BOCA (historical) BOCA (circumstantial) BOCA (normative)1199051(sec) S
we want to produce a set with a large number of efficientsolutions 119909 isin 119883 a 119878 value close to 100 (ideal) and a small 119905For the sake of easy reading we have split the set of instancesinto two subsets (instances type 119860 and 119861)
We then compare our BOCA implementations with theone presented in [18] Tables 5 and 6 show a comparisonbetween those algorithms As we can see when compared interms of its 119878 value (the bigger the better) BOCA algorithm
using historical knowledge (BOCAH) performs consistentlybetter than the ones using circumstantial (BOCAC) andnormative (BOCAN) knowledge In fact BOCAH obtains a 119878
value that is in average 58 bigger than the one obtainedby BOCAC and 65 bigger than the 119878 value obtained byBOCAN When compared in terms of the CPU time neededto reach the number of iterations (generations) BOCAH isin average faster than both BOCAC and BOCAN algorithms
The Scientific World Journal 7
Table 3 Results obtained bywell-knownMOEA algorithmsNSGA-II and PAES for instances of class 119860
We can note that for 119861 instances times required by BOCAH
and BOCAC are in average quite similar (only 16 ofdifference) Finally when we look at the number of efficientsolutions found by each algorithm (|119883| column) we can seethat again BOCAH outperforms both BOCAC and BOCAN
algorithms In this case the average number of efficient
Table 5 Comparison among our BOCA implementations (119860instances)
solutions found by the BOCAH algorithm is about 20biggerthan the one obtained by the other two approaches
Results above are consistent with the good performanceobtained by the BOCAH approach in [18] Moreover resultsshow that performance of the BOCA algorithm dependslargely on the selected knowledge and it can make thedifference in terms of 119878 value time and number of efficientsolutions found by the algorithmThis is an important finding
8 The Scientific World Journal
Table 7 Comparison among our BOCAwith circumstantial knowl-edge and NSGA-II and PAES
as it points out the relevance of the choice of a specific type ofknowledge
We now compare BOCAC and BOCAN algorithms tothe well-known NSGA-II and PAES algorithms Tables 7 and8 show a comparison between our BOCAC algorithm andthe NSGA-II and PAES algorithms As we can see althoughBOCAC obtains in average a 119878 value 68 lower than the
Table 9 Comparison among our BOCAwith normative knowledgeand NSGA-II and PAES
one obtained by the NSGA-II algorithm it is more thanthree times faster Moreover when BOCAC is comparedto PAES algorithm the obtained 119878 values are in averageequivalent while BOCAC is around 30 faster than PAESPAES obtains in average more efficient points than BOCAC
though (932)Finally Tables 9 and 10 show a comparison between
BOCAN and NSGA-II and PAES algorithms BOCAN per-forms quite similar to PAES algorithm with respect to both119878 value and the number of obtained efficient solutionsHowever BOCAN is faster than PAES Similar situationoccurs when BOCAN is compared to NSGA-II algorithmAlthough NSGA-II obtains better values for both 119878 and |119883|BOCAN is much faster than NSGA-II This situation canbe explained by the very fast performance that our BOCAN
algorithm obtains for the set of small instances When welook further at the results we can note that if we onlyconsider both medium and large size instances executiontimes obtained by both algorithms are quite similar to eachotherThis result confirmswhat is outlined in [18] in the senseof the good performance that the BOCA algorithm showsFurthermore our results confirm this good performancewith respect to other well-known EMO algorithms does notdepend on which type of knowledge is considered Howeveras we mentioned before the choice of the knowledge usedon the BOCA algorithm is an important issue and it has animpact on the algorithm performance
4 Conclusions and Future Work
Evolutionary algorithms are a very good alternative to solvecomplex combinatorial optimisation problems In this paper
The Scientific World Journal 9
Table 10 Comparison among our BOCA with normative knowl-edge and NSGA-II and PAES
we have implemented a biobjective cultural algorithm to solvethe well-known BOUFLP We have considered two differentsources of knowledge namely circumstantial and normativeand compare them with a previously implemented historicalknowledge Furthermore we compare our BOCAapproacheswith two well-known EAs namely NSGA-II and PAES
Although BOCA approaches using both normative andcircumstantial knowledge could not improve the resultsobtained by the BOCA algorithm with the historical knowl-edge results pointed out that performance of the BOCAalgorithm depends largely on the selected knowledge andit can make the difference in terms of 119878 value time andnumber of efficient solutions found by the algorithm Thisis an important finding as it points out the relevance of thechoice of a specific type of knowledge Moreover our resultsalso confirm the good performance showed by the BOCAalgorithmwith respect to other well-known EMO algorithmssuch as NSGA-II and PAES algorithmsThe BOCA algorithmis very competitive when compared to those EMOalgorithmsindependently of the type of knowledge implemented
As a future work we think that more investigation isneeded in order to find patterns that allow us to get theright knowledge implemented depending on the problemfeatures As we mentioned before the knowledge choice hasan impact on the performance of the BOCA algorithm andtherefore it must be studied in depth Also as future workhybrid knowledge could be implemented in order to exploitthe advantages of each kind of knowledge at the same timeMoreover our BOCA algorithm can be used to solve other
interesting MOPs arising in the logistic field such as routingor scheduling problems
Appendix
Result Tables
In this appendix section obtained results are presentedColumns 119878
sdotshow the 119878 value obtained by algorithm sdot as
Algorithms are indexed as follows The original BOCAalgorithm is indexed by 1 BOCA algorithms using cir-cumstantial and normative knowledge are indexed by 2
and 3 respectively Finally the other EAs considered inthis paper namely NSGA-II and PAES are indexed by 4
and 5 respectively Columns 119905sdotshow the time obtained by
each algorithm in seconds Columns |119883sdot| show the number
of efficient solutions found by the corresponding algorithmFinally operator Δsdot
sdotsdotshows a value that is equivalent to (sdot minus
sdotsdot)sdot times 100
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] D Maravall and J de Lope ldquoMulti-objective dynamic opti-mization with genetic algorithms for automatic parkingrdquo SoftComputing vol 11 no 3 pp 249ndash257 2007
[2] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multiobjectiveoptimization Nsga IIrdquo in Parallel Problem Solving from NaturePPSN VI M Schoenauer K Deb G Rudolph et al Edsvol 1917 of Lecture Notes in Computer Science pp 849ndash858Springer Berlin Germany 2000
[3] I Borgulya ldquoAn algorithm for the capacitated vehicle routingproblem with route balancingrdquo Central European Journal ofOperations Research vol 16 no 4 pp 331ndash343 2008
[4] P J Angeline Z Michalewicz M Schoenauer X Yao and AZalzala Eds The Pareto Archived Evolution Strategy A NewBaseline Algorithm for Pareto Multiobjective Optimisation vol1 IEEE Press 1999
[5] C A Coello Coello and M Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 2002
[6] W K Mashwani ldquoComprehensive survey of the hybrid evolu-tionary algorithmsrdquo International Journal of Applied Evolution-ary Computation vol 4 pp 1ndash19 2013
[7] R Bhattacharya and S Bandyopadhyay ldquoSolving conflicting bi-objective facility location problem by NSGA II evolutionaryalgorithmrdquo International Journal of Advanced ManufacturingTechnology vol 51 no 1ndash4 pp 397ndash414 2010
[8] B Crawford C Lagos C Castro and F Paredes ldquoA culturalalgorithm for solving the set covering problemrdquo inAnalysis andDesign of Intelligent Systems using Soft Computing TechniquesP Melin O Castillo E Ramırez J Kacprzyk and W PedryczEds vol 41 of Advances in Soft Computing pp 408ndash415Springer Berlin Germany 2007
10 The Scientific World Journal
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014
number of optimisation problems (see [6]) A comprehensiveliterature review related to EMO algorithms can be found in[7]
EMO algorithms have some problems that must be takeninto account though For instance they tend to fall intopremature convergence with low evolution efficiency [8]This is because of implicit information embodied in theevolution process and domain knowledge corresponding tooptimisation problems which are not fully included in thesolution approach [9] To overcome these problems onecan make use of implicit evolution information Reynolds[10] proposes an EA called cultural algorithm (CA) whichis inspired from human culture evolution process and thatmakes use of the implicit evolution information generated ateach iterationTheCAs have a dual evolution structure whichconsists of two spaces population and belief space On theone hand the population space works as in any other EAthat is using evolutionary operators such as mutation andcrossover On the other hand in the belief space implicitknowledge is extracted from selected individuals in thepopulation and stored in a different way Then they are usedto guide the whole evolution process in the population spacesuch that they can induce the population to escape fromlocal optimal solutions It has been proved that CAs caneffectively improve the evolution performance [9] Althoughless studied CAs have been also used to solve MO problemsCoello et al [11] andCoello et al [12] two remarkable surveysonCAs onlymentionCoello and Landa [13] work as exampleof a CA application solving MO problems More recentlyZhang et al [14] present a CA which is enhanced by usinga particle swarm optimisation algorithm This enhanced CAis applied to fuel distribution MO problem Srinivasan andRamakrishnan [15] present a MO cultural algorithm that isapplied on data mining domain In [16] the authors applieda CA to a biobjective portfolio selection problem usingnormative knowledge in the belief space In [17] authorspresent a formal framework to implement MO culturalalgorithms To the best of our knowledge [18] is the onlyarticle that uses CAs to solve the biobjective uncapacitatedfacility location problem (BOUFLP) Furthermore we didnot find any article which compares the performance of CAsusing different evolutionary strategies at the belief space levelThus in this paper we present an extension of the biobjectivecultural algorithm (BOCA) developed in [18] We use twodifferent strategies at the belief space level and compare theperformance of our new algorithms with the performanceof the previous one We also compare its results with otherwell-known EMO algorithms such as PAES and NSGA-IIObtained solutions were compared using the hypervolume Smetric proposed in [19]
The remaining of this paper is organised as followsSection 2 shows an overview on MO focused on EMOalgorithms Section 21 presents briefly the BOUFLP andsome of its distinctive features In Section 3 we describe ourimplementation for the BOCA algorithm We describe themain differences between our implementation and the one in[18] Section 32 presents the BOCA algorithm applied to a setof well-known instances from the literature Finally Section 4presents the conclusions of this work
2 Overview
In this section we show an overview of topics relatedto this paper In Section 21 MO concepts are presentedemphasizing EMO algorithms and its state of art Morespecifically we focus on the development of EMO algorithmsfor MO combinatorial optimisation (MOCO) problems InSection 22 we present the BOUFLP formulation based on acost-coverage approach Finally in Section 23 we present anoverview of CAs and its multiobjective extension Details ofour CA implementation are also presented at the end of thissection
21 (Evolutionary) Multiobjective Optimisation In this sec-tionwe briefly introduce themain principles ofMOproblemsand particularly MOCO problems For a comprehensivereview of this topic see [20 21] In this paper we will makeuse of the following notation for the comparison of vectors(solutions) Let 119910 and 119910
1015840isin R119901 We say that 119910 ≦ 119910
1015840 if119910119896≦ 1199101015840
119896forall119896 = 1 119901 Similarly we will say that 119910 le 119910
1015840 if119910 ≦ 119910
1015840 but 119910 = 1199101015840 Finally we say that 119910 lt 119910
1015840 if 119910119896lt 1199101015840
119896forall119896 =
1 119901 A solution 119909 isin R119899 with 119899 being equal to the numberof decision variables is called an efficient solution and itsimage 119891(119909) with 119891 R119899 rarr R119901 a nondominated point ofthe MO problem if there is no 119909 isin R119899 with 119909 = 119909 such that119891(119909) le 119891(119909) In [22] the author describes several excellencerelations These relations establish strict partial orders in theset of all nondominated points related to different aspectsof their quality Previously in [23 24] the authors considerseveral outperformance relations to address the closenessof the set of nondominated points found by an algorithmto the actual set of nondominated points called ParetoFrontier (PF) In [25] a comprehensive explanation of thedesirable features of an approximation to the PF is presentedIn this paper we choose the S metric which is properlyexplained in [24] The S metric calculates the hypervolumeof a multidimensional region [19] and allows the integrationof aspects that are individually measured by other metricsAn advantage of the S metric is that each algorithm can beassessed independently of the other algorithms involved inthe study However the S values of two setsA andB namelySA and SB respectively cannot be used to derive whethereither set entirely dominates the other Figure 1(a) shows asituation where SA gt SB and setA completely dominates setB Figure 1(b) shows a situation where SB gt SA but neitherA dominatesB norB dominatesA
In this paper we use EAs to solve the BOUFLP An EAis a stochastic search procedure inspired by the evolutionprocess in nature In this process individuals evolve and thefitter ones have a better chance of reproduction and survivalThe reproduction mechanisms favour the characteristics ofthe stronger parents and hopefully produce better childrenguaranteeing the presence of those characteristics in futuregenerations [26] EAs have been successfully applied to alarge number of both single- andmultiobjective optimisationproblems Comprehensive reviews of EMO algorithms arepresented in [11 24] andmore recently in [12] Amore general
Figure 1 S metric represented as the area that is dominated by a setof (approximately) nondominated points
review of hybrid heuristics solving MOCO problems whereEMO algorithms are also included is presented in [27]
22 Biobjective Uncapacitated Facility Location ProblemFacility location problem (FLP) is one of the most importantproblems for companies with the aim of distributing productsto their customers The problem consists of selecting sites toinstall plants warehouses and distribution centres allocatingcustomers to serving facilities and interconnecting facilitiesby flow assignment decisions Comprehensive reviews andanalysis of the FLP are presented in [28ndash30]
In this paper we consider a two-level supply chain wherea single plant serves a set of warehouses which serve a set ofend customers or retailers Figure 2 shows this configuration
Two main (conflicting) objectives can be identified in theFLP
(i) minimise the total cost associated with the facilityinstallation and customer allocation and
(ii) maximise the customers rate coverage
Several works on single-objective optimisation have beencarried out considering these two objectives separately Onthe one hand uncapacitated FLP (UFLP) is one of moststudied FLPs in the literature In the UFLP the main goalis to minimise the location-allocation cost of the networkOn the other hand 119901 median FLPs are one of the mostcommon FLPs among those that are focused on coveragemaximisationMost important FLPmodels arewell describedand formalised in [29] MO FLPs have been also well studiedin the literature during the last two decades A survey on thistopic can be found in [31]
As we mentioned before in this paper we solve theBOUFLP BOUFLP has been modelled with minisum andmaxisumobjectives (cost and coverage)The followingmodelformulation is based on [32] Let 119868 = 1 119898 be the set ofpotential facilities and 119869 = 1 119899 the set of customersLet FC
119894be the fixed cost of opening facility 119894 and 119889
119895the
demand of customer 119895 Let 119888119894119895be the cost of assigning the
customer 119895 to facility 119894 and ℎ119894119895the distance between facility 119894
and customer 119895 Let119863MAX be the maximal covering distancethat is customers within this distance to an open facility areconsidered well served Let 119876
119895= 119894 isin 119868 ℎ
119894119895le 119863MAX
be the set of facilities that could serve customer 119895 within themaximal covering distance 119863MAX Let 119909119894 be 1 if facility 119894 isopen and 0 otherwise Let 119910
119894119895be 1 if the whole demand of
customer 119895 is served by facility 119894 and 0 otherwise Consider
with 119891 R119899 rarr R119901 and 119901 = 2 Objective functions 1198911and 119891
2
are as follows
1198911(119909 119910) = sum
119894isin119868
FC119894119909119894+ sum
119894isin119868
sum
119895isin119869
119888119894119895119910119894119895 (6)
1198912(119909 119910) = minussum
119895isin119869
119889119895sum
119894isin119876119895
119910119894119895 (7)
Equation (6) represents total operating cost the first termcorresponds to location cost that is the sum of the fixed costsof all the open facilities and the second term represents theallocation cost that is the cost of attending customer demandby an open facility Equation (7) measures coverage as thesum of the demand of customers attended by open facilitieswithin the maximal covering distance Equations (2) and (3)
4 The Scientific World Journal
Plant or centralwarehouse
Warehouse 1
Warehouse 2
Warehouse 3
Customers(clusters)
Figure 2 A two-level supply chain network configuration
ensure that each customer is attended by only one facilityEquation (3) also forces customer to be assigned to an openfacility Finally equations (4) and (5) set decision variables asbinary
23 Biobjective Cultural Algorithm The experience andbeliefs accepted by a community in a social system are themain motivations for the creation of the CAs Originally pro-posed by Reynolds [10] CAs model the evolution of culturalsystems based on the principles of human social evolutionIn this case evolution is seen as an optimisation process [10]The CAs guide the evolution of the population based on theknowledge Knowledge acquired during previous iterationsis provided to future generations allowing accelerating theconvergence of the algorithm to good solutions [33] Domainknowledge is modelled separately from the populationbecause there is certain independence between them whichallows us to work and model them separately in order toenhance the overall algorithm performance Figure 3 showsthis interaction
CAs are mainly characterised by presenting two inher-itance systems one at population level called populationspace and the other at knowledge level called belief spaceThis key feature is designed to increase the learning ratesand convergence of the algorithm and thus to do a moreresponsive system for a number of problems [34] Moreoverit allows us to identify two significant levels of knowledge
a microevolutionary level (represented by the populationspace) andmacroevolutionary level (represented by the spaceof beliefs) [35]
CAs have the following components population space(set of individuals who have independent features) [35] beliefspace (stored individuals acquired in previous generations)[34] computer protocol connecting the two spaces anddefining the rules on the type of information to be exchangedbetween them by using the acceptance and influence func-tions and finally knowledge sources which are describedin terms of their ability to coordinate the distribution ofindividuals depending on the nature of a problem instance[35] These knowledge sources can be of the followingtypes circumstantial normative domain topographic andhistorical
The most distinctive feature of CAs is the use of thebelief space which through an influence function affectsfuture generations For this reason in this paper we focuson the effect on the algorithm performance of changes insuch an influence function To do this we have consideredresults obtained previously in [18] where the authors usedan influence function based on historical knowledge andwe compare those results with our BOCA implementationwhich considers two influence functions the first one basedon circumstantial knowledge and the second one basedon normative knowledge Algorithm 1 shows the generalprocedure of our BOCA algorithm
The Scientific World Journal 5
Belief space
Population space
Update ()
Accept ()Influence ()
New generation ()
Best individuals ()
2
2
2
1
1
1
n minus 1 n
n minus 1 n
n minus 1 n
middot middot middot
middot middot middot
middot middot middot
middot middot middot
Figure 3 CA general diagram
begin119896 = 0initialise Popuation 119875
119896
initialise BeliefSpace 119861119896
while 119896 lt 119894119905119890119903119886119905119894119900119899119873119906119898119887119890119903 doEvaluate(119875
To initialise the population we use a semirandom func-tion In its first phase this function defines in a stochasticway the set of facilities that will be opened (selected facili-ties) Then we allocate each customer to a selected facilityminimising the cost function 119891
1while avoiding minimising
the coverage function 1198912 This strategy provides better results
than using completely random initial populations and itscomputational time additional cost is marginal
To obtain the next generation two parents are used in arecombination process To avoid local optimal values we donot overuse the culture Thus a parent is selected from thepopulation to obtain diversity and the other parent is selected
from the belief space to influence the next generation Thebelief space keeps a list of all the individuals which meetsome criteria These criteria depend on what knowledgethe algorithm implements In this paper the circumstantialknowledge selects the best individuals found so far foreach objective function Thus one individual will give usinformation on the best value found for 119891
1and the other will
do the same for 1198912 The historical knowledge stores a list of
individuals with the best fitness value found so farThe fitnessvalue is calculated as the hypervolume S that is covered by anindividual Finally normative knowledge considers a list ofindividuals which are pairwise nondominated with respect tothe other individuals of their generation
Let |119869| be the number of available facilities and let |119868|
be the number of customers of our BOUFLP In this paperdecisions variables 119909 and 119910 are represented by a binary |119869|-length vector and |119869| times |119868| matrix respectively Comparingtwo different solutions (individuals) needs an evaluationcriterion In this paper we use the same criterion explainedin [18]
3 Computational Experiments
In this section we present the set of instances that are usedin this study as well as results obtained by our BOCAimplementation
31 Instances Presentation The instances that were used inthis paper correspond to random instances using a problemgenerator that follows the methodology from UflLib [36]Previous works in the literature have also used this problemgenerator to create their test instances [18 26]
The BOCA algorithm has several parameters that need tobe set As in [18] the number of generations considered inthis paper is equal to 100 Population size 119871 is set equal to 100mutation probability in the population space 119875ps is equal to02 and probability of mutation in the belief space 119875bs is 004Both 119871 and 119875ps values are different from the values used in[18] These values are chosen as they all together yield to thebest performance of the algorithm given some test instancesThus although resulting values are different from that used in[18] the method we use to set them is the same as that usedin that work This is important in order to fairly compare thedifferent BOCA algorithms
32 Results and Discussion In this section we compare theresults obtained by the previous BOCAalgorithm (see [18] forfurther details) and our approach Moreover a comparisonbetween results obtained by well-known EMO algorithmssuch as NSGA-II and PAES and our BOCA algorithm is alsopresented in this section
Tables 1 and 2 show the results obtained by the BOCAimplementations using historical [18] circumstantial andnormative knowledge respectively In the same way Tables 3and 4 present the results obtained by the well-known NSGA-II and PAES algorithms For each algorithm 119878 value () time119905 (in seconds) and the number of efficient solutions 119909 isin 119883
have been included in these tables As we mentioned before
6 The Scientific World Journal
Table 1 Results obtained by the BOCA implementations for instance of class 119860
Instance BOCA (historical) BOCA (circumstantial) BOCA (normative)1199051(sec) S
we want to produce a set with a large number of efficientsolutions 119909 isin 119883 a 119878 value close to 100 (ideal) and a small 119905For the sake of easy reading we have split the set of instancesinto two subsets (instances type 119860 and 119861)
We then compare our BOCA implementations with theone presented in [18] Tables 5 and 6 show a comparisonbetween those algorithms As we can see when compared interms of its 119878 value (the bigger the better) BOCA algorithm
using historical knowledge (BOCAH) performs consistentlybetter than the ones using circumstantial (BOCAC) andnormative (BOCAN) knowledge In fact BOCAH obtains a 119878
value that is in average 58 bigger than the one obtainedby BOCAC and 65 bigger than the 119878 value obtained byBOCAN When compared in terms of the CPU time neededto reach the number of iterations (generations) BOCAH isin average faster than both BOCAC and BOCAN algorithms
The Scientific World Journal 7
Table 3 Results obtained bywell-knownMOEA algorithmsNSGA-II and PAES for instances of class 119860
We can note that for 119861 instances times required by BOCAH
and BOCAC are in average quite similar (only 16 ofdifference) Finally when we look at the number of efficientsolutions found by each algorithm (|119883| column) we can seethat again BOCAH outperforms both BOCAC and BOCAN
algorithms In this case the average number of efficient
Table 5 Comparison among our BOCA implementations (119860instances)
solutions found by the BOCAH algorithm is about 20biggerthan the one obtained by the other two approaches
Results above are consistent with the good performanceobtained by the BOCAH approach in [18] Moreover resultsshow that performance of the BOCA algorithm dependslargely on the selected knowledge and it can make thedifference in terms of 119878 value time and number of efficientsolutions found by the algorithmThis is an important finding
8 The Scientific World Journal
Table 7 Comparison among our BOCAwith circumstantial knowl-edge and NSGA-II and PAES
as it points out the relevance of the choice of a specific type ofknowledge
We now compare BOCAC and BOCAN algorithms tothe well-known NSGA-II and PAES algorithms Tables 7 and8 show a comparison between our BOCAC algorithm andthe NSGA-II and PAES algorithms As we can see althoughBOCAC obtains in average a 119878 value 68 lower than the
Table 9 Comparison among our BOCAwith normative knowledgeand NSGA-II and PAES
one obtained by the NSGA-II algorithm it is more thanthree times faster Moreover when BOCAC is comparedto PAES algorithm the obtained 119878 values are in averageequivalent while BOCAC is around 30 faster than PAESPAES obtains in average more efficient points than BOCAC
though (932)Finally Tables 9 and 10 show a comparison between
BOCAN and NSGA-II and PAES algorithms BOCAN per-forms quite similar to PAES algorithm with respect to both119878 value and the number of obtained efficient solutionsHowever BOCAN is faster than PAES Similar situationoccurs when BOCAN is compared to NSGA-II algorithmAlthough NSGA-II obtains better values for both 119878 and |119883|BOCAN is much faster than NSGA-II This situation canbe explained by the very fast performance that our BOCAN
algorithm obtains for the set of small instances When welook further at the results we can note that if we onlyconsider both medium and large size instances executiontimes obtained by both algorithms are quite similar to eachotherThis result confirmswhat is outlined in [18] in the senseof the good performance that the BOCA algorithm showsFurthermore our results confirm this good performancewith respect to other well-known EMO algorithms does notdepend on which type of knowledge is considered Howeveras we mentioned before the choice of the knowledge usedon the BOCA algorithm is an important issue and it has animpact on the algorithm performance
4 Conclusions and Future Work
Evolutionary algorithms are a very good alternative to solvecomplex combinatorial optimisation problems In this paper
The Scientific World Journal 9
Table 10 Comparison among our BOCA with normative knowl-edge and NSGA-II and PAES
we have implemented a biobjective cultural algorithm to solvethe well-known BOUFLP We have considered two differentsources of knowledge namely circumstantial and normativeand compare them with a previously implemented historicalknowledge Furthermore we compare our BOCAapproacheswith two well-known EAs namely NSGA-II and PAES
Although BOCA approaches using both normative andcircumstantial knowledge could not improve the resultsobtained by the BOCA algorithm with the historical knowl-edge results pointed out that performance of the BOCAalgorithm depends largely on the selected knowledge andit can make the difference in terms of 119878 value time andnumber of efficient solutions found by the algorithm Thisis an important finding as it points out the relevance of thechoice of a specific type of knowledge Moreover our resultsalso confirm the good performance showed by the BOCAalgorithmwith respect to other well-known EMO algorithmssuch as NSGA-II and PAES algorithmsThe BOCA algorithmis very competitive when compared to those EMOalgorithmsindependently of the type of knowledge implemented
As a future work we think that more investigation isneeded in order to find patterns that allow us to get theright knowledge implemented depending on the problemfeatures As we mentioned before the knowledge choice hasan impact on the performance of the BOCA algorithm andtherefore it must be studied in depth Also as future workhybrid knowledge could be implemented in order to exploitthe advantages of each kind of knowledge at the same timeMoreover our BOCA algorithm can be used to solve other
interesting MOPs arising in the logistic field such as routingor scheduling problems
Appendix
Result Tables
In this appendix section obtained results are presentedColumns 119878
sdotshow the 119878 value obtained by algorithm sdot as
Algorithms are indexed as follows The original BOCAalgorithm is indexed by 1 BOCA algorithms using cir-cumstantial and normative knowledge are indexed by 2
and 3 respectively Finally the other EAs considered inthis paper namely NSGA-II and PAES are indexed by 4
and 5 respectively Columns 119905sdotshow the time obtained by
each algorithm in seconds Columns |119883sdot| show the number
of efficient solutions found by the corresponding algorithmFinally operator Δsdot
sdotsdotshows a value that is equivalent to (sdot minus
sdotsdot)sdot times 100
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] D Maravall and J de Lope ldquoMulti-objective dynamic opti-mization with genetic algorithms for automatic parkingrdquo SoftComputing vol 11 no 3 pp 249ndash257 2007
[2] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multiobjectiveoptimization Nsga IIrdquo in Parallel Problem Solving from NaturePPSN VI M Schoenauer K Deb G Rudolph et al Edsvol 1917 of Lecture Notes in Computer Science pp 849ndash858Springer Berlin Germany 2000
[3] I Borgulya ldquoAn algorithm for the capacitated vehicle routingproblem with route balancingrdquo Central European Journal ofOperations Research vol 16 no 4 pp 331ndash343 2008
[4] P J Angeline Z Michalewicz M Schoenauer X Yao and AZalzala Eds The Pareto Archived Evolution Strategy A NewBaseline Algorithm for Pareto Multiobjective Optimisation vol1 IEEE Press 1999
[5] C A Coello Coello and M Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 2002
[6] W K Mashwani ldquoComprehensive survey of the hybrid evolu-tionary algorithmsrdquo International Journal of Applied Evolution-ary Computation vol 4 pp 1ndash19 2013
[7] R Bhattacharya and S Bandyopadhyay ldquoSolving conflicting bi-objective facility location problem by NSGA II evolutionaryalgorithmrdquo International Journal of Advanced ManufacturingTechnology vol 51 no 1ndash4 pp 397ndash414 2010
[8] B Crawford C Lagos C Castro and F Paredes ldquoA culturalalgorithm for solving the set covering problemrdquo inAnalysis andDesign of Intelligent Systems using Soft Computing TechniquesP Melin O Castillo E Ramırez J Kacprzyk and W PedryczEds vol 41 of Advances in Soft Computing pp 408ndash415Springer Berlin Germany 2007
10 The Scientific World Journal
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014
Figure 1 S metric represented as the area that is dominated by a setof (approximately) nondominated points
review of hybrid heuristics solving MOCO problems whereEMO algorithms are also included is presented in [27]
22 Biobjective Uncapacitated Facility Location ProblemFacility location problem (FLP) is one of the most importantproblems for companies with the aim of distributing productsto their customers The problem consists of selecting sites toinstall plants warehouses and distribution centres allocatingcustomers to serving facilities and interconnecting facilitiesby flow assignment decisions Comprehensive reviews andanalysis of the FLP are presented in [28ndash30]
In this paper we consider a two-level supply chain wherea single plant serves a set of warehouses which serve a set ofend customers or retailers Figure 2 shows this configuration
Two main (conflicting) objectives can be identified in theFLP
(i) minimise the total cost associated with the facilityinstallation and customer allocation and
(ii) maximise the customers rate coverage
Several works on single-objective optimisation have beencarried out considering these two objectives separately Onthe one hand uncapacitated FLP (UFLP) is one of moststudied FLPs in the literature In the UFLP the main goalis to minimise the location-allocation cost of the networkOn the other hand 119901 median FLPs are one of the mostcommon FLPs among those that are focused on coveragemaximisationMost important FLPmodels arewell describedand formalised in [29] MO FLPs have been also well studiedin the literature during the last two decades A survey on thistopic can be found in [31]
As we mentioned before in this paper we solve theBOUFLP BOUFLP has been modelled with minisum andmaxisumobjectives (cost and coverage)The followingmodelformulation is based on [32] Let 119868 = 1 119898 be the set ofpotential facilities and 119869 = 1 119899 the set of customersLet FC
119894be the fixed cost of opening facility 119894 and 119889
119895the
demand of customer 119895 Let 119888119894119895be the cost of assigning the
customer 119895 to facility 119894 and ℎ119894119895the distance between facility 119894
and customer 119895 Let119863MAX be the maximal covering distancethat is customers within this distance to an open facility areconsidered well served Let 119876
119895= 119894 isin 119868 ℎ
119894119895le 119863MAX
be the set of facilities that could serve customer 119895 within themaximal covering distance 119863MAX Let 119909119894 be 1 if facility 119894 isopen and 0 otherwise Let 119910
119894119895be 1 if the whole demand of
customer 119895 is served by facility 119894 and 0 otherwise Consider
with 119891 R119899 rarr R119901 and 119901 = 2 Objective functions 1198911and 119891
2
are as follows
1198911(119909 119910) = sum
119894isin119868
FC119894119909119894+ sum
119894isin119868
sum
119895isin119869
119888119894119895119910119894119895 (6)
1198912(119909 119910) = minussum
119895isin119869
119889119895sum
119894isin119876119895
119910119894119895 (7)
Equation (6) represents total operating cost the first termcorresponds to location cost that is the sum of the fixed costsof all the open facilities and the second term represents theallocation cost that is the cost of attending customer demandby an open facility Equation (7) measures coverage as thesum of the demand of customers attended by open facilitieswithin the maximal covering distance Equations (2) and (3)
4 The Scientific World Journal
Plant or centralwarehouse
Warehouse 1
Warehouse 2
Warehouse 3
Customers(clusters)
Figure 2 A two-level supply chain network configuration
ensure that each customer is attended by only one facilityEquation (3) also forces customer to be assigned to an openfacility Finally equations (4) and (5) set decision variables asbinary
23 Biobjective Cultural Algorithm The experience andbeliefs accepted by a community in a social system are themain motivations for the creation of the CAs Originally pro-posed by Reynolds [10] CAs model the evolution of culturalsystems based on the principles of human social evolutionIn this case evolution is seen as an optimisation process [10]The CAs guide the evolution of the population based on theknowledge Knowledge acquired during previous iterationsis provided to future generations allowing accelerating theconvergence of the algorithm to good solutions [33] Domainknowledge is modelled separately from the populationbecause there is certain independence between them whichallows us to work and model them separately in order toenhance the overall algorithm performance Figure 3 showsthis interaction
CAs are mainly characterised by presenting two inher-itance systems one at population level called populationspace and the other at knowledge level called belief spaceThis key feature is designed to increase the learning ratesand convergence of the algorithm and thus to do a moreresponsive system for a number of problems [34] Moreoverit allows us to identify two significant levels of knowledge
a microevolutionary level (represented by the populationspace) andmacroevolutionary level (represented by the spaceof beliefs) [35]
CAs have the following components population space(set of individuals who have independent features) [35] beliefspace (stored individuals acquired in previous generations)[34] computer protocol connecting the two spaces anddefining the rules on the type of information to be exchangedbetween them by using the acceptance and influence func-tions and finally knowledge sources which are describedin terms of their ability to coordinate the distribution ofindividuals depending on the nature of a problem instance[35] These knowledge sources can be of the followingtypes circumstantial normative domain topographic andhistorical
The most distinctive feature of CAs is the use of thebelief space which through an influence function affectsfuture generations For this reason in this paper we focuson the effect on the algorithm performance of changes insuch an influence function To do this we have consideredresults obtained previously in [18] where the authors usedan influence function based on historical knowledge andwe compare those results with our BOCA implementationwhich considers two influence functions the first one basedon circumstantial knowledge and the second one basedon normative knowledge Algorithm 1 shows the generalprocedure of our BOCA algorithm
The Scientific World Journal 5
Belief space
Population space
Update ()
Accept ()Influence ()
New generation ()
Best individuals ()
2
2
2
1
1
1
n minus 1 n
n minus 1 n
n minus 1 n
middot middot middot
middot middot middot
middot middot middot
middot middot middot
Figure 3 CA general diagram
begin119896 = 0initialise Popuation 119875
119896
initialise BeliefSpace 119861119896
while 119896 lt 119894119905119890119903119886119905119894119900119899119873119906119898119887119890119903 doEvaluate(119875
To initialise the population we use a semirandom func-tion In its first phase this function defines in a stochasticway the set of facilities that will be opened (selected facili-ties) Then we allocate each customer to a selected facilityminimising the cost function 119891
1while avoiding minimising
the coverage function 1198912 This strategy provides better results
than using completely random initial populations and itscomputational time additional cost is marginal
To obtain the next generation two parents are used in arecombination process To avoid local optimal values we donot overuse the culture Thus a parent is selected from thepopulation to obtain diversity and the other parent is selected
from the belief space to influence the next generation Thebelief space keeps a list of all the individuals which meetsome criteria These criteria depend on what knowledgethe algorithm implements In this paper the circumstantialknowledge selects the best individuals found so far foreach objective function Thus one individual will give usinformation on the best value found for 119891
1and the other will
do the same for 1198912 The historical knowledge stores a list of
individuals with the best fitness value found so farThe fitnessvalue is calculated as the hypervolume S that is covered by anindividual Finally normative knowledge considers a list ofindividuals which are pairwise nondominated with respect tothe other individuals of their generation
Let |119869| be the number of available facilities and let |119868|
be the number of customers of our BOUFLP In this paperdecisions variables 119909 and 119910 are represented by a binary |119869|-length vector and |119869| times |119868| matrix respectively Comparingtwo different solutions (individuals) needs an evaluationcriterion In this paper we use the same criterion explainedin [18]
3 Computational Experiments
In this section we present the set of instances that are usedin this study as well as results obtained by our BOCAimplementation
31 Instances Presentation The instances that were used inthis paper correspond to random instances using a problemgenerator that follows the methodology from UflLib [36]Previous works in the literature have also used this problemgenerator to create their test instances [18 26]
The BOCA algorithm has several parameters that need tobe set As in [18] the number of generations considered inthis paper is equal to 100 Population size 119871 is set equal to 100mutation probability in the population space 119875ps is equal to02 and probability of mutation in the belief space 119875bs is 004Both 119871 and 119875ps values are different from the values used in[18] These values are chosen as they all together yield to thebest performance of the algorithm given some test instancesThus although resulting values are different from that used in[18] the method we use to set them is the same as that usedin that work This is important in order to fairly compare thedifferent BOCA algorithms
32 Results and Discussion In this section we compare theresults obtained by the previous BOCAalgorithm (see [18] forfurther details) and our approach Moreover a comparisonbetween results obtained by well-known EMO algorithmssuch as NSGA-II and PAES and our BOCA algorithm is alsopresented in this section
Tables 1 and 2 show the results obtained by the BOCAimplementations using historical [18] circumstantial andnormative knowledge respectively In the same way Tables 3and 4 present the results obtained by the well-known NSGA-II and PAES algorithms For each algorithm 119878 value () time119905 (in seconds) and the number of efficient solutions 119909 isin 119883
have been included in these tables As we mentioned before
6 The Scientific World Journal
Table 1 Results obtained by the BOCA implementations for instance of class 119860
Instance BOCA (historical) BOCA (circumstantial) BOCA (normative)1199051(sec) S
we want to produce a set with a large number of efficientsolutions 119909 isin 119883 a 119878 value close to 100 (ideal) and a small 119905For the sake of easy reading we have split the set of instancesinto two subsets (instances type 119860 and 119861)
We then compare our BOCA implementations with theone presented in [18] Tables 5 and 6 show a comparisonbetween those algorithms As we can see when compared interms of its 119878 value (the bigger the better) BOCA algorithm
using historical knowledge (BOCAH) performs consistentlybetter than the ones using circumstantial (BOCAC) andnormative (BOCAN) knowledge In fact BOCAH obtains a 119878
value that is in average 58 bigger than the one obtainedby BOCAC and 65 bigger than the 119878 value obtained byBOCAN When compared in terms of the CPU time neededto reach the number of iterations (generations) BOCAH isin average faster than both BOCAC and BOCAN algorithms
The Scientific World Journal 7
Table 3 Results obtained bywell-knownMOEA algorithmsNSGA-II and PAES for instances of class 119860
We can note that for 119861 instances times required by BOCAH
and BOCAC are in average quite similar (only 16 ofdifference) Finally when we look at the number of efficientsolutions found by each algorithm (|119883| column) we can seethat again BOCAH outperforms both BOCAC and BOCAN
algorithms In this case the average number of efficient
Table 5 Comparison among our BOCA implementations (119860instances)
solutions found by the BOCAH algorithm is about 20biggerthan the one obtained by the other two approaches
Results above are consistent with the good performanceobtained by the BOCAH approach in [18] Moreover resultsshow that performance of the BOCA algorithm dependslargely on the selected knowledge and it can make thedifference in terms of 119878 value time and number of efficientsolutions found by the algorithmThis is an important finding
8 The Scientific World Journal
Table 7 Comparison among our BOCAwith circumstantial knowl-edge and NSGA-II and PAES
as it points out the relevance of the choice of a specific type ofknowledge
We now compare BOCAC and BOCAN algorithms tothe well-known NSGA-II and PAES algorithms Tables 7 and8 show a comparison between our BOCAC algorithm andthe NSGA-II and PAES algorithms As we can see althoughBOCAC obtains in average a 119878 value 68 lower than the
Table 9 Comparison among our BOCAwith normative knowledgeand NSGA-II and PAES
one obtained by the NSGA-II algorithm it is more thanthree times faster Moreover when BOCAC is comparedto PAES algorithm the obtained 119878 values are in averageequivalent while BOCAC is around 30 faster than PAESPAES obtains in average more efficient points than BOCAC
though (932)Finally Tables 9 and 10 show a comparison between
BOCAN and NSGA-II and PAES algorithms BOCAN per-forms quite similar to PAES algorithm with respect to both119878 value and the number of obtained efficient solutionsHowever BOCAN is faster than PAES Similar situationoccurs when BOCAN is compared to NSGA-II algorithmAlthough NSGA-II obtains better values for both 119878 and |119883|BOCAN is much faster than NSGA-II This situation canbe explained by the very fast performance that our BOCAN
algorithm obtains for the set of small instances When welook further at the results we can note that if we onlyconsider both medium and large size instances executiontimes obtained by both algorithms are quite similar to eachotherThis result confirmswhat is outlined in [18] in the senseof the good performance that the BOCA algorithm showsFurthermore our results confirm this good performancewith respect to other well-known EMO algorithms does notdepend on which type of knowledge is considered Howeveras we mentioned before the choice of the knowledge usedon the BOCA algorithm is an important issue and it has animpact on the algorithm performance
4 Conclusions and Future Work
Evolutionary algorithms are a very good alternative to solvecomplex combinatorial optimisation problems In this paper
The Scientific World Journal 9
Table 10 Comparison among our BOCA with normative knowl-edge and NSGA-II and PAES
we have implemented a biobjective cultural algorithm to solvethe well-known BOUFLP We have considered two differentsources of knowledge namely circumstantial and normativeand compare them with a previously implemented historicalknowledge Furthermore we compare our BOCAapproacheswith two well-known EAs namely NSGA-II and PAES
Although BOCA approaches using both normative andcircumstantial knowledge could not improve the resultsobtained by the BOCA algorithm with the historical knowl-edge results pointed out that performance of the BOCAalgorithm depends largely on the selected knowledge andit can make the difference in terms of 119878 value time andnumber of efficient solutions found by the algorithm Thisis an important finding as it points out the relevance of thechoice of a specific type of knowledge Moreover our resultsalso confirm the good performance showed by the BOCAalgorithmwith respect to other well-known EMO algorithmssuch as NSGA-II and PAES algorithmsThe BOCA algorithmis very competitive when compared to those EMOalgorithmsindependently of the type of knowledge implemented
As a future work we think that more investigation isneeded in order to find patterns that allow us to get theright knowledge implemented depending on the problemfeatures As we mentioned before the knowledge choice hasan impact on the performance of the BOCA algorithm andtherefore it must be studied in depth Also as future workhybrid knowledge could be implemented in order to exploitthe advantages of each kind of knowledge at the same timeMoreover our BOCA algorithm can be used to solve other
interesting MOPs arising in the logistic field such as routingor scheduling problems
Appendix
Result Tables
In this appendix section obtained results are presentedColumns 119878
sdotshow the 119878 value obtained by algorithm sdot as
Algorithms are indexed as follows The original BOCAalgorithm is indexed by 1 BOCA algorithms using cir-cumstantial and normative knowledge are indexed by 2
and 3 respectively Finally the other EAs considered inthis paper namely NSGA-II and PAES are indexed by 4
and 5 respectively Columns 119905sdotshow the time obtained by
each algorithm in seconds Columns |119883sdot| show the number
of efficient solutions found by the corresponding algorithmFinally operator Δsdot
sdotsdotshows a value that is equivalent to (sdot minus
sdotsdot)sdot times 100
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] D Maravall and J de Lope ldquoMulti-objective dynamic opti-mization with genetic algorithms for automatic parkingrdquo SoftComputing vol 11 no 3 pp 249ndash257 2007
[2] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multiobjectiveoptimization Nsga IIrdquo in Parallel Problem Solving from NaturePPSN VI M Schoenauer K Deb G Rudolph et al Edsvol 1917 of Lecture Notes in Computer Science pp 849ndash858Springer Berlin Germany 2000
[3] I Borgulya ldquoAn algorithm for the capacitated vehicle routingproblem with route balancingrdquo Central European Journal ofOperations Research vol 16 no 4 pp 331ndash343 2008
[4] P J Angeline Z Michalewicz M Schoenauer X Yao and AZalzala Eds The Pareto Archived Evolution Strategy A NewBaseline Algorithm for Pareto Multiobjective Optimisation vol1 IEEE Press 1999
[5] C A Coello Coello and M Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 2002
[6] W K Mashwani ldquoComprehensive survey of the hybrid evolu-tionary algorithmsrdquo International Journal of Applied Evolution-ary Computation vol 4 pp 1ndash19 2013
[7] R Bhattacharya and S Bandyopadhyay ldquoSolving conflicting bi-objective facility location problem by NSGA II evolutionaryalgorithmrdquo International Journal of Advanced ManufacturingTechnology vol 51 no 1ndash4 pp 397ndash414 2010
[8] B Crawford C Lagos C Castro and F Paredes ldquoA culturalalgorithm for solving the set covering problemrdquo inAnalysis andDesign of Intelligent Systems using Soft Computing TechniquesP Melin O Castillo E Ramırez J Kacprzyk and W PedryczEds vol 41 of Advances in Soft Computing pp 408ndash415Springer Berlin Germany 2007
10 The Scientific World Journal
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014
Figure 2 A two-level supply chain network configuration
ensure that each customer is attended by only one facilityEquation (3) also forces customer to be assigned to an openfacility Finally equations (4) and (5) set decision variables asbinary
23 Biobjective Cultural Algorithm The experience andbeliefs accepted by a community in a social system are themain motivations for the creation of the CAs Originally pro-posed by Reynolds [10] CAs model the evolution of culturalsystems based on the principles of human social evolutionIn this case evolution is seen as an optimisation process [10]The CAs guide the evolution of the population based on theknowledge Knowledge acquired during previous iterationsis provided to future generations allowing accelerating theconvergence of the algorithm to good solutions [33] Domainknowledge is modelled separately from the populationbecause there is certain independence between them whichallows us to work and model them separately in order toenhance the overall algorithm performance Figure 3 showsthis interaction
CAs are mainly characterised by presenting two inher-itance systems one at population level called populationspace and the other at knowledge level called belief spaceThis key feature is designed to increase the learning ratesand convergence of the algorithm and thus to do a moreresponsive system for a number of problems [34] Moreoverit allows us to identify two significant levels of knowledge
a microevolutionary level (represented by the populationspace) andmacroevolutionary level (represented by the spaceof beliefs) [35]
CAs have the following components population space(set of individuals who have independent features) [35] beliefspace (stored individuals acquired in previous generations)[34] computer protocol connecting the two spaces anddefining the rules on the type of information to be exchangedbetween them by using the acceptance and influence func-tions and finally knowledge sources which are describedin terms of their ability to coordinate the distribution ofindividuals depending on the nature of a problem instance[35] These knowledge sources can be of the followingtypes circumstantial normative domain topographic andhistorical
The most distinctive feature of CAs is the use of thebelief space which through an influence function affectsfuture generations For this reason in this paper we focuson the effect on the algorithm performance of changes insuch an influence function To do this we have consideredresults obtained previously in [18] where the authors usedan influence function based on historical knowledge andwe compare those results with our BOCA implementationwhich considers two influence functions the first one basedon circumstantial knowledge and the second one basedon normative knowledge Algorithm 1 shows the generalprocedure of our BOCA algorithm
The Scientific World Journal 5
Belief space
Population space
Update ()
Accept ()Influence ()
New generation ()
Best individuals ()
2
2
2
1
1
1
n minus 1 n
n minus 1 n
n minus 1 n
middot middot middot
middot middot middot
middot middot middot
middot middot middot
Figure 3 CA general diagram
begin119896 = 0initialise Popuation 119875
119896
initialise BeliefSpace 119861119896
while 119896 lt 119894119905119890119903119886119905119894119900119899119873119906119898119887119890119903 doEvaluate(119875
To initialise the population we use a semirandom func-tion In its first phase this function defines in a stochasticway the set of facilities that will be opened (selected facili-ties) Then we allocate each customer to a selected facilityminimising the cost function 119891
1while avoiding minimising
the coverage function 1198912 This strategy provides better results
than using completely random initial populations and itscomputational time additional cost is marginal
To obtain the next generation two parents are used in arecombination process To avoid local optimal values we donot overuse the culture Thus a parent is selected from thepopulation to obtain diversity and the other parent is selected
from the belief space to influence the next generation Thebelief space keeps a list of all the individuals which meetsome criteria These criteria depend on what knowledgethe algorithm implements In this paper the circumstantialknowledge selects the best individuals found so far foreach objective function Thus one individual will give usinformation on the best value found for 119891
1and the other will
do the same for 1198912 The historical knowledge stores a list of
individuals with the best fitness value found so farThe fitnessvalue is calculated as the hypervolume S that is covered by anindividual Finally normative knowledge considers a list ofindividuals which are pairwise nondominated with respect tothe other individuals of their generation
Let |119869| be the number of available facilities and let |119868|
be the number of customers of our BOUFLP In this paperdecisions variables 119909 and 119910 are represented by a binary |119869|-length vector and |119869| times |119868| matrix respectively Comparingtwo different solutions (individuals) needs an evaluationcriterion In this paper we use the same criterion explainedin [18]
3 Computational Experiments
In this section we present the set of instances that are usedin this study as well as results obtained by our BOCAimplementation
31 Instances Presentation The instances that were used inthis paper correspond to random instances using a problemgenerator that follows the methodology from UflLib [36]Previous works in the literature have also used this problemgenerator to create their test instances [18 26]
The BOCA algorithm has several parameters that need tobe set As in [18] the number of generations considered inthis paper is equal to 100 Population size 119871 is set equal to 100mutation probability in the population space 119875ps is equal to02 and probability of mutation in the belief space 119875bs is 004Both 119871 and 119875ps values are different from the values used in[18] These values are chosen as they all together yield to thebest performance of the algorithm given some test instancesThus although resulting values are different from that used in[18] the method we use to set them is the same as that usedin that work This is important in order to fairly compare thedifferent BOCA algorithms
32 Results and Discussion In this section we compare theresults obtained by the previous BOCAalgorithm (see [18] forfurther details) and our approach Moreover a comparisonbetween results obtained by well-known EMO algorithmssuch as NSGA-II and PAES and our BOCA algorithm is alsopresented in this section
Tables 1 and 2 show the results obtained by the BOCAimplementations using historical [18] circumstantial andnormative knowledge respectively In the same way Tables 3and 4 present the results obtained by the well-known NSGA-II and PAES algorithms For each algorithm 119878 value () time119905 (in seconds) and the number of efficient solutions 119909 isin 119883
have been included in these tables As we mentioned before
6 The Scientific World Journal
Table 1 Results obtained by the BOCA implementations for instance of class 119860
Instance BOCA (historical) BOCA (circumstantial) BOCA (normative)1199051(sec) S
we want to produce a set with a large number of efficientsolutions 119909 isin 119883 a 119878 value close to 100 (ideal) and a small 119905For the sake of easy reading we have split the set of instancesinto two subsets (instances type 119860 and 119861)
We then compare our BOCA implementations with theone presented in [18] Tables 5 and 6 show a comparisonbetween those algorithms As we can see when compared interms of its 119878 value (the bigger the better) BOCA algorithm
using historical knowledge (BOCAH) performs consistentlybetter than the ones using circumstantial (BOCAC) andnormative (BOCAN) knowledge In fact BOCAH obtains a 119878
value that is in average 58 bigger than the one obtainedby BOCAC and 65 bigger than the 119878 value obtained byBOCAN When compared in terms of the CPU time neededto reach the number of iterations (generations) BOCAH isin average faster than both BOCAC and BOCAN algorithms
The Scientific World Journal 7
Table 3 Results obtained bywell-knownMOEA algorithmsNSGA-II and PAES for instances of class 119860
We can note that for 119861 instances times required by BOCAH
and BOCAC are in average quite similar (only 16 ofdifference) Finally when we look at the number of efficientsolutions found by each algorithm (|119883| column) we can seethat again BOCAH outperforms both BOCAC and BOCAN
algorithms In this case the average number of efficient
Table 5 Comparison among our BOCA implementations (119860instances)
solutions found by the BOCAH algorithm is about 20biggerthan the one obtained by the other two approaches
Results above are consistent with the good performanceobtained by the BOCAH approach in [18] Moreover resultsshow that performance of the BOCA algorithm dependslargely on the selected knowledge and it can make thedifference in terms of 119878 value time and number of efficientsolutions found by the algorithmThis is an important finding
8 The Scientific World Journal
Table 7 Comparison among our BOCAwith circumstantial knowl-edge and NSGA-II and PAES
as it points out the relevance of the choice of a specific type ofknowledge
We now compare BOCAC and BOCAN algorithms tothe well-known NSGA-II and PAES algorithms Tables 7 and8 show a comparison between our BOCAC algorithm andthe NSGA-II and PAES algorithms As we can see althoughBOCAC obtains in average a 119878 value 68 lower than the
Table 9 Comparison among our BOCAwith normative knowledgeand NSGA-II and PAES
one obtained by the NSGA-II algorithm it is more thanthree times faster Moreover when BOCAC is comparedto PAES algorithm the obtained 119878 values are in averageequivalent while BOCAC is around 30 faster than PAESPAES obtains in average more efficient points than BOCAC
though (932)Finally Tables 9 and 10 show a comparison between
BOCAN and NSGA-II and PAES algorithms BOCAN per-forms quite similar to PAES algorithm with respect to both119878 value and the number of obtained efficient solutionsHowever BOCAN is faster than PAES Similar situationoccurs when BOCAN is compared to NSGA-II algorithmAlthough NSGA-II obtains better values for both 119878 and |119883|BOCAN is much faster than NSGA-II This situation canbe explained by the very fast performance that our BOCAN
algorithm obtains for the set of small instances When welook further at the results we can note that if we onlyconsider both medium and large size instances executiontimes obtained by both algorithms are quite similar to eachotherThis result confirmswhat is outlined in [18] in the senseof the good performance that the BOCA algorithm showsFurthermore our results confirm this good performancewith respect to other well-known EMO algorithms does notdepend on which type of knowledge is considered Howeveras we mentioned before the choice of the knowledge usedon the BOCA algorithm is an important issue and it has animpact on the algorithm performance
4 Conclusions and Future Work
Evolutionary algorithms are a very good alternative to solvecomplex combinatorial optimisation problems In this paper
The Scientific World Journal 9
Table 10 Comparison among our BOCA with normative knowl-edge and NSGA-II and PAES
we have implemented a biobjective cultural algorithm to solvethe well-known BOUFLP We have considered two differentsources of knowledge namely circumstantial and normativeand compare them with a previously implemented historicalknowledge Furthermore we compare our BOCAapproacheswith two well-known EAs namely NSGA-II and PAES
Although BOCA approaches using both normative andcircumstantial knowledge could not improve the resultsobtained by the BOCA algorithm with the historical knowl-edge results pointed out that performance of the BOCAalgorithm depends largely on the selected knowledge andit can make the difference in terms of 119878 value time andnumber of efficient solutions found by the algorithm Thisis an important finding as it points out the relevance of thechoice of a specific type of knowledge Moreover our resultsalso confirm the good performance showed by the BOCAalgorithmwith respect to other well-known EMO algorithmssuch as NSGA-II and PAES algorithmsThe BOCA algorithmis very competitive when compared to those EMOalgorithmsindependently of the type of knowledge implemented
As a future work we think that more investigation isneeded in order to find patterns that allow us to get theright knowledge implemented depending on the problemfeatures As we mentioned before the knowledge choice hasan impact on the performance of the BOCA algorithm andtherefore it must be studied in depth Also as future workhybrid knowledge could be implemented in order to exploitthe advantages of each kind of knowledge at the same timeMoreover our BOCA algorithm can be used to solve other
interesting MOPs arising in the logistic field such as routingor scheduling problems
Appendix
Result Tables
In this appendix section obtained results are presentedColumns 119878
sdotshow the 119878 value obtained by algorithm sdot as
Algorithms are indexed as follows The original BOCAalgorithm is indexed by 1 BOCA algorithms using cir-cumstantial and normative knowledge are indexed by 2
and 3 respectively Finally the other EAs considered inthis paper namely NSGA-II and PAES are indexed by 4
and 5 respectively Columns 119905sdotshow the time obtained by
each algorithm in seconds Columns |119883sdot| show the number
of efficient solutions found by the corresponding algorithmFinally operator Δsdot
sdotsdotshows a value that is equivalent to (sdot minus
sdotsdot)sdot times 100
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] D Maravall and J de Lope ldquoMulti-objective dynamic opti-mization with genetic algorithms for automatic parkingrdquo SoftComputing vol 11 no 3 pp 249ndash257 2007
[2] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multiobjectiveoptimization Nsga IIrdquo in Parallel Problem Solving from NaturePPSN VI M Schoenauer K Deb G Rudolph et al Edsvol 1917 of Lecture Notes in Computer Science pp 849ndash858Springer Berlin Germany 2000
[3] I Borgulya ldquoAn algorithm for the capacitated vehicle routingproblem with route balancingrdquo Central European Journal ofOperations Research vol 16 no 4 pp 331ndash343 2008
[4] P J Angeline Z Michalewicz M Schoenauer X Yao and AZalzala Eds The Pareto Archived Evolution Strategy A NewBaseline Algorithm for Pareto Multiobjective Optimisation vol1 IEEE Press 1999
[5] C A Coello Coello and M Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 2002
[6] W K Mashwani ldquoComprehensive survey of the hybrid evolu-tionary algorithmsrdquo International Journal of Applied Evolution-ary Computation vol 4 pp 1ndash19 2013
[7] R Bhattacharya and S Bandyopadhyay ldquoSolving conflicting bi-objective facility location problem by NSGA II evolutionaryalgorithmrdquo International Journal of Advanced ManufacturingTechnology vol 51 no 1ndash4 pp 397ndash414 2010
[8] B Crawford C Lagos C Castro and F Paredes ldquoA culturalalgorithm for solving the set covering problemrdquo inAnalysis andDesign of Intelligent Systems using Soft Computing TechniquesP Melin O Castillo E Ramırez J Kacprzyk and W PedryczEds vol 41 of Advances in Soft Computing pp 408ndash415Springer Berlin Germany 2007
10 The Scientific World Journal
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014
To initialise the population we use a semirandom func-tion In its first phase this function defines in a stochasticway the set of facilities that will be opened (selected facili-ties) Then we allocate each customer to a selected facilityminimising the cost function 119891
1while avoiding minimising
the coverage function 1198912 This strategy provides better results
than using completely random initial populations and itscomputational time additional cost is marginal
To obtain the next generation two parents are used in arecombination process To avoid local optimal values we donot overuse the culture Thus a parent is selected from thepopulation to obtain diversity and the other parent is selected
from the belief space to influence the next generation Thebelief space keeps a list of all the individuals which meetsome criteria These criteria depend on what knowledgethe algorithm implements In this paper the circumstantialknowledge selects the best individuals found so far foreach objective function Thus one individual will give usinformation on the best value found for 119891
1and the other will
do the same for 1198912 The historical knowledge stores a list of
individuals with the best fitness value found so farThe fitnessvalue is calculated as the hypervolume S that is covered by anindividual Finally normative knowledge considers a list ofindividuals which are pairwise nondominated with respect tothe other individuals of their generation
Let |119869| be the number of available facilities and let |119868|
be the number of customers of our BOUFLP In this paperdecisions variables 119909 and 119910 are represented by a binary |119869|-length vector and |119869| times |119868| matrix respectively Comparingtwo different solutions (individuals) needs an evaluationcriterion In this paper we use the same criterion explainedin [18]
3 Computational Experiments
In this section we present the set of instances that are usedin this study as well as results obtained by our BOCAimplementation
31 Instances Presentation The instances that were used inthis paper correspond to random instances using a problemgenerator that follows the methodology from UflLib [36]Previous works in the literature have also used this problemgenerator to create their test instances [18 26]
The BOCA algorithm has several parameters that need tobe set As in [18] the number of generations considered inthis paper is equal to 100 Population size 119871 is set equal to 100mutation probability in the population space 119875ps is equal to02 and probability of mutation in the belief space 119875bs is 004Both 119871 and 119875ps values are different from the values used in[18] These values are chosen as they all together yield to thebest performance of the algorithm given some test instancesThus although resulting values are different from that used in[18] the method we use to set them is the same as that usedin that work This is important in order to fairly compare thedifferent BOCA algorithms
32 Results and Discussion In this section we compare theresults obtained by the previous BOCAalgorithm (see [18] forfurther details) and our approach Moreover a comparisonbetween results obtained by well-known EMO algorithmssuch as NSGA-II and PAES and our BOCA algorithm is alsopresented in this section
Tables 1 and 2 show the results obtained by the BOCAimplementations using historical [18] circumstantial andnormative knowledge respectively In the same way Tables 3and 4 present the results obtained by the well-known NSGA-II and PAES algorithms For each algorithm 119878 value () time119905 (in seconds) and the number of efficient solutions 119909 isin 119883
have been included in these tables As we mentioned before
6 The Scientific World Journal
Table 1 Results obtained by the BOCA implementations for instance of class 119860
Instance BOCA (historical) BOCA (circumstantial) BOCA (normative)1199051(sec) S
we want to produce a set with a large number of efficientsolutions 119909 isin 119883 a 119878 value close to 100 (ideal) and a small 119905For the sake of easy reading we have split the set of instancesinto two subsets (instances type 119860 and 119861)
We then compare our BOCA implementations with theone presented in [18] Tables 5 and 6 show a comparisonbetween those algorithms As we can see when compared interms of its 119878 value (the bigger the better) BOCA algorithm
using historical knowledge (BOCAH) performs consistentlybetter than the ones using circumstantial (BOCAC) andnormative (BOCAN) knowledge In fact BOCAH obtains a 119878
value that is in average 58 bigger than the one obtainedby BOCAC and 65 bigger than the 119878 value obtained byBOCAN When compared in terms of the CPU time neededto reach the number of iterations (generations) BOCAH isin average faster than both BOCAC and BOCAN algorithms
The Scientific World Journal 7
Table 3 Results obtained bywell-knownMOEA algorithmsNSGA-II and PAES for instances of class 119860
We can note that for 119861 instances times required by BOCAH
and BOCAC are in average quite similar (only 16 ofdifference) Finally when we look at the number of efficientsolutions found by each algorithm (|119883| column) we can seethat again BOCAH outperforms both BOCAC and BOCAN
algorithms In this case the average number of efficient
Table 5 Comparison among our BOCA implementations (119860instances)
solutions found by the BOCAH algorithm is about 20biggerthan the one obtained by the other two approaches
Results above are consistent with the good performanceobtained by the BOCAH approach in [18] Moreover resultsshow that performance of the BOCA algorithm dependslargely on the selected knowledge and it can make thedifference in terms of 119878 value time and number of efficientsolutions found by the algorithmThis is an important finding
8 The Scientific World Journal
Table 7 Comparison among our BOCAwith circumstantial knowl-edge and NSGA-II and PAES
as it points out the relevance of the choice of a specific type ofknowledge
We now compare BOCAC and BOCAN algorithms tothe well-known NSGA-II and PAES algorithms Tables 7 and8 show a comparison between our BOCAC algorithm andthe NSGA-II and PAES algorithms As we can see althoughBOCAC obtains in average a 119878 value 68 lower than the
Table 9 Comparison among our BOCAwith normative knowledgeand NSGA-II and PAES
one obtained by the NSGA-II algorithm it is more thanthree times faster Moreover when BOCAC is comparedto PAES algorithm the obtained 119878 values are in averageequivalent while BOCAC is around 30 faster than PAESPAES obtains in average more efficient points than BOCAC
though (932)Finally Tables 9 and 10 show a comparison between
BOCAN and NSGA-II and PAES algorithms BOCAN per-forms quite similar to PAES algorithm with respect to both119878 value and the number of obtained efficient solutionsHowever BOCAN is faster than PAES Similar situationoccurs when BOCAN is compared to NSGA-II algorithmAlthough NSGA-II obtains better values for both 119878 and |119883|BOCAN is much faster than NSGA-II This situation canbe explained by the very fast performance that our BOCAN
algorithm obtains for the set of small instances When welook further at the results we can note that if we onlyconsider both medium and large size instances executiontimes obtained by both algorithms are quite similar to eachotherThis result confirmswhat is outlined in [18] in the senseof the good performance that the BOCA algorithm showsFurthermore our results confirm this good performancewith respect to other well-known EMO algorithms does notdepend on which type of knowledge is considered Howeveras we mentioned before the choice of the knowledge usedon the BOCA algorithm is an important issue and it has animpact on the algorithm performance
4 Conclusions and Future Work
Evolutionary algorithms are a very good alternative to solvecomplex combinatorial optimisation problems In this paper
The Scientific World Journal 9
Table 10 Comparison among our BOCA with normative knowl-edge and NSGA-II and PAES
we have implemented a biobjective cultural algorithm to solvethe well-known BOUFLP We have considered two differentsources of knowledge namely circumstantial and normativeand compare them with a previously implemented historicalknowledge Furthermore we compare our BOCAapproacheswith two well-known EAs namely NSGA-II and PAES
Although BOCA approaches using both normative andcircumstantial knowledge could not improve the resultsobtained by the BOCA algorithm with the historical knowl-edge results pointed out that performance of the BOCAalgorithm depends largely on the selected knowledge andit can make the difference in terms of 119878 value time andnumber of efficient solutions found by the algorithm Thisis an important finding as it points out the relevance of thechoice of a specific type of knowledge Moreover our resultsalso confirm the good performance showed by the BOCAalgorithmwith respect to other well-known EMO algorithmssuch as NSGA-II and PAES algorithmsThe BOCA algorithmis very competitive when compared to those EMOalgorithmsindependently of the type of knowledge implemented
As a future work we think that more investigation isneeded in order to find patterns that allow us to get theright knowledge implemented depending on the problemfeatures As we mentioned before the knowledge choice hasan impact on the performance of the BOCA algorithm andtherefore it must be studied in depth Also as future workhybrid knowledge could be implemented in order to exploitthe advantages of each kind of knowledge at the same timeMoreover our BOCA algorithm can be used to solve other
interesting MOPs arising in the logistic field such as routingor scheduling problems
Appendix
Result Tables
In this appendix section obtained results are presentedColumns 119878
sdotshow the 119878 value obtained by algorithm sdot as
Algorithms are indexed as follows The original BOCAalgorithm is indexed by 1 BOCA algorithms using cir-cumstantial and normative knowledge are indexed by 2
and 3 respectively Finally the other EAs considered inthis paper namely NSGA-II and PAES are indexed by 4
and 5 respectively Columns 119905sdotshow the time obtained by
each algorithm in seconds Columns |119883sdot| show the number
of efficient solutions found by the corresponding algorithmFinally operator Δsdot
sdotsdotshows a value that is equivalent to (sdot minus
sdotsdot)sdot times 100
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] D Maravall and J de Lope ldquoMulti-objective dynamic opti-mization with genetic algorithms for automatic parkingrdquo SoftComputing vol 11 no 3 pp 249ndash257 2007
[2] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multiobjectiveoptimization Nsga IIrdquo in Parallel Problem Solving from NaturePPSN VI M Schoenauer K Deb G Rudolph et al Edsvol 1917 of Lecture Notes in Computer Science pp 849ndash858Springer Berlin Germany 2000
[3] I Borgulya ldquoAn algorithm for the capacitated vehicle routingproblem with route balancingrdquo Central European Journal ofOperations Research vol 16 no 4 pp 331ndash343 2008
[4] P J Angeline Z Michalewicz M Schoenauer X Yao and AZalzala Eds The Pareto Archived Evolution Strategy A NewBaseline Algorithm for Pareto Multiobjective Optimisation vol1 IEEE Press 1999
[5] C A Coello Coello and M Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 2002
[6] W K Mashwani ldquoComprehensive survey of the hybrid evolu-tionary algorithmsrdquo International Journal of Applied Evolution-ary Computation vol 4 pp 1ndash19 2013
[7] R Bhattacharya and S Bandyopadhyay ldquoSolving conflicting bi-objective facility location problem by NSGA II evolutionaryalgorithmrdquo International Journal of Advanced ManufacturingTechnology vol 51 no 1ndash4 pp 397ndash414 2010
[8] B Crawford C Lagos C Castro and F Paredes ldquoA culturalalgorithm for solving the set covering problemrdquo inAnalysis andDesign of Intelligent Systems using Soft Computing TechniquesP Melin O Castillo E Ramırez J Kacprzyk and W PedryczEds vol 41 of Advances in Soft Computing pp 408ndash415Springer Berlin Germany 2007
10 The Scientific World Journal
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014
we want to produce a set with a large number of efficientsolutions 119909 isin 119883 a 119878 value close to 100 (ideal) and a small 119905For the sake of easy reading we have split the set of instancesinto two subsets (instances type 119860 and 119861)
We then compare our BOCA implementations with theone presented in [18] Tables 5 and 6 show a comparisonbetween those algorithms As we can see when compared interms of its 119878 value (the bigger the better) BOCA algorithm
using historical knowledge (BOCAH) performs consistentlybetter than the ones using circumstantial (BOCAC) andnormative (BOCAN) knowledge In fact BOCAH obtains a 119878
value that is in average 58 bigger than the one obtainedby BOCAC and 65 bigger than the 119878 value obtained byBOCAN When compared in terms of the CPU time neededto reach the number of iterations (generations) BOCAH isin average faster than both BOCAC and BOCAN algorithms
The Scientific World Journal 7
Table 3 Results obtained bywell-knownMOEA algorithmsNSGA-II and PAES for instances of class 119860
We can note that for 119861 instances times required by BOCAH
and BOCAC are in average quite similar (only 16 ofdifference) Finally when we look at the number of efficientsolutions found by each algorithm (|119883| column) we can seethat again BOCAH outperforms both BOCAC and BOCAN
algorithms In this case the average number of efficient
Table 5 Comparison among our BOCA implementations (119860instances)
solutions found by the BOCAH algorithm is about 20biggerthan the one obtained by the other two approaches
Results above are consistent with the good performanceobtained by the BOCAH approach in [18] Moreover resultsshow that performance of the BOCA algorithm dependslargely on the selected knowledge and it can make thedifference in terms of 119878 value time and number of efficientsolutions found by the algorithmThis is an important finding
8 The Scientific World Journal
Table 7 Comparison among our BOCAwith circumstantial knowl-edge and NSGA-II and PAES
as it points out the relevance of the choice of a specific type ofknowledge
We now compare BOCAC and BOCAN algorithms tothe well-known NSGA-II and PAES algorithms Tables 7 and8 show a comparison between our BOCAC algorithm andthe NSGA-II and PAES algorithms As we can see althoughBOCAC obtains in average a 119878 value 68 lower than the
Table 9 Comparison among our BOCAwith normative knowledgeand NSGA-II and PAES
one obtained by the NSGA-II algorithm it is more thanthree times faster Moreover when BOCAC is comparedto PAES algorithm the obtained 119878 values are in averageequivalent while BOCAC is around 30 faster than PAESPAES obtains in average more efficient points than BOCAC
though (932)Finally Tables 9 and 10 show a comparison between
BOCAN and NSGA-II and PAES algorithms BOCAN per-forms quite similar to PAES algorithm with respect to both119878 value and the number of obtained efficient solutionsHowever BOCAN is faster than PAES Similar situationoccurs when BOCAN is compared to NSGA-II algorithmAlthough NSGA-II obtains better values for both 119878 and |119883|BOCAN is much faster than NSGA-II This situation canbe explained by the very fast performance that our BOCAN
algorithm obtains for the set of small instances When welook further at the results we can note that if we onlyconsider both medium and large size instances executiontimes obtained by both algorithms are quite similar to eachotherThis result confirmswhat is outlined in [18] in the senseof the good performance that the BOCA algorithm showsFurthermore our results confirm this good performancewith respect to other well-known EMO algorithms does notdepend on which type of knowledge is considered Howeveras we mentioned before the choice of the knowledge usedon the BOCA algorithm is an important issue and it has animpact on the algorithm performance
4 Conclusions and Future Work
Evolutionary algorithms are a very good alternative to solvecomplex combinatorial optimisation problems In this paper
The Scientific World Journal 9
Table 10 Comparison among our BOCA with normative knowl-edge and NSGA-II and PAES
we have implemented a biobjective cultural algorithm to solvethe well-known BOUFLP We have considered two differentsources of knowledge namely circumstantial and normativeand compare them with a previously implemented historicalknowledge Furthermore we compare our BOCAapproacheswith two well-known EAs namely NSGA-II and PAES
Although BOCA approaches using both normative andcircumstantial knowledge could not improve the resultsobtained by the BOCA algorithm with the historical knowl-edge results pointed out that performance of the BOCAalgorithm depends largely on the selected knowledge andit can make the difference in terms of 119878 value time andnumber of efficient solutions found by the algorithm Thisis an important finding as it points out the relevance of thechoice of a specific type of knowledge Moreover our resultsalso confirm the good performance showed by the BOCAalgorithmwith respect to other well-known EMO algorithmssuch as NSGA-II and PAES algorithmsThe BOCA algorithmis very competitive when compared to those EMOalgorithmsindependently of the type of knowledge implemented
As a future work we think that more investigation isneeded in order to find patterns that allow us to get theright knowledge implemented depending on the problemfeatures As we mentioned before the knowledge choice hasan impact on the performance of the BOCA algorithm andtherefore it must be studied in depth Also as future workhybrid knowledge could be implemented in order to exploitthe advantages of each kind of knowledge at the same timeMoreover our BOCA algorithm can be used to solve other
interesting MOPs arising in the logistic field such as routingor scheduling problems
Appendix
Result Tables
In this appendix section obtained results are presentedColumns 119878
sdotshow the 119878 value obtained by algorithm sdot as
Algorithms are indexed as follows The original BOCAalgorithm is indexed by 1 BOCA algorithms using cir-cumstantial and normative knowledge are indexed by 2
and 3 respectively Finally the other EAs considered inthis paper namely NSGA-II and PAES are indexed by 4
and 5 respectively Columns 119905sdotshow the time obtained by
each algorithm in seconds Columns |119883sdot| show the number
of efficient solutions found by the corresponding algorithmFinally operator Δsdot
sdotsdotshows a value that is equivalent to (sdot minus
sdotsdot)sdot times 100
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] D Maravall and J de Lope ldquoMulti-objective dynamic opti-mization with genetic algorithms for automatic parkingrdquo SoftComputing vol 11 no 3 pp 249ndash257 2007
[2] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multiobjectiveoptimization Nsga IIrdquo in Parallel Problem Solving from NaturePPSN VI M Schoenauer K Deb G Rudolph et al Edsvol 1917 of Lecture Notes in Computer Science pp 849ndash858Springer Berlin Germany 2000
[3] I Borgulya ldquoAn algorithm for the capacitated vehicle routingproblem with route balancingrdquo Central European Journal ofOperations Research vol 16 no 4 pp 331ndash343 2008
[4] P J Angeline Z Michalewicz M Schoenauer X Yao and AZalzala Eds The Pareto Archived Evolution Strategy A NewBaseline Algorithm for Pareto Multiobjective Optimisation vol1 IEEE Press 1999
[5] C A Coello Coello and M Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 2002
[6] W K Mashwani ldquoComprehensive survey of the hybrid evolu-tionary algorithmsrdquo International Journal of Applied Evolution-ary Computation vol 4 pp 1ndash19 2013
[7] R Bhattacharya and S Bandyopadhyay ldquoSolving conflicting bi-objective facility location problem by NSGA II evolutionaryalgorithmrdquo International Journal of Advanced ManufacturingTechnology vol 51 no 1ndash4 pp 397ndash414 2010
[8] B Crawford C Lagos C Castro and F Paredes ldquoA culturalalgorithm for solving the set covering problemrdquo inAnalysis andDesign of Intelligent Systems using Soft Computing TechniquesP Melin O Castillo E Ramırez J Kacprzyk and W PedryczEds vol 41 of Advances in Soft Computing pp 408ndash415Springer Berlin Germany 2007
10 The Scientific World Journal
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014
We can note that for 119861 instances times required by BOCAH
and BOCAC are in average quite similar (only 16 ofdifference) Finally when we look at the number of efficientsolutions found by each algorithm (|119883| column) we can seethat again BOCAH outperforms both BOCAC and BOCAN
algorithms In this case the average number of efficient
Table 5 Comparison among our BOCA implementations (119860instances)
solutions found by the BOCAH algorithm is about 20biggerthan the one obtained by the other two approaches
Results above are consistent with the good performanceobtained by the BOCAH approach in [18] Moreover resultsshow that performance of the BOCA algorithm dependslargely on the selected knowledge and it can make thedifference in terms of 119878 value time and number of efficientsolutions found by the algorithmThis is an important finding
8 The Scientific World Journal
Table 7 Comparison among our BOCAwith circumstantial knowl-edge and NSGA-II and PAES
as it points out the relevance of the choice of a specific type ofknowledge
We now compare BOCAC and BOCAN algorithms tothe well-known NSGA-II and PAES algorithms Tables 7 and8 show a comparison between our BOCAC algorithm andthe NSGA-II and PAES algorithms As we can see althoughBOCAC obtains in average a 119878 value 68 lower than the
Table 9 Comparison among our BOCAwith normative knowledgeand NSGA-II and PAES
one obtained by the NSGA-II algorithm it is more thanthree times faster Moreover when BOCAC is comparedto PAES algorithm the obtained 119878 values are in averageequivalent while BOCAC is around 30 faster than PAESPAES obtains in average more efficient points than BOCAC
though (932)Finally Tables 9 and 10 show a comparison between
BOCAN and NSGA-II and PAES algorithms BOCAN per-forms quite similar to PAES algorithm with respect to both119878 value and the number of obtained efficient solutionsHowever BOCAN is faster than PAES Similar situationoccurs when BOCAN is compared to NSGA-II algorithmAlthough NSGA-II obtains better values for both 119878 and |119883|BOCAN is much faster than NSGA-II This situation canbe explained by the very fast performance that our BOCAN
algorithm obtains for the set of small instances When welook further at the results we can note that if we onlyconsider both medium and large size instances executiontimes obtained by both algorithms are quite similar to eachotherThis result confirmswhat is outlined in [18] in the senseof the good performance that the BOCA algorithm showsFurthermore our results confirm this good performancewith respect to other well-known EMO algorithms does notdepend on which type of knowledge is considered Howeveras we mentioned before the choice of the knowledge usedon the BOCA algorithm is an important issue and it has animpact on the algorithm performance
4 Conclusions and Future Work
Evolutionary algorithms are a very good alternative to solvecomplex combinatorial optimisation problems In this paper
The Scientific World Journal 9
Table 10 Comparison among our BOCA with normative knowl-edge and NSGA-II and PAES
we have implemented a biobjective cultural algorithm to solvethe well-known BOUFLP We have considered two differentsources of knowledge namely circumstantial and normativeand compare them with a previously implemented historicalknowledge Furthermore we compare our BOCAapproacheswith two well-known EAs namely NSGA-II and PAES
Although BOCA approaches using both normative andcircumstantial knowledge could not improve the resultsobtained by the BOCA algorithm with the historical knowl-edge results pointed out that performance of the BOCAalgorithm depends largely on the selected knowledge andit can make the difference in terms of 119878 value time andnumber of efficient solutions found by the algorithm Thisis an important finding as it points out the relevance of thechoice of a specific type of knowledge Moreover our resultsalso confirm the good performance showed by the BOCAalgorithmwith respect to other well-known EMO algorithmssuch as NSGA-II and PAES algorithmsThe BOCA algorithmis very competitive when compared to those EMOalgorithmsindependently of the type of knowledge implemented
As a future work we think that more investigation isneeded in order to find patterns that allow us to get theright knowledge implemented depending on the problemfeatures As we mentioned before the knowledge choice hasan impact on the performance of the BOCA algorithm andtherefore it must be studied in depth Also as future workhybrid knowledge could be implemented in order to exploitthe advantages of each kind of knowledge at the same timeMoreover our BOCA algorithm can be used to solve other
interesting MOPs arising in the logistic field such as routingor scheduling problems
Appendix
Result Tables
In this appendix section obtained results are presentedColumns 119878
sdotshow the 119878 value obtained by algorithm sdot as
Algorithms are indexed as follows The original BOCAalgorithm is indexed by 1 BOCA algorithms using cir-cumstantial and normative knowledge are indexed by 2
and 3 respectively Finally the other EAs considered inthis paper namely NSGA-II and PAES are indexed by 4
and 5 respectively Columns 119905sdotshow the time obtained by
each algorithm in seconds Columns |119883sdot| show the number
of efficient solutions found by the corresponding algorithmFinally operator Δsdot
sdotsdotshows a value that is equivalent to (sdot minus
sdotsdot)sdot times 100
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] D Maravall and J de Lope ldquoMulti-objective dynamic opti-mization with genetic algorithms for automatic parkingrdquo SoftComputing vol 11 no 3 pp 249ndash257 2007
[2] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multiobjectiveoptimization Nsga IIrdquo in Parallel Problem Solving from NaturePPSN VI M Schoenauer K Deb G Rudolph et al Edsvol 1917 of Lecture Notes in Computer Science pp 849ndash858Springer Berlin Germany 2000
[3] I Borgulya ldquoAn algorithm for the capacitated vehicle routingproblem with route balancingrdquo Central European Journal ofOperations Research vol 16 no 4 pp 331ndash343 2008
[4] P J Angeline Z Michalewicz M Schoenauer X Yao and AZalzala Eds The Pareto Archived Evolution Strategy A NewBaseline Algorithm for Pareto Multiobjective Optimisation vol1 IEEE Press 1999
[5] C A Coello Coello and M Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 2002
[6] W K Mashwani ldquoComprehensive survey of the hybrid evolu-tionary algorithmsrdquo International Journal of Applied Evolution-ary Computation vol 4 pp 1ndash19 2013
[7] R Bhattacharya and S Bandyopadhyay ldquoSolving conflicting bi-objective facility location problem by NSGA II evolutionaryalgorithmrdquo International Journal of Advanced ManufacturingTechnology vol 51 no 1ndash4 pp 397ndash414 2010
[8] B Crawford C Lagos C Castro and F Paredes ldquoA culturalalgorithm for solving the set covering problemrdquo inAnalysis andDesign of Intelligent Systems using Soft Computing TechniquesP Melin O Castillo E Ramırez J Kacprzyk and W PedryczEds vol 41 of Advances in Soft Computing pp 408ndash415Springer Berlin Germany 2007
10 The Scientific World Journal
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014
as it points out the relevance of the choice of a specific type ofknowledge
We now compare BOCAC and BOCAN algorithms tothe well-known NSGA-II and PAES algorithms Tables 7 and8 show a comparison between our BOCAC algorithm andthe NSGA-II and PAES algorithms As we can see althoughBOCAC obtains in average a 119878 value 68 lower than the
Table 9 Comparison among our BOCAwith normative knowledgeand NSGA-II and PAES
one obtained by the NSGA-II algorithm it is more thanthree times faster Moreover when BOCAC is comparedto PAES algorithm the obtained 119878 values are in averageequivalent while BOCAC is around 30 faster than PAESPAES obtains in average more efficient points than BOCAC
though (932)Finally Tables 9 and 10 show a comparison between
BOCAN and NSGA-II and PAES algorithms BOCAN per-forms quite similar to PAES algorithm with respect to both119878 value and the number of obtained efficient solutionsHowever BOCAN is faster than PAES Similar situationoccurs when BOCAN is compared to NSGA-II algorithmAlthough NSGA-II obtains better values for both 119878 and |119883|BOCAN is much faster than NSGA-II This situation canbe explained by the very fast performance that our BOCAN
algorithm obtains for the set of small instances When welook further at the results we can note that if we onlyconsider both medium and large size instances executiontimes obtained by both algorithms are quite similar to eachotherThis result confirmswhat is outlined in [18] in the senseof the good performance that the BOCA algorithm showsFurthermore our results confirm this good performancewith respect to other well-known EMO algorithms does notdepend on which type of knowledge is considered Howeveras we mentioned before the choice of the knowledge usedon the BOCA algorithm is an important issue and it has animpact on the algorithm performance
4 Conclusions and Future Work
Evolutionary algorithms are a very good alternative to solvecomplex combinatorial optimisation problems In this paper
The Scientific World Journal 9
Table 10 Comparison among our BOCA with normative knowl-edge and NSGA-II and PAES
we have implemented a biobjective cultural algorithm to solvethe well-known BOUFLP We have considered two differentsources of knowledge namely circumstantial and normativeand compare them with a previously implemented historicalknowledge Furthermore we compare our BOCAapproacheswith two well-known EAs namely NSGA-II and PAES
Although BOCA approaches using both normative andcircumstantial knowledge could not improve the resultsobtained by the BOCA algorithm with the historical knowl-edge results pointed out that performance of the BOCAalgorithm depends largely on the selected knowledge andit can make the difference in terms of 119878 value time andnumber of efficient solutions found by the algorithm Thisis an important finding as it points out the relevance of thechoice of a specific type of knowledge Moreover our resultsalso confirm the good performance showed by the BOCAalgorithmwith respect to other well-known EMO algorithmssuch as NSGA-II and PAES algorithmsThe BOCA algorithmis very competitive when compared to those EMOalgorithmsindependently of the type of knowledge implemented
As a future work we think that more investigation isneeded in order to find patterns that allow us to get theright knowledge implemented depending on the problemfeatures As we mentioned before the knowledge choice hasan impact on the performance of the BOCA algorithm andtherefore it must be studied in depth Also as future workhybrid knowledge could be implemented in order to exploitthe advantages of each kind of knowledge at the same timeMoreover our BOCA algorithm can be used to solve other
interesting MOPs arising in the logistic field such as routingor scheduling problems
Appendix
Result Tables
In this appendix section obtained results are presentedColumns 119878
sdotshow the 119878 value obtained by algorithm sdot as
Algorithms are indexed as follows The original BOCAalgorithm is indexed by 1 BOCA algorithms using cir-cumstantial and normative knowledge are indexed by 2
and 3 respectively Finally the other EAs considered inthis paper namely NSGA-II and PAES are indexed by 4
and 5 respectively Columns 119905sdotshow the time obtained by
each algorithm in seconds Columns |119883sdot| show the number
of efficient solutions found by the corresponding algorithmFinally operator Δsdot
sdotsdotshows a value that is equivalent to (sdot minus
sdotsdot)sdot times 100
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] D Maravall and J de Lope ldquoMulti-objective dynamic opti-mization with genetic algorithms for automatic parkingrdquo SoftComputing vol 11 no 3 pp 249ndash257 2007
[2] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multiobjectiveoptimization Nsga IIrdquo in Parallel Problem Solving from NaturePPSN VI M Schoenauer K Deb G Rudolph et al Edsvol 1917 of Lecture Notes in Computer Science pp 849ndash858Springer Berlin Germany 2000
[3] I Borgulya ldquoAn algorithm for the capacitated vehicle routingproblem with route balancingrdquo Central European Journal ofOperations Research vol 16 no 4 pp 331ndash343 2008
[4] P J Angeline Z Michalewicz M Schoenauer X Yao and AZalzala Eds The Pareto Archived Evolution Strategy A NewBaseline Algorithm for Pareto Multiobjective Optimisation vol1 IEEE Press 1999
[5] C A Coello Coello and M Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 2002
[6] W K Mashwani ldquoComprehensive survey of the hybrid evolu-tionary algorithmsrdquo International Journal of Applied Evolution-ary Computation vol 4 pp 1ndash19 2013
[7] R Bhattacharya and S Bandyopadhyay ldquoSolving conflicting bi-objective facility location problem by NSGA II evolutionaryalgorithmrdquo International Journal of Advanced ManufacturingTechnology vol 51 no 1ndash4 pp 397ndash414 2010
[8] B Crawford C Lagos C Castro and F Paredes ldquoA culturalalgorithm for solving the set covering problemrdquo inAnalysis andDesign of Intelligent Systems using Soft Computing TechniquesP Melin O Castillo E Ramırez J Kacprzyk and W PedryczEds vol 41 of Advances in Soft Computing pp 408ndash415Springer Berlin Germany 2007
10 The Scientific World Journal
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014
we have implemented a biobjective cultural algorithm to solvethe well-known BOUFLP We have considered two differentsources of knowledge namely circumstantial and normativeand compare them with a previously implemented historicalknowledge Furthermore we compare our BOCAapproacheswith two well-known EAs namely NSGA-II and PAES
Although BOCA approaches using both normative andcircumstantial knowledge could not improve the resultsobtained by the BOCA algorithm with the historical knowl-edge results pointed out that performance of the BOCAalgorithm depends largely on the selected knowledge andit can make the difference in terms of 119878 value time andnumber of efficient solutions found by the algorithm Thisis an important finding as it points out the relevance of thechoice of a specific type of knowledge Moreover our resultsalso confirm the good performance showed by the BOCAalgorithmwith respect to other well-known EMO algorithmssuch as NSGA-II and PAES algorithmsThe BOCA algorithmis very competitive when compared to those EMOalgorithmsindependently of the type of knowledge implemented
As a future work we think that more investigation isneeded in order to find patterns that allow us to get theright knowledge implemented depending on the problemfeatures As we mentioned before the knowledge choice hasan impact on the performance of the BOCA algorithm andtherefore it must be studied in depth Also as future workhybrid knowledge could be implemented in order to exploitthe advantages of each kind of knowledge at the same timeMoreover our BOCA algorithm can be used to solve other
interesting MOPs arising in the logistic field such as routingor scheduling problems
Appendix
Result Tables
In this appendix section obtained results are presentedColumns 119878
sdotshow the 119878 value obtained by algorithm sdot as
Algorithms are indexed as follows The original BOCAalgorithm is indexed by 1 BOCA algorithms using cir-cumstantial and normative knowledge are indexed by 2
and 3 respectively Finally the other EAs considered inthis paper namely NSGA-II and PAES are indexed by 4
and 5 respectively Columns 119905sdotshow the time obtained by
each algorithm in seconds Columns |119883sdot| show the number
of efficient solutions found by the corresponding algorithmFinally operator Δsdot
sdotsdotshows a value that is equivalent to (sdot minus
sdotsdot)sdot times 100
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
References
[1] D Maravall and J de Lope ldquoMulti-objective dynamic opti-mization with genetic algorithms for automatic parkingrdquo SoftComputing vol 11 no 3 pp 249ndash257 2007
[2] K Deb S Agrawal A Pratap and T Meyarivan ldquoA fast elitistnon-dominated sorting genetic algorithm for multiobjectiveoptimization Nsga IIrdquo in Parallel Problem Solving from NaturePPSN VI M Schoenauer K Deb G Rudolph et al Edsvol 1917 of Lecture Notes in Computer Science pp 849ndash858Springer Berlin Germany 2000
[3] I Borgulya ldquoAn algorithm for the capacitated vehicle routingproblem with route balancingrdquo Central European Journal ofOperations Research vol 16 no 4 pp 331ndash343 2008
[4] P J Angeline Z Michalewicz M Schoenauer X Yao and AZalzala Eds The Pareto Archived Evolution Strategy A NewBaseline Algorithm for Pareto Multiobjective Optimisation vol1 IEEE Press 1999
[5] C A Coello Coello and M Lechuga ldquoMOPSO a proposal formultiple objective particle swarm optimizationrdquo in Proceedingsof the Congress on Evolutionary Computation (CEC rsquo02) vol 2pp 1051ndash1056 2002
[6] W K Mashwani ldquoComprehensive survey of the hybrid evolu-tionary algorithmsrdquo International Journal of Applied Evolution-ary Computation vol 4 pp 1ndash19 2013
[7] R Bhattacharya and S Bandyopadhyay ldquoSolving conflicting bi-objective facility location problem by NSGA II evolutionaryalgorithmrdquo International Journal of Advanced ManufacturingTechnology vol 51 no 1ndash4 pp 397ndash414 2010
[8] B Crawford C Lagos C Castro and F Paredes ldquoA culturalalgorithm for solving the set covering problemrdquo inAnalysis andDesign of Intelligent Systems using Soft Computing TechniquesP Melin O Castillo E Ramırez J Kacprzyk and W PedryczEds vol 41 of Advances in Soft Computing pp 408ndash415Springer Berlin Germany 2007
10 The Scientific World Journal
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014
[9] Y Guo J Cheng Y Cao and Y Lin ldquoA novel multi-populationcultural algorithm adopting knowledge migrationrdquo Soft Com-puting vol 15 no 5 pp 897ndash905 2011
[10] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 World Scientic 1994
[11] C A C Coello G B Lamont and D A V VeldhuizenEvolutionary Algorithms for Solving Multi-Objective Problems(Genetic and Evolutionary Computation) Springer SecaucusNJ USA 2006
[12] C A Coello C Dhaenens and L Jourdan Advances in Multi-Objective Nature Inspired Computing Springer 1st edition 2010
[13] C A Coello and R Landa ldquoEvolutionary multiobjective opti-mization using a cultural algorithmrdquo in Proceedings of the IEEESwarm Intelligence Symposium pp 6ndash13 IEEE Service CenterPiscataway NJ USA 2003
[14] R Zhang J Zhou L Mo S Ouyang and X Liao ldquoEconomicenvironmental dispatch using an enhanced multi-objectivecultural algorithmrdquo Electric Power Systems Research vol 99 pp18ndash29 2013
[15] S Srinivasan and S Ramakrishnan ldquoA social intelligent systemfor multi-objective optimization of classification rules usingcultural algorithmsrdquo Computing vol 95 no 4 pp 327ndash3502013
[16] G G Cabrera C Vasconcellos R Soto J M Rubio F Paredesand B Crawford ldquoAn evolutionary multi-objective optimiza-tion algorithm for portfolio selection problemrdquo InternationalJournal of Physical Sciences vol 6 no 22 pp 5316ndash5327 2011
[17] R Reynolds and D Liu ldquoMulti-objective cultural algorithmsrdquoinProceedings of the IEEECongress of EvolutionaryComputation(CEC rsquo11) pp 1233ndash1241 June 2011
[18] G Cabrera J M Rubio D Dıaz B Fernandez C Cubillosand R Soto ldquoA cultural algorithm applied in a BiObjectiveuncapacitated facility location problemrdquo in Evolutionary Multi-Criterion Optimization R Takahashi K Deb EWanner and SGreco Eds vol 6576 of Lecture Notes in Computer Science pp477ndash491 Springer Berlin Germany 2011
[19] J Knowles and D Corne ldquoOn metrics for comparing non-dominated setsrdquo in Proceedings of the Congress on EvolutionaryComputation (CEC rsquo02) vol 1 pp 711ndash716 Honolulu HawaiiUSA May 2002
[20] I Kaliszewski Soft Computing for Complex Multiple CriteriaDecision Making vol 85 of International Series in OperationsResearch amp Management Science Springer 2006
[21] M Ehrgott Multicriteria Optimization Springer Berlin Ger-many 2nd edition 2005
[22] A Farhang-mehr and S Azarm ldquoMinimal sets of qualitymetricsrdquo in Proceedings of the 2nd International Conference onEvolutionary Multi-Criterion Optimization (EMO rsquo03) LectureNotes in Computer Science pp 405ndash417 Springer 2003
[23] M P Hansen and A Jaszkiewicz ldquoEvaluating the quality ofapproximations to the non-dominated setrdquo Tech Rep IMM-REP-1998-7 Institute of Mathematical Modelling TechnicalUniversity of Denmark 1998
[24] E Zitzler Evolutionary algorithms for multiobjective optimiza-tion methods and applications [PhD thesis] Swiss FederalInstitute of Technology (ETH) Zurich Switzerland 1999
[25] E Zitzler K Deb and LThiele ldquoComparison of multiobjectiveevolutionary algorithms empirical resultsrdquo Evolutionary Com-putation vol 8 no 2 pp 173ndash195 2000
[26] J G Villegas F Palacios andA LMedaglia ldquoSolutionmethodsfor the bi-objective (cost-coverage) unconstrained facility loca-tion problemwith an illustrative examplerdquoAnnals of OperationsResearch vol 147 pp 109ndash141 2006
[27] M Ehrgott and X Gandibleux ldquoHybrid metaheuristics formulti-objective combinatorial optimizationrdquo in Hybrid Meta-heuristics C Blum M J B Aguilera A Roli and M SampelsEds vol 114 of Studies in Computational Intelligence pp 221ndash259 Springer Berlin Germany 2008
[28] J Bramel and D Simchi-Levi The Logic of Logistics The-ory Algorithms and Applications for Logistics ManagementSpringer New York NY USA 1997
[29] M S Daskin Network and Discrete Location Models Algo-rithms and Applications Wiley-Interscience New York NYUSA 1st edition 1995
[30] Z Drezner and H Hamacher Facility Location Applicationsand Theory Springer Berlin Germany 2002
[31] R Z Farahani M SteadieSeifi and N Asgari ldquoMultiple criteriafacility location problems a surveyrdquo Applied MathematicalModelling vol 34 no 7 pp 1689ndash1709 2010
[32] C S Revelle and G Laporte ldquoThe plant location problem newmodels and research prospectsrdquo Operations Research vol 44no 6 pp 864ndash874 1996
[33] R G Reynolds New Ideas in Optimization McGraw-HillMaidenhead UK 1999
[34] R Landa Becerra and C A Coello Coello ldquoA cultural algorithmwith differential evolution to solve constrained optimizationproblemsrdquo in Advances in Artificial Intelligence (IBERAMIArsquo04) C Lemaıtre C Reyes and J A Gonzalez Eds vol 3315 ofLectureNotes inComputer Science pp 881ndash890 Springer BerlinGermany 2004
[35] C Soza R Landa M Riff and C Coello ldquoA cultural algo-rithm with operator parameters control for solving timetablingproblemsrdquo in Foundations of Fuzzy Logic and Soft ComputingP Melin O Castillo L Aguilar J Kacprzyk and W PedryczEds vol 4529 of Lecture Notes in Computer Science pp 810ndash819 Springer Berlin Germany 2007
[36] M Hoefer ldquoUflLib Benchmark Instances for the UncapacitatedFacility Location Problemrdquo 2014