TELK OMNIKA , V ol. 15, No . 3, September 2017, pp . 1247 1256 ISSN: 1693-6930, accredited A b y DIKTI, Decree No: 58/DIKTI/K ep/2013 DOI: 10.12928/telk omnika.v15.i3.3656 1247 HABCO: A Rob ust Ag ent on Hybrid Ant-Bee Colon y Optimization Abba Suganda Gir sang* 1 , Chun-W ei Tsai 2 , and Chu-Sing Y ang 2 1 Master in Computer Science , Bina Nusantar a Univ ersity , J akar ta, Indonesia 2 Depar tment of Computer Science and Engineer ing, National Chung-Hsing U niv ersity , T aichung, T aiw an 3 Institute of Computer and Comm unication Engineer ing, Depar tment of Electr i cal Engineer ing, National Cheng K ung Univ ersity , T ainan, T aiw an *Corresponding author , e-mail: agirsang@bin us .edu Abstract The pur pose of this research is to gener ate a rob ust agent b y combinin g bee colon y optimization (BCO) and ELU-Ants f or solving tr a v elin g salesman prob lem (TSP), called HABCO . The rob ust agents , called ant-bees , firstly are g rouped into three types scout, f ollo w er , recr uiter at each stages . Then, the bad agents are high proba b ly discarded, while the good agents are high probab ly dup licated in ear lier steps . This first tw o steps mimic BCO algor ithm. Ho w e v er , constr ucting tours such as choosing nodes , and updating pheromone are b uilt b y ELU-Ants method.T o e v aluate the perf or mance of the proposed algor ithm, HABCO is perf or med on se v er al benchmar k datasets and compared to A CS and BCO . The e xper imental results sho w that HABCO achie v es the better solution, either with or without 2opt. K e yw or ds: Ant Colon y Optimization, Bee Colon y Optimization, Hybr id, Rob ust Agent, T r a v eling S alesman Prob lem Cop yright c 2017 Univer sitas Ahmad Dahlan. All rights reser ved. 1. Intr oduction Intelligent algor ithm is popular to solv e se v er al optimization prob lems because if its per- f or mance [1-11].Some intelligent algor ithms such as ant colon y optimization [1-2], genetic algo- r ithm [3], par ticle s w ar m optimization [4-5], sim ulated annealing [6], cat s w ar m optimization [7], bee colon y optimization [8-10], and so f or th w ere proposed to solv e tr a v elling salesman prob lem (TSP). In ant colon y optimization (A CO), ants la y a pheromone tr ail in their paths while tr a v el- ing from their nest to f ood source [1] to find the shor test tour f or the f ood. Since then, man y researchers also contin ued this approach f or better results on v ar ious optimization prob lems [11- 13]. In [12] p roposed an adaptiv e par ameter control scheme f or de v eloping an adaptiv e A CS namely AA CS . The par ameters v alue in AA CS are adaptiv ely controlled based on inf or mation of pheromone tr ails distr ib ution in each iter ation that is utiliz ed to estimate the optimization state . Naimi and T aher inejad proposed ELU-Ants and KCC-Ants to modify A CS b y introduce ne w local update [13]. This research emphasiz ed that the update pheromone on ear lier step is more than the later step . The other researchers presented a combining A CO with the other methods [14-15]. Lucic et al [16-17] then proposed another intelligent algor ithm bee colon y opti- mization (BCO),which is designed to create the m ulti-agent system (colon y of ar tificial bees) f or solving the combinator ial optimization prob lems [17-24]. This research is tr ied to adopt ELU-Ants algor ithm which considers the beginning step as the more impor tant step to constr uct the tour . Logically , an agent with bad choice of tr a v eling some cities (par t of tour) should be discarded b ut an agent with good choice should be duplicated at the beginning step . The discarded and duplicated agent in ear lier step guar antees the agent will be more competitiv e . Theref ore the update pheromone on ear lier step is more than the later step . In spite of star ting with good or bad choice , ants m ust contin ue selecting a path f or completing their tour . Receiv ed Ma y 17, 2016; Re vised Ma y 4, 2017; Accepted J une 1, 2017 Evaluation Warning : The document was created with Spire.PDF for Python.
1248 ISSN: 1693-6930 Figure 1. Illustr ation agent M tr a v els cities with tw o step constr uctions . Fig. 1 descr ibes the a gents M constr ucts the tour which is split into tw o impor tant step constr ucting. Suppose in beginning step , agents M constr uct the par t tour which consists i cities . Each agents tr a v els cities C 1 , C 2 , C 3 ,..., C i at beginning step . While constr ucting the tour , the ant is possib le to get bad choice at first stage and leads tr apped on bad constr ucting as w ell. At last it probab ly br ings to the bad-tour . In BCO , after tr a v eling some nectars or stage , each bees bac ks to the hiv e to comm u- nicate their w a y . In the hiv e , the y g roup into three type a nd perf or m w aggle dance to identify their quality of f ood source . The longer dance indicates the better quality of nectars f ound. Bees e xchange inf or mation to each other f or their ne xt stage-tour . Bees finding bad-source f ood tend to f ollo w the others with good-source f ood at ne xt stage . This step sho ws that BCO discards bee with bad stage-tour in order to get the better tour f or the ne xt stage , and in other side automatically the bee with good stage-tour is duplicated. So based on discussion abo v e , the proposed algor ithm, HABCO , is gener ate d b y con- str ucting the rob ust agent as BCO algor ithm, ho w e v er the process of selecting nodes (cities) is done b y ELU-Ants . HABCO w as first presented b y Ab ba et al in a proceeding of a conf erence [25]. Ho w e v er this paper presented mor e detail of the related w or k and processing of judgment agent. 2. Related W ork 2.1. Ant Colon y System Ant colon y system (A CS) is one of the most widely used and best-perf or ming A CO . While ants search f ood, the y will deposit the pheromone on the path where the y passed. The ants will tend to choose the path whose the highest amount pheromone on it, and a v oid the lo w pheromone path. There are some substantial par ts to descr ibe A CS br iefly [1]. Constr uction T our : While the kth ant is in city i , the ne xt city j is selected from the un visited city set J k (i) according to the tw o possibilities , P k (i,j) , in Eq. (1) and in Eq. (2) : a. if q q 0 (e xploitation) j = ar g max b ( i; u ) : ( i; u ) c ; 2 J k ( i ) ; (1) TELK OMNIKA V ol. 15, No . 3, September 2017 : 1247 1256 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 1693-6930 1249 b . if q > q 0 (bias e xplor ation) P k ( i; j ) = [ ( i; j )] : [ ( i; j )] 2 J k ( i ) [ ( i; j )] : [ ( i; j )] ; (2) where q is a r andom n umber ; q 0 is a par ameter ; represents pheromone v alue . Local pheromone updating : Ants visit edge i , j and change the pheromone (i,j) edge le v el b y local updating r ule in Eq. (3) : ( i; j ) = (1 ) ( i; j ) + ( i; j ) ; (3) Global pheromone updating : The pheromone of edges on the globally best tour is updat ed b y global updating r ules Eq. (4) and Eq. (5). ( i; j ) = (1 ) ( i; j ) + ( i; j ) ; (4) ( i; j ) = 8 < : ( L g b ) 1 if ( i; j ) 2 global best tour 0 Otherwise (5) 0 < < 1 is the pheromone deca y par ameter , and L g b is the length of the globally best tour . 2.2. ELU-Ants ELU-Ants [4] is a modified of the or iginal A CS . Hosein in his research off ered t w o equation to implement the algor ithm, one is named Kcc-Ants and the another one is ELU-Ants . Al though both of them used the similar approach, the y represented their proposed on diff erent equation. The results of both equation are almost similar . This research uses the ELU-Ants approach. The main idea of this research is the ants ha v e more ability to add more the ph eromone of the pr imar y links than the last links . While in ear ly steps , the ant has a lot of options to choose which city tr a v eled, because it has passed just a f e w cities . As a consequence , it mak es just a f e w edge are prohibited to be selected. Theref ore , the ant can freely choose the most desir ab le link (with more pheromone and less length) as its ne xt link. Contr ar y on final steps of tour , the ant already passed most of the cities , and the current selecte d link ma y not ha v e a significant eff ect on the quality of the tour , so it seems logical to reduce its ability of changing the last link pheromone . In other w ords , ants ha v e more eff ect on pheromone update where the y are in their initial steps and less eff ect when the y are going to finish the tour . Based on abo v e discussion, local update or iginal (Eq.(3)) is modified to be ELU-Ants (Eq. (6)) and Kcc-Ants (Eq. (7)). ( i; j ) = ( i; j ) + 0 :e 5 :cc n ; (6) ( i; j ) = ( i; j ) + K :cc: 0 cl cc ! ; (7) where cc represents the n umber of cities passed till no w; n is total n umber of cities ; K and ! denote tw o par ameters giv en; Cl represents c u rrent length of passed path of each ants and represents initial pheromone . 2.3. Bee Colon y Optimization Bee colon y optimization (BCO) is designed to create the m ulti agent system to solv e the combinator ial optimization prob lems . Lucic et al [16-17] proposed BCO to solv e TSP . On implementation BCO on TSP , the tour consists n umerous stages . A stage is a collection of a cer tain amount of nectars . In TSP model, nectars can represent cities . In constr ucting the tour , a be e star ts as an unemplo y ed f or ager without kno wledge about the f ood ( nectar). Bees star t e xplor ing from nectar to nectars or tr a v el. This step is called f orw ard pass (FP). As a notion, not all bees star t f or aging in the first stage . Some of them star t at the second stage , third stage , etc. HABCO: A Rob ust Agent on Hybr id Ant-Bee Colon y Optimization (Ab ba Suganda Girsang) Evaluation Warning : The document was created with Spire.PDF for Python.
1250 ISSN: 1693-6930 Once a bee is on, she will be activ e till the end of iter ation. After perf or ming one stage , a bee retur ns to the hiv e in order to unload and store the nectar . It is called bac kw ard pass (BP). In BP , bees will perf or m w aggle dance to sho w their quality tour and star t e xchanging inf or mation. The kind of comm unication betw een individual bees contr ib utes to the f or mation of the bee colon y . Bef ore comm unicating, BCO selects r andomly with a par ticular probability (e .g.10 %) to be scout. The bees which retain the pre vious stage and contin ue the ne xt stage without inter acting with the others are named scout. The remind bees will be categor iz ed into f ollo w er and recr uiter . BCO uses Eq. (8) f or deter mining the probability that the bee k in ne xt stage (u+1) with the same par tial tour at stage u in iter ation z as descr ibed belo w : P k ( u + 1 ; z ) = e L k ( u;z ) min r 2 w ( u;z ) ( L r ( u;z )) uz (8) where L k (u,z) is the length stage-tour that is disco v ered b y bee k in stage u in iter ation z . As a result of inter action, in addition of scout, the bees ha v e the tw o other types , recr uiter and f ollo w er . The bees which retain their pre vious stage and recr uit the f ollo w er bees on their w a y are named recr uiter . The bee s which abandon the f ood source (pre vious stage) and f ollo w the recr uiter bees on their w a y are named f ollo w er . Since the probabilit y of best path bee will be 1 ( P k = 1), she will use absolutely the same par tial and categor iz ed as g roup of recr uiter . BCO introduced the probability of par tial tour will be chosen b y an y bee that decided to choose the ne w route in one stage . The probability is gener ated regarding tw o main attr ib ute the total length tour and n umber bee that are adv er tising the par tial tour . The smaller the nor maliz ed v alue of the total length, the better is the par tial tour , while the bigger nor maliz ed v alue of n umber bees , the better is the par tial tour . Procedure BCO can be descr ibed as on Fig. 2. Procedure BCO f or solving TSP Set par ameters and n umber stage F or each stages F or each bees Constr uct Solution Classify bee into scout, recr uiter , f ollo w er End bee All bees e xchange inf or mation End stage Local Search End Figure 2. BCO Algor ithm. 3. Pr oposed Algorithm 3.1. The Concept Since there is a consider ation f or beginning step and last step in constr ucting tour , the tour should be split into some par ts of tour . This par t of tour can be identified as the stage in BCO which consists some cities . The stage-tour agent constr ucted is e xpected to judge the quality agent. On BCO , after tr a v eling one stage , bees bac k to hiv e and g roup into three types f ollo w er , recr uiter , and scout based on their quality . This algor ithm uses ”ant-bee” as its agent and mimics the beha vior of bee on BCO . F ollo w er ant-bee is identified as the agent which perf or ms bad choice on its stage-tour . Recr uit ant-bee is identified as the agent which perf or m good choice on its stage-tour . Scout ant-bee is identified as the agent which alw a ys find the ne w alter nativ e tour . Fur ther , f ollo w er will change their st age-tour and f ollo w the stage-tour of recr uiter agents . Scouts neither f ollo w nor recr uit the other agent, y et k eep their stage-tour . Fig. 3 giv es the illustr ation ho w gener ates the competitiv e agents . Suppose there are three agents M 1 , M 2 and M 3 and three cities at first stage . S 11 , S 12 , S 13 are the collection of TELK OMNIKA V ol. 15, No . 3, September 2017 : 1247 1256 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 1693-6930 1251 Figure 3. Illustr ation ho w gener ates the competitiv e agent. cities tr a v eled a t the first stage b y agent M 1 , M 2 and M 3 respectiv ely . C 11 - C 21 - C 31 , C 12 - C 22 - C 32 , C 13 - C 23 - C 33 are the sequence cities tr a v eled at first stage b y M 1 , M 2 and M 3 respectiv ely . After tr a v eling the first stage , all agents are e v aluated to conclude their type . Suppose M 1 , M 2 and M 3 become recr uiter , f ollo w er and scout respectiv ely . At the end of first stage , M 1 as recr uiter k eeps its stage-tour , M 2 as f ollo w er changes its stage-tour and f ollo w M 1 s stage-tour , M 3 as scout k eeps its stage-tour . Theref ore , at beginning second stage the agents M 1 , M 2 and M 3 should contin ue the tour from city C 31 , C 31 , C 33 respectiv ely . The interesting question is ho w to mak e a good judgment in order to classify the quality of agents on constr ucting stage-tour . The sub section 3.3 will e xplain that the total amount pheromon e gathered on each stage can be used as a good judgment. 3.2. Construct T our In constr uction tour f or solving TSP , the ant-bee chooses cities with A CS algor ithm (Eq. (1) and Eq. (2)). After each ant-bees tr a v eling one stage using the r ule of A CS , it bac ks to hiv e to inter acting with each other (supposing distance f rom hiv e to each city=0), and calculating t he total amount of pheromone . HABCO also categor iz es ant-bees into three types based on pheromone gathered namely scout( S ), f ollo w er( F ), and recr uiter( R ) as BCO . Scout retains the pre vious stage and contin ues the ne xt stage without inter acting with the others , f ollo w er abandons the pre vious stage and f ollo ws the recr uiter , recr uiter retains her pre vious stage and recr uits the f ollo w er to join her pre vious stage tour . On HABCO , the judgment of quality of agents is based on the amount pheromone that agent gathers in each stages . J udgment is used to classify ant-bees into F ollo w er , and Recr uiter . Scout agent is firstly chosen 10 % from all agents . The scout agent guar antees the alter nativ e tour is k ept on the process . The remainder agents is categor iz ed as f ollo w er and recr uiter .The probability ant-bee into f ollo w er after tr a v eling one stage is used b y Eq. (9). P F = max i P M i =1 ( max i ) ; (9) HABCO: A Rob ust Agent on Hybr id Ant-Bee Colon y Optimization (Ab ba Suganda Girsang) Evaluation Warning : The document was created with Spire.PDF for Python.
1252 ISSN: 1693-6930 where i is agent inde x; M is the total n umber agent ant-bees; i represents pheromone gathered at each stages b y agent; max represents pherom one maxim um gathered at each stages b y agent. F rom Eq. (9) sho ws that agent which gathers the high pheromone has lo w probability to be f ollo w er . F or the highest pheromone ( max ) agent is impossib le to be f ollo w er ( P F =0). In other w ords , it will absolutely become a recr uiter . After categor iz ed the agents into f ollo w er and recr uiter , the f ollo w er will change he r tour and f ollo w one of the f ollo w er agent r andomly . After e xchanging inf or mation on hiv e , the agent ant-bees contin ue the ne xt stages until finish the tour . 3.3. Pher omone updating There are 3 types pheromone updating on HABCO . The y are local update , semi-global update and global update . Local updat e is perf or med as local updating r ule in A CS al gor ithm (Eq. (3)). While ant-bees choosing their edge , the y also modify their pheromone . After perf or ming one sta ge , HABCO updates the pheromone called semi-global update . This update is based on modified ELU-Ants algor ithm that ants ha v e more eff ect on pheromone update where the y are in their ear lier steps and less eff ect when the y are going to finish the tour . ( i; j ) = ( i; j ) + 0 :e 5 S j S j ; (10) where S is current stage; j S j is total n umber stage . Ob viously from Eq. (10), the pheromone after perf or ming the first stage ( S=1 ) is added more than the ne xt stage . So the ant-bees pla y f e w er impact in update pheromone when the y are in their final par t of tour . In last stage (final par t), S= total n umber stage , it mak es the impact pheromone update to w ard z ero . At last, the ph eromone on the edges are tr a v elled b y th e best ant-bee will be updated with global updating r ule as the A CS Eq. (4) and Eq. (5). 3.4. Ruin local optima In the obser v ation, the best length tour is possib le same in some iter ations in a ro w especially on the relativ e small benchmar k. The best ant-bee probab ly tr a v els on same edge in each iter ations . So , it might the process tr aps to be local optimal. This situation mak es the algor ithm sometimes f ails to get the optimal solution. T o a v oid this prob lem (r uin local optimal), the pheromone on that edges is reduced/e v apor ated, when the best of length tour is same f or t times iter ations ( t=100 ) in a ro w . r s = x: r s ; (11) where 0 < x < 1. V ar iab le is consider ed as a f air n umber . When x is set too high, r u ining of local optim um will f ail to be obtained. Contr ar ily , if x is set too small, it will decrease pheromone significantly and aff ect the constr uction of tour . The algor ithm of HABCO is descr ibed as Fig. 4. 4. P erf ormance and Anal ysis Results 4.1. Anal ysis Constructing the Rob ust Ag ent As af erementioned, the pheromone and distance nodes are tw o v ar iab les impor tant in A CO . This section probes whether the tw o v ar iab les can be to a judgment of the quality agent. One benchmar k TSP , st70 , is in v estigated to see impact of the distance and amount pheromone at each stages that is perf or med b y ants . This actions are tr ied on three stages while n umber agents are ten. Firstly the agent (ant) tr a v els all cities stage b y stage with f ollo w the A CS r ule on Eq. (2). W e obser v er tw o kind the agents . The first is the best agent which get the best complete tour . The second one is the w orst agent which get the w orst complete tour . While tr a v eling, the total distance and the amount of pheromone at each edges is added and calculated in one stage . After tr a v eling one stage , each agent contin ues ne xt stage without inter action each other . At last, after each agents completes the ir tour , the length complete tour and pheromone of each ants also be calculated and r ank ed. In t he obser v ation, consider ing b y calculating the distance tr a v ele d at one stage is not reliab le . It sho ws the best ant which gets the shor test distance tour has v ar ious position r ank on first stage . Fur ther , the w orst ant which tr a v els the longest distance has not alw a ys bad position TELK OMNIKA V ol. 15, No . 3, September 2017 : 1247 1256 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 1693-6930 1253 Set par ameters and n umber stage F or each stages F or each ant-bees constr uct solution; local update End ant-bees Semi-global update; Rank ed ant-bees F or each ant-bees //Bac k to hiv e Classify ant-bees into scout, recr uiter , f ollo w er End ant-bees All ant-bees e xchange inf or mation If (stage=n umber stage) Calculating the length complete tour of each ant-bees Global update b y the best ant-bee Ruin local optima End if End Stage Local Search Figure 4. HABCO Algor ithm. on first stage . It means that an ant can not guar antee getting the relativ ely shor t in the complete tour when tr a v eling relativ ely shor t at the first stage . Contr ar ily , when an ant tr a v els relativ ely long at first stage , she can not guar antee getting relativ ely long in the complete tour . It is because each ants is possib le to choose se v er al cities which ha v e shor t distance in the beginning, b ut she has no choice e xcept contin uing the long distance at last. Thus , the total tour might be still relativ e long at last. The second obser v ation is to rely on the total amount of pheromone in one stage-tour tr a v eled b y each ants . In this e xper iment, it sho ws that the ant which gathered a lot of pheromone at first stage tends to be a good or best ant in the complete tour . The other hand, the ant which get the shor test distance probab ly gets the high pheromone on its first stage . So , it is more reasonab le that total amount of pheromone at each stage indicates the quality sub-tour . The g reater total pheromone that the ant achie v es in one stage , the ant has high probability of getting the good complete tour . So if the amount at ear ly stage-tour is higher , the ant tends to get the best complete tour . Contr ar y , if the pheromone at ear ly stage-tour is lo w er , the ant tends to the w orst complete tour . As a notion, this research is perf or med on three stages . In our in v estigation, the result of f our or more stages is not be reliab le . It mean s the pheromone gathered b y agents at first stage can not be a good judgment of quality agents . Exper iments are conducted to e v aluate the perf or mance of HABCO . T ab le 1 sho ws the settings of v ar ious par ameters f or the proposed algor it hm.This paper details tw o compar ison.The first one is a compar ison HABCO and BCO , and the second one is a compar ison HBCO and A CS . All algor ithms are e x ecuted 10 times and nine dataset benchmar k TSP . T ab le 1. P ar ameter Setting of HABCO algor ithm P ar ameter Symbol Set v alue n umber ant-bee M 10 n umber stage S 3 iter ation nc 1000 deg ree pheromone B 2 par ameter local update 0.1 pheromone deca y 0.1 limit e xploitation q 0 10 % f actor r uin local optima X 0.95 T ab le 2 displa y the results of all algor ithms and optimal solution (BKS/best kno wn solu- HABCO: A Rob ust Agent on Hybr id Ant-Bee Colon y Optimization (Ab ba Suganda Girsang) Evaluation Warning : The document was created with Spire.PDF for Python.
1254 ISSN: 1693-6930 tion) as w ell. The first compar ison is HABCO and BCO . T ab le 4 sho ws HABCO+2opt outper- f or ms BCO+2opt. On some datasets (eil51, kroa1 50, a280), HABCO without 2opt outperf or ms BCO+2opt. At first HABCO uses A CS method (Eq. (1) and Eq. (2)) to constr uct the tour . A CS has an adv antage to mak e the edge more dynamic b y perf or m local update . With local update , pheromone o n edge visited is diminished and mak e it less desir ab le . Theref ore the premature con v erge can be a v oided. As a consequence , the agent (ant) has man y v ar ious options to de- v elop its tour . With only ten agents , A CS can get the good quality tour . In other hand, BCO has main adv antage on e xchange inf or mation agents (bees) on their hiv e . Bef ore bac king to hiv e , constr uction bee depends only the role on Eq. (7). It mak es the selection candidate node is more static. BCO has also a slightly comple x f or m ula than A CS , and also has man y v ar iab les to set. The diff erence setting sometimes gener ates the diff erent sign ificant results . The second compar i- son should be discussed is compar ing HABCO and A CS . T ab le 4 sho ws that HABCO outperf or ms A CS either with 2opt or without 2opt. HABCO also sho ws slightly better either in a v er age or best case . Although A CS and HABCO uses the same role to constr uct the tour (Eq. (1), Eq. (2)) at each stages , HABCO has an adv antage in producing the competitiv e tour of their agents b y dupli- cating the good sub tour and discarding the bad sub tour in their hiv e . Theref ore , in constr ucting the tour , HABCO has smar ter agents than A CS . T ab le 2. P erf or mance HABCO , A CS and BCO on nine benchmar k datasets Bench BKS A CS A CS+2opt BCO+2opt HABCO HABCO+2opt mar k A vg Best A vg Best A vg Best A vg Best A vg Best eil51 426 432.4 430 430.3 428 432.1 429 430.7 428 429.1 427 ber lin52 7542 7715.8 7542 7611.4 7542 7652.6 7542 7577.9 7542 7543.5 7542 st70 675 693.4 684 686.2 680 683.2 675 686.5 680 681.1 678 kroa100 21282 22054.8 21716 21568 21369 21927.3 21369 21789.9 21402 21413.5 21282 eil101 629 649.9 641 643.9 636 656.2 645 643 634 641.7 634 kroa150 26524 27476.9 26889 27219.1 26912 27452.4 27063 27139.1 26881 27015.9 26781 d198 15780 16480.2 16202 16207 15918 16351.5 16021 16421.5 16182 16030 15895 a280 2579 2739.6 2686 2669.9 2640 2834.1 2701 2702.4 2653 2665.1 2630 u724 41910 46363 44688 44136.4 43781 46516.7 44091 45117.8 43901 44002.3 43432 5. Conc lusion HABCO as combination three algor ithms , A CS , BCO and ELU-Ants , is an aff ectiv e algo- r ithm to solv e the TSP prob lem. In solving TSP , HABCO divides the tour into some stages lik es on BCO algor ithm. In tour ing some cities in a stage , HABCO uses the A CS method from choos- ing the nodes , and updating local and global updating pheromone . The quality of each agents HABCO is judged based on the pheromone gathered at each stages . The pheromone gathered in A CS algor ithm classifies agent into f ollo w er , recr uiter and scout agent lik e on BCO algor ithm. A good judgment on premature step mak es the probab ly good agent (ide ntified as recr uiter) du- plicated, and t he probability bad agent (identified as f ollo w er) is discarded. The scout agent is e xist to maintain the ne w alter nativ e t our . This process gener ates the competitiv e agent in each iter ation. This algor ithm also proposed the semi-global update pheromone to adopt the ELU-Ants algor ithm.Compar ing A CS and BCO , HABCO can solv e TSP prob lem better . It can be seen from solving some dataset TSP prob lem with local or without local search. 6. Ac kno wledg ement This ar ticle is etx ended v ersion of proceeding [25] with entitled ”A Hybr id Ant-Bee Colon y Optimization f or Solving T r a v eling Salesman Prob lem with Competitiv e Agents . Ref erences [1] Dor igo M, Gambardella L M. Ant Colon y System: A Cooper ativ e Lear ning Approach to The T r a v eling Salesman Prob lem. IEEE T r ansactions on Ev olutionar y Computation . 1997; 1(1): 5366. TELK OMNIKA V ol. 15, No . 3, September 2017 : 1247 1256 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA ISSN: 1693-6930 1255 [2] Girsang A S , Tsai C W and Y ang C S . Rectifying the Inconsistent Fuzzy Pref erence Matr ix in AHP Using a Multi-Objectiv e Bicr iter ionAnt. Neur al Processing Letters .2016; 44(2) : 519-538. [3] P otvin J V , Genetic Algor ithms f or the T r a v eling Salesman P rob lem. Annals of Oper ations Research . 1996; 63 : 339-370. [4] Clerc M. Discrete P ar ticle Sw ar m Optimization Illustr ated b y The T r a v eling Salesman Prob lem. Ne w optimization T echniques in Engineer ing . 2004; 219-239. [5] W ang K P , Hu ang L, Zhou C G, P ang W . P ar ticle Sw ar m Optimization f or T r a v eling Salesman Prob lem . Inter national Conf erence on Machine Lear ning and Cyber netics . 2003; 15831585. [6] Chu S C , Tsai P W , P an, J S . Cat S w ar m Optimization . PRICAI 2006: T rends in Ar tificial Intelligence . 2006; 854-858. [7] Lucic P . Modeling T r anspor tation Prob lems Using Concepts of Sw ar m Intelligence and Soft Computing. PhD Thesis . Virginia : F aculty of the Virginia P olytechnic Institute and State Uni- v ersity; 2002. [8] T eodoro vic P , Lucic P , Mar k o vic G, DellOrco M. Bee Colon y Optimization: Pr inciples and Ap- plications . 8th Seminar on Neur al Netw or k Applications in Electr ical Engineer ing, NEUREL. 2006. [9] W ong L P , Lo w M Y H, Chong C S . A Bee Colon y Optimization Algor ithm f or T r a v elling Sales- man Prob lem , Proceedings of Second Asia Inter national Conf erence on Modelling and Sim u- lation (AMS). 2008; 818-823. [10] Lucic P , T eodoro vic D . V ehicle Routing Prob lem with Un cer tain Demand at Nodes: The Bee System and Fuzzy Logic Approach. Fuzzy Sets in Optimization . 2003; 67-82. [11] Girsang A S , Tsai C W , Y ang, C S . A F ast Bee Colon y Optimization f or T r a v eling Sales- man Prob lem . Third Inter national Conf erence on Inno v ations in Bio-Inspired Computing and Applications (IBICA).Kaohsiung. 2012; 7-12. [12] Y u W J , and Zhang J . Pheromone-distr ib ution-based adaptiv e Ant Colon y System . Proceed- ings of the 12th ann ual conf erence on Genetic and e v olutionar y computation. P or tland. 2010. [13] Naimi H M, T aher inejad N. Ne w Rob ust and Efficient Ant Colon y Algor ithms: Using Ne w Inter pretation of Local Updating Process . Exper t Systems with Applications . 2009; 36 (1): 481-488. [14] Chen S M, Chien C Y . P ar alleliz ed Genetic Ant Colon y Systems f or Solving The T r a v eling Salesman Prob lem. Exper t Systems with Applications . 2011; 38 (4): 3873-3883. [15] Chen S M, Chien C Y . Solving The T r a v eling Salesman Prob lem B ased on The Genetic Sim ulated Annealing Ant Colon y System with P ar ticle Sw ar m Optimization T echniques . Exper t Systems with Applications . 2011; 38 (12): 14439-14450. [16] Lucic P . Modeling T r anspor tation Prob lems Using Concepts of Sw ar m Intell igence and Soft Computing . Virginia. 2002. [17] T eodoro vic D , Lucic P , Mar k o vic P , Orco M D . Bee Colon y Optimization: Pr inciples and Ap- plications . 8th Seminar on Neur al Netw or k Applications in Electr ical Engineer ing, NEUREL , 2006. [18] Lucic P , T eodoro vic D . V ehicle Routing Prob lem with Un c e r tain Demand at Nodes: The Bee System and Fuzzy Logic Approach. Studies in Fuzziness and Soft Computing . 2003; 126 : 67-82. [19] Kar aboga D , Bastur k B . A P o w erful and Efficient Algor ithm f or Numer ical Fu nction Opti- mization: Ar tificial Bee Colon y (ABC) Algor ithm. Jour nal of Global Optimization . 2007; 39(3): 459-471. [20] Benatchba K, Admane L, K oudil M. Using Bees to Solv e a Data-Mining Prob lem Expressed as a Max-Sat One . Ar tificial Intelligence and Kno wledge Engineer ing Applications: A Bioin- spired Approach . 2005 ; 75-86. [21] Chong C S , Lo w M Y H, Siv akumar A I, Ga y K L. A Bee Colon y Optimization Algor ithm to Job Shop Scheduling . Proceedings of Winter Sim ulation Conf erence . 2006; 1954-1961. [22] F athian M, Amir i B , Maroosi A. Application of Hone y-Bee Mating Optimization Algor ithm on Cluster ing. Applied Mathematics and Computation . 2007; 190 (2), 1502-1513. [23] W ang J , Zhan g D . Image Denoising Based on Ar tificial Bee Colon y and BP Neur al Netw or k. HABCO: A Rob ust Agent on Hybr id Ant-Bee Colon y Optimization (Ab ba Suganda Girsang) Evaluation Warning : The document was created with Spire.PDF for Python.
1256 ISSN: 1693-6930 TELK OMNIKA (T elecomm unication Computing Electronics and Control) . 2015; 13(2) : 614- 623. [24] Meng L, Y in S , Hu X. A Ne w Method Used f or T r a v eling Salesman Prob lem Based on Dis- crete Ar tificial Bee Colon y Algor ithm. TELK OMNIKA (T elecomm unication Computing Electron- ics and Control) . 2016; 14(1):342-348. [25] Girsang A S , Tsai C W , Y ang C S . A Hybr id Ant-Bee Colon y Optimization f or Solving T r a v eling Salesman Prob lem with Competitiv e Agents Proceeding in Mobile , Ubiquitous , and Intelligent Computing. 2014; 643-648. TELK OMNIKA V ol. 15, No . 3, September 2017 : 1247 1256 Evaluation Warning : The document was created with Spire.PDF for Python.