hs s a repostory copy of AN ADAPIVE LOCALIZAION SYSEM USING PARICLE SWARM OPIMIZAION IN A CIRCULAR DISRIBUION FORM. Whte Rose Research Onlne URL for ths paper: http://eprnts.whterose.ac.uk/118699/ Verson: Publshed Verson Artcle: Alhammad, Abdulraqeb, Hashm, Fazrulhsyam, Fadlee, Mohd et al. (1 more author) (2016) AN ADAPIVE LOCALIZAION SYSEM USING PARICLE SWARM OPIMIZAION IN A CIRCULAR DISRIBUION FORM. JURNAL EKNOLOGI. pp. 105-110. Reuse Items deposted n Whte Rose Research Onlne are protected by copyrght, wth all rghts reserved unless ndcated otherwse. hey may be downloaded and/or prnted for prvate study, or other acts as permtted by natonal copyrght laws. he publsher or other rghts holders may allow further reproducton and re-use of the full text verson. hs s ndcated by the lcence nformaton on the Whte Rose Research Onlne record for the tem. akedown If you consder content n Whte Rose Research Onlne to be n breach of UK law, please notfy us by emalng eprnts@whterose.ac.uk ncludng the URL of the record and the reason for the wthdrawal request. eprnts@whterose.ac.uk https://eprnts.whterose.ac.uk/
Jurnal eknolog AN ADAPIVE LOCALIZAION SYSEM USING PARICLE SWARM OPIMIZAION IN A CIRCULAR DISRIBUION FORM Abdulraqeb Alhammad a, Fazrulhsyam Hashm a*, Mohd Fadlee a, areq M. Sham b a Faculty of Engneerng, Unversty Putra Malaysa, 43300, Serdang, Malaysa b Faculty of Engneerng, Multmeda Unversty, 36100, Cyberjaya, Malaysa Full Paper Artcle hstory Receved 1 February 2015 Receved n revsed form 24 March 2015 Accepted 1 August 2015 *Correspondng author Fazrul@upm.edu.my Abstract rackng the user locaton n ndoor envronment becomes substantal ssue n recent research Hgh accuracy and fast convergence are very mportant ssues for a good localzaton system. One of the technques that are used n localzaton systems s partcle swarm optmzaton (PSO). hs technque s a stochastc optmzaton based on the movement and velocty of partcles. In ths paper, we ntroduce an algorthm usng PSO for ndoor localzaton system. he proposed algorthm uses PSO to generate several partcles that have crcular dstrbuton around one access pont (AP). he PSO generates partcles where the dstance from each partcle to the AP s the same dstance from the AP to the target. he partcle whch acheves correct dstances (dstances from each AP to target) s selected as the target. Four PSO varants, namely standard PSO (SPSO), lnearly decreasng nerta weght PSO (LDIW PSO), self-organzng herarchcal PSO wth tme acceleraton coeffcents (HPSO-VAC), and constrcton factor PSO (CFPSO) are used to fnd the mnmum dstance error. he smulaton results show the proposed method usng HPSO- VAC varant acheves very low dstance error of 0.19 meter. Keywords: Indoor localzaton system, partcle swarm optmzaton, Eucldean dstance 2016 Penerbt UM Press. All rghts reserved 1.0 INRODUCION Nowadays, navgaton systems have been wdely used n outdoor and ndoor envronments. here are many types of navgaton systems such as marne navgaton system, global poston system (GPS), wreless sensor network and robotc mappng. hese systems have been desgned to estmate the target at partcular places wthn network coverage. However, an ndoor localzaton system that usually covers a small area compared to outdoor envronment has a small locaton error. Recently, ndoor localzaton systems have become recent nterested research due to ncrease the trackng applcatons. here are many technques that are used for trackng the user such as rado frequency fngerprntng [1-2], angle of arrval (AOA) [3] and trangulaton [4]. he typcal localzaton system should have a good accuracy wth less complexty. However, the accuracy of the system depends on the type and sze of the envronments. hus, ndoor envronment requres a good accuracy compare to outdoor envronment. he ndoor localzaton systems requre dfferent types of pre-knowledge nformaton such as physcal testbed, receved sgnal strength (RSS) and orentaton of the user. Moreover, fngerprntng technque consders the most accurate technque that has mnor error dstance compare to other technques [5-6]. However, fngerprntng technque s sufferng from hgh system complexty whch conssts of two operatng phases (offlne and onlne phase) [7]. Besdes, t requres bg sze of rado map n order to obtan hgh locaton accuracy. here are several algorthms that are used for mprovng the accuracy of estmated locaton. One of the optmzaton algorthms s partcle swarm optmzaton whch was developed by [8]. PSO s a collecton of huge partcles whch have dfferent 78: 9 3 (2016) 105 110 www.jurnalteknolog.utm.my eissn 2180 3722
106 Abdulraqeb et al. / Jurnal eknolog (Scences & Engneerng) 78: 9 3 (2016) 105 110 postons and velocty. It used to mprove the accuracy of localzaton system. Recently, POS has consdered by many research communtes due to hgh convergence, more fxable and less complexty. he hgh accuracy obtans when the propagaton condton lne of sght (LOS) whereas non lne of sght (NLOS) condton acheves low accuracy [9]. In ndoor localzaton, the poston of the partcle represents the optmzed poston of the target, whereas the velocty represents the partcle movement wthn a partcular envronment. In ths paper, we propose a method to enhance the system accuracy by usng PSO whch generates a swarm of partcles around each AP crcularly. hs paper s organzed as follows. Secton 2 ntroduces the related work that has been done n the ndoor trackng envronment. Secton 3 explans n detals the methodology that has been proposed to mnmze the dstance error usng PSO. Secton 4 descrbes the smulaton setup of the ndoor envronment. Secton 5 shows the results obtaned and compare the results for the four PSO varants. Lastly, secton 6 summarzes ths work. 2.0 RELAED WORK he usage of PSO depends on the level of localzaton problem. Some of the researchers used a sngle object optmzaton model to solve smple problems of localzaton system. Random tme varable PSO algorthm s an adaptve PSO proposed by [10], whch called random tme varyng nerta weght and acceleraton coeffcents (PSO-RVIWAC). hs algorthm combned of two dfferent algorthms whch are: the random nerta weght (PSO-RANDIW) and tme-varyng acceleraton coeffcents (PSO-VAC). It used to enhance the performance of the orgnal PSO. However, ths algorthm does not compatble for NLOS envronments. In [11], the authors proposed a new dstrbuted selecton technque to mnmze the number of selected nodes n wreless sensor network. hese selected number of nodes consdered the man part of the technque to establsh the grd coordnate system. However, the localzaton accuracy depends on the grd nodes n the system. Unlke [12], a dstrbuted and cooperatve algorthm for multpath envronments n wreless sensor network s proposed. hs algorthm based on the multpath propagaton that allows sensors to cooperatvely self-localze wth respect to a sngle anchor node n the whole network. he sngle anchor node s computed by usng the range and drecton of arrval measurements. he authors n [13] proposed group dscrmnant (GD) algorthm to mprove the accuracy of the locaton fngerprntng. GS depends on the AP selecton approach whch focusng on measurng the localzaton capabltes of each group of APs. he mean error obtaned by GS s 3 meters for fve APs. Unlke the tradtonal technques that treat the APs based on ther ndvdual mportance [14-15]. he workng [16-20] focus on mult-objectve partcle swarm optmzaton to solve the localzaton ssues. he mult-objectve partcle swarm optmzaton mproved the accuracy and convergence of the localzaton system. However, n [16] the system suffered from slow convergence and lmtless sze of the archve. hus, the authors n [20] addressed these ssues by consderng the geometrc topology constrant to ncrease the accuracy of localzaton system and usng the global optmum soluton n order to get better convergence. However, ths method obtaned an average localzaton error of 10 meters when many number of anchor ponts were used. In ths work, we apply the SPSO [8], LDIW-PSO [21], HPSO-VAC [22], and CFPSO [23] n order to fnd the mnmum dstance error. In LDIW-PSO, the Inerta weght s decreasng durng the searchng process based on the followng formula: where t wt w w w w max s 0.9, max mn mn w mn t s 0.4, (1) s the maxmum number of teratons, and s the current teraton. HPSO-VAC mproved the performance of the SPSO by controllng the acceleraton coeffcents. he values of where c 1 c 1 and c 2 are expressed as follows: c, c 1 2 t c1 c1 f c1 c1 (2) t c2 c2 f c2 c2 (3) s 2.5, s 0.5, s 0.5, and s 2.5. c 1 f he velocty of the SPSO has been modfed by Clerc and kenndey [23] resultng n a new PSO varant named constrcton factor PSO. he velocty of the constrcton factor PSO s wrtten as follows: vd t c1r 1 Pbestd t xd t vd t1 (4) c2r2gbestd t xd t where 2k 2 2 4 c 2 and c1 c2, k (0,1] 3.0 MEHODOLOGY PSO s used to estmate the user locaton n ndoor envronments. We assume that there are fve access ponts, whch are deployed n the ndoor envronment wth dmenson of 80 m x 50 m. he coordnates of the four access ponts are determned. At the ntal stage the dstance between access ponts and the target locaton s calculated usng Equaton (5), c 2 f
107 Abdulraqeb et al. / Jurnal eknolog (Scences & Engneerng) 78: 9 3 (2016) 105 110 D X x Y y Max d * AP, t AP t AP t m n AP 1,,5 where (X,Y ) and (x,y ) Indcates the locaton of deployed access ponts and estmated target locaton respectvely. Maxd s the maxmum value of measured dstance error. represents the random number whch s defned as a unform dstrbuton between. he objectve of ths work to mnmze the dstance error usng the followng equaton: error mn( ( t e) ( t e) ) (5) d x x y y (6) where x and ndcate estmated locaton of the e target. he dstance from the access pont to the target can be measured usng RSS. However, the poston of the target cannot be determned snce there are other coordnates that can produce the same dstance. o overcome ths problem we proposed a method that generates N partcles n a crcular dstrbuton usng PSO. he numbers of partcles that are generated n the crcle are abbrevated as NPC. It should be noted that each partcle represents a poston. Each of these partcles s a potental to be the target (snce all the generated partcles have the same dstance as the dstance from the AP to the target). o confrm that our soluton produces the exact locaton of the target, we test every generated partcle n the crcle by calculatng the dstance from each AP to that generated partcle and confrmng that the selected partcle satsfes all dstances (d1,d2,.d5) (It s noteworthy that we choose the AP that acheves the hghest RSS whch has the mnmum measurement dstance error). At the begnnng of the process, we generate N partcles randomly n the area. In order to get the best partcle that has the same dstance as DAP,t. hen, n each teraton, PSO s used to fnd the partcle whch has the closest dstance as DAP,t by usng the followng ftness functon: y e f ( d) mn abs( D D ) Ap, t Ap, P (7) Where DAP,p s the dstance from the AP to the partcle whch s wrtten as follows: D, ( x x ) ( y y ) (8) Ap p Ap p AP p he accuracy of the system depends on the optmzaton ftness functon. In other word, the optmzaton ftness functon s nversely proportonal to the system accuracy. At the end of the PSO teratons, PSO records the best partcle that has the same dstance as DAP,t. hs process s repeated K tmes where n each tme a new partcle (poston) s generated around the crcle. After that, we test each generated partcle to confrm that t satsfes all other dstances. Our methodology for the SPSO s explaned n the followng steps: SEP 1: Defne the swarm sze N. SEP 2: Locate the postons of APs randomly n the area SEP 3: Measure the dstances from each AP (d1,d2,., dx) to the target usng RSS. SEP 4: Determne the AP that acheves the hghest RSS (the mnmum measurement dstance error) and record ts dstance from target SEP 5: For =1:N. SEP 6: Generate the postons of N partcles and defne them as X and set Pbest =X and V=0 SEP 7: For t=1:. SEP 8: Calculate the dstance from AP (hghest RSS) to the poston of the partcle ( ) usng Equaton 8. SEP 9: If f( x ) Pbest < x f ( Pbest ) x then do End SEP 10: Calculate the dstance from AP (hghest RSS) to the poston of the partcle usng Pbest SEP 11: Record the gbest whch s the best partcle that acheves mn abs( DAp, t DAp, P ) SEP 12: Update the velocty of the partcle usng Equaton 9. SEP 13: Update the poston of the partcle usng Equaton 10. SEP 14: Next and SEP 15: Stop when all the N partcles have been generated. SEP 16: For each generated partcle, fnd the partcle that satsfes all the dstances from the APs. SEP 17: End for loop. SEP 18: Return the poston of ths partcle and calculate the dstance error usng Equaton 6. he flowchart of the proposed method that used to estmate the target locaton s shown n Fgure 2. PSO uses poston and velocty update equatons whch can be wrtten as follows: v wv ( t) c r ( Pbest ( t) d d 11 d x ( t)) c r ( gbest ( t) x ( t)) d d d (9) x ( t 1) x ( t) v ( t 1) (10) d d d
108 Abdulraqeb et al. / Jurnal eknolog (Scences & Engneerng) 78: 9 3 (2016) 105 110 where c1 and c2 ndcates to the acceleraton constants, w s the nerta weght, r1, r2, are unformly dstrbuted random numbers between 0 and 1. 4.0 SIMULAION SEUP he testbed dmenson of envronment s 80m X 50m wth fve APs are deployed. he coordnates of fve APs that deployed n the envronment are (10, 10), (70, 10), (10, 40), (70, 40) and (40, 25) as shown n Fgure 1. he partcles were ntalzed for each AP wth sze of 100. he dstance between partcles and AP s calculated by Equaton (11). D, ( X x ) ( Y y ) AP p AP p AP p p 1,2,...,100 (11) where xp and yp are the poston of partcles at each AP. ntalzed Fgure 2 Flowchart of the proposed method 5.0 RESULS AND DISCUSSION he proposed method has been evaluated to nvestgate mnmzng the dstance error usng four PSO varants that have been mentoned n secton 3. Fgure 1 estbed wth fve deployed APs he number of partcles that used n the smulaton s 100 partcles wth maxmum of 100 teratons for all four PSO varants as shown n able 1. able 1 Specfcaton of PSO parameters Approach Parameter Settng All Number of teratons 100 Runs 20 Swarm sze 100 SPSO c1, c2, w 2,2,0.9 LDIW-PSO c1, c2, w 2,2,0.9-0.4 HPSO-VAC c1, c1 f, c 2,c 2.5,0.5,0.5,2.5 2 f CFPSO c1, c2, k 2.05,2.05,1 Fgure 3 Cumulatve dstrbuton functon of four PSO varants when NPC=30 We study the effects of three dfferent cases of NPC whch are 30, 60, and 100. able 2 shows the statstcs of four PSO varants.
109 Abdulraqeb et al. / Jurnal eknolog (Scences & Engneerng) 78: 9 3 (2016) 105 110 Fgure 3 shows the comparson of PSO varants versus the dstance error when NPC s 30. It can be observed that the SPSO acheved a maxmum and mnmum dstance error of 14.59 meters and 0.30 meters, respectvely. However, the maxmum dstance error of 14.59 s consdered a hgh dstance error that has a negatve effect on the system accuracy. Lkewse, LDIW-PSO acheved slght mprovement compared to SPSO. Unlke CFPSO and HPSO-VAC, the mnmum and maxmum of dstance error s sgnfcantly mproved by 82.14% compared to the prevous two PSO varants. Besdes, CFPSO outperformed HPSO- VAC and mnmzed the maxmum dstance error by 45.26%. Fgure 5 shows the comparson of PSO varants when NPC s 100. he SPSO obtaned the optmum value of the maxmum dstance error (4.60 meters when NPC=100) compared to NPC=30 AND 60. However, LDIW-PSO does not postvely affected by ncreasng the NPC from 60 to 100. he last two PSO varants (CFPSO and HPSO-VAC) obtaned the smallest dstance error compared to the prevous results. Specally, HPSO-VAC consdered the best varant that mnmzed the maxmum dstance error to 0.5 meter. able 2 he dstance error of the four PSO varants when NPC=30, 60 and 100 PSO varant NPC Mn (m) SPSO LDIW-PSO CFPSP HPSO-VAC Max (m) Mean (m) Std. 30 0.30 14.59 4.20 5.35 60 0.23 7.06 1.32 1.94 100 0.11 4.60 1.58 1.71 30 0.97 13.69 3.76 4.40 60 0.15 2.66 0.98 0.84 100 0.21 2.13 0.84 0.59 30 0.09 1.79 0.67 0.53 60 0.06 0.67 0.40 0.22 100 0.17 1.53 0.44 0.38 30 0.09 3.27 0.80 1.08 60 0.07 1.08 0.38 0.32 100 0.025 0.50 0.19 0.15 Fgure 4 Cumulatve dstrbuton functon of four PSO varants when NPC=60 Fgure 4, shows the dstance errors for the four PSO varants when NPC s 60. It can be seen that SPSO mproved ts performance by reducng the mnmum and maxmum dstance error by 23.33% and 51.61%, respectvely. However, LDIW-PSO sgnfcantly mproved wth approxmately 6 tmes better than SPSO. Fnally, CFPSO and HPSO-VAC acheved remarkable mprovements whch the maxmum dstance error closes to 1 meter. herefore, the best PSO varants that have been tested to mnmze the dstance error are CFPSO and HPSO-VAC. Fgure 6 Average error of the four PSO varants for NPS=30, 60 and 100 Fgure 5 Cumulatve dstrbuton functon of four PSO varants when NPC=100 he localzaton accuracy depends on the PSO varants as well as NPC. Fgure 6 and able 2 shows the HPSO-VAC acheved the best average accuracy of 0.19 meter whle the SPSO acheved the poor average accuracy of 1.58 meter when NPC s 100. o conclude, When NPC ncreases the mnmum dstance error decreases. HPSO-VAC acheved the best results n terms of accuracy as well stablty when NPC s 100 followed by CFPSO, LDIW-PSO, and SPSO. he most stable PSO varant s HPSO-VAC snce t has a very low standard devaton.
110 Abdulraqeb et al. / Jurnal eknolog (Scences & Engneerng) 78: 9 3 (2016) 105 110 6.0 CONCLUSION Localzaton systems are stll a consderable ssue n wreless communcaton, especally n ndoor envronments. PSO s a well-known optmzaton algorthm that s used for obtanng good results. In ths paper, we proposed a method that uses PSO to mprove the system accuracy n the ndoor envronments. he dea of our proposed method s to generate N partcles n a crcular dstrbuton where the partcle dstances are the same dstance from the AP to the target. hen, we test each partcle f t satsfes the other dstances (dstances from each AP to the target) and not only the dstance from the selected AP to the target. We select the partcle whch satsfes all dstances as the target. he proposed method has been evaluated n smulaton wth the deployment of fve APs n a dmenson area of 80 50 m 2. Four PSO varants, namely, SPSO, LDIW-PSO, CFPSO and HPSO- VAC are used to observe ther effect on mnmzng the dstance error. Based on the obtaned results, HPSO-VAC s the best PSO varant that mnmzes the dstance error wth average dstance error of 0.19 meters when NPC s 100. Other PSO varants can be used to further mnmze the dstance error snce t s not optmal yet. References [1] Al-Ahmad, A. S. M., Omer, A. I. A. and Kamarudn, M. R., and Rahman,. A. 2010. Mult-floor Indoor Postonng System Usng Bayesan Graphcal Models. Progress In Electromagnetcs Research B. 25(2010): 241-259. [2] Alhammad, A., Alas, M. Y., an, S. W. and Sapumohott, C. 2013. Enhanced Indoor Localzaton System for Mult-Floor Envronment usng Clusterng echnques. 1st Internatonal Conference of Recent rends n Informaton and Communcaton echnologes, Malaysa, 12-15 September 204. 37-47. [3] L. M, Kwok. S. H and Gordon, H. 2010. Accurate Angle-of- Arrval Measurement Usng Partcle Swarm Optmzaton. Wreless Sensor Network. 2(5): 358-364. [4] Rady, S., Wagner,A. and Badreddn, E. 2010. Herarchcal Localzaton Usng Entropy-Based Feature Map And rangulaton echnques. 2010 IEEE Internatonal Conference on Systems Man and Cybernetcs (SMC). 10-13 October 2010. 519-525. [5] Dng, G., Zhang, J., Zhang, L. and an, Z., 2013. Overvew Of Receved Sgnal Strength Based Fngerprntng Localzaton In Indoor Wreless LAN Envronments. 2013 IEEE 5th Internatonal Symposum on Mcrowave, Antenna, Propagaton and EMC echnologes for Wreless Communcatons (MAPE). 29 October 2013. 160-164. [6] Alhammad, A., Fazrulhysam, M. Fadlee and S. Alrah. 2016. Effects of Dfferent ypes Of RSS Data On he System Accuracy Of Indoor Localzaton System. 2016 IEEE Regon 10 Symposum (ENSYMP). 9-11 May 2016. 311-314. [7] Machaj, J. and Brda, P. 2012. Optmzaton Of Rank Based Fngerprntng Localzaton Algorthm. Internatonal x Conference on Indoor Postonng and Indoor Navgaton (IPIN). 13-15 November 2012. 1-7. [8] Kennedy, J. and Eberhart, R. 1995. Partcle Swarm Optmzaton. Proc. IEEE Internatonal Conference on Neural Networks. 27 Nov.-1 Dec. 1995. 1942-1948. [9] Yan, W. and Nan, H. 2014. A Novel Partcle Swarm Optmzaton Based Non-lne of Sght Moble Node Localzaton Algorthm. Journal of Computatonal Informaton Systems. 8759-87661. [10] Zhu, H. Ngah, S., Xu Y., anabe, Y. and Baba,. 2008. A Random me-varyng Partcle Swarm Optmzaton for Local Postonng Systems. Internatonal Journal of Computer Scence and Network Securty. 8)6(: 49-60. [11] sa, P., Ln, Chun., Chen C., and Wang J. 2013. A Scalable Localzaton Scheme usng Partcle Swarm Approach for Sensor Networks. 7 th Internatonal Conference on Sensor echnologes and Applcatons, 25-31 August 2013. 21-26. [12] Leng, M., ay, W. P., Quek,. Q. S. 2012. Cooperatve And Dstrbuted Localzaton For Wreless Sensor Networks In Multpath Envronments. IEEE Internatonal Conference on Acoustcs, Speech and Sgnal Processng (ICASSP). 25-30 March 2012. 3125-3128. [13] Ln,. N., Fang, S. H., seng, W. H., Lee, C. W. and Hseh, J. W. 2014. A Group-Dscrmnaton-Based Access Pont Selecton for WLAN Fngerprntng Localzaton. IEEE ransactons on Vehcular echnology. 63(8): 3967-3976. [14] Chen, Y., Yang, Q., Yn J., and Cha X., 2006. Power-Effcent Access-Pont Selecton For Indoor Locaton Estmaton. IEEE ransactons on Knowledge Data Engngeerng. 18(7): 877-888. [15] Kushk, A., Platanots, K. N., and Venetsanopoulos, A. N., 2007. Kernelbased Postonng In Wreless Local Area Networks. IEEE ransactons Moble Computng. 6(6): 689-705. [16] Veccho, M., López-Valcarce, R. and Marcellon, F. 2012. A wo-objectve Evolutonary Approach Based On opologcal Constrants For Node Localzaton In Wreless Sensor Networks. Appled Soft Computng Journal. 12(7): 1891-1901. [17] Shokran, M. and Hgh, K. A. 2014. Applcaton Of A Mult Objectve Mult-Leader Partcle Swarm Optmzaton Algorthm On NLP And MINLP Problems. Computers and Chemcal Engneerng. 60(2014): 57-75. [18] Zhang, E., Wu, Y., and Chen, Q. 2014. A Practcal Approach For Solvng Mult-Objectve Relablty Redundancy Allocaton Problems Usng Extended Bare-Bones Partcle Swarm Optmzaton. Journal of Relablty Engneerng and System Safety. 127(2014): 65-76. [19] Duan, C., Wang, X., Shu, S., Jng, C., and Chang, H. 2014. hermodynamc Desgn Of Strlng Engne Usng Mult- Objectve Partcle Swarm Optmzaton Algorthm. Energy Converson and Management. 84(204): 88-96. [20] Sun, Z., ao, L., Wang, X., and Zhou, Z. 2014. Localzaton Algorthm n Wreless Sensor Networks Based on Multobjectve Partcle Swarm Optmzaton. Internatonal Journal of Dstrbuted Sensor Networks. 11(8): 1-9. [21] Yuhu, S. and Eberhart, R. C. 1999. Emprcal Study Of Partcle Swarm Optmzaton. Proceedngs of the 1999 Congress on Evolutonary Computaton. 6-9 July 1999. 1950. [22] Ratnaweera, A., Halgamuge, S., and Watson, H. C. 2004. Self-organzng Herarchcal Partcle Swarm Optmzer Wth me-varyng Acceleraton Coeffcents. IEEE ransactons on Evolutonary Computaton. 8(3): 240-255. [23] Clerc, M. and Kennedy, J. 2002. he Partcle Swarm - Exploson, Stablty, And Convergence In A Multdmensonal Complex Space. IEEE ransactons on Evolutonary Computaton. 6(1): 58-73.