A pa icle swa m op imiza ion algo i hm o op imal ca -call alloca ion in ele a o g oup
con ol sys ems
Be na Bola
Yildiz Technical Uni e si y, Facul y o Mechanical Enginee ing,
Mechanical Enginee ing Depa men , Yildiz, TR-34349, Is anbul, Tu key.
Email: [email protected].
Oğuz Al un
Yildiz Technical Uni e si y, Facul y o Compu e Enginee ing,
Compu e Enginee ing Depa men , Yildiz, TR-34349, Is anbul, Tu key.
Email: ogu[email p o ec ed].
Pablo Co és
Uni e si y o Se ille, Escuela Técnica Supe io de Ingenie ía, Ingenie ía O ganización
Camino de los Descub imien os s/n, Se illa 41092, Spain.
Email: [email protected]
Abs ac .- High- ise buildings equi e he ins alla ion o complex ele a o g oup con ol
sys ems (EGCS). In e ical anspo a ion, when a passenge makes a hall call by p essing a
landing call bu on ins alled a he loo and loca ed nea he ca s o he ele a o g oup, he
EGCS mus alloca e one o he ca s o he g oup o he hall call. We de elop a Pa icle Swa m
Op imiza ion (PSO) algo i hm o deal wi h his ca -call alloca ion p oblem. The PSO algo i hm
is compa ed o o he so compu ing echniques such as gene ic algo i hm and abu sea ch
app oaches ha ha e been p o ed as e icien algo i hms o his p oblem. The p oposed PSO
algo i hm was es ed in high- ise buildings om 10 o 24 loo s, and se e al ca con igu a ions
om 2 o 6 ca s. Resul s om ials show ha he p oposed PSO algo i hm esul s in be e
a e age jou ney imes and compu a ional imes compa ed o gene ic and abu sea ch
app oaches.
Keywo ds: Ele a o ; li ; pa icle swa m op imiza ion; ele a o g oup con ol sys em; e ical
anspo a ion
1. In oduc ion
High- ise buildings equi e he ins alla ion o la ge ele a o g oups. The e o e, he e ical
anspo a ion indus y g ows oge he wi h he mo e and mo e inc ease o such high- ise
buildings. Such si ua ions equi e he managemen o mul iple ele a o s in a coo dina ed way in
o de o e icien ly anspo passenge s h oughou he building. This managemen is done by
he Ele a o G oup Con ol Sys em (EGCS).
The EGCS pe o mance depends on he a ic pa e n in he building. I ep esen s he mobili y
o a building popula ion in i s necessi ies o e ical anspo a ion. Each building add esses a
speci ic shape o i s own a ic pa e n. Typically, in a p o essional building he a ic pa e n
will p esen a la ge han a e age numbe o up landing calls a he s a o he day. These a e
due o he building’s wo ke s a i ing o s a wo k. This phase is called uppeak a ic. On he
con a y, la e in day he e is he opposi e phenomenon, and a la ge han a e age numbe o
down landing calls akes place. I co esponds o he building’s popula ion wan ing o go home
a e a wo king day. This a ic pa e n is called downpeak. In he middle o he day he e a e
wo join phenomena, because he appea ance o up and down peaks. I depic s a si ua ion o
people wan ing o lea e he building o lunch and people coming back a e lunch. This pe iod
is called lunchpeak a ic. Finally he es o he day does no show any special endency om
any speci ic loo o om any speci ic s eam. Gene ally, less a ic is egis e ed oo. I is
called in e loo a ic.
I has been p o en (Ba ney e al. 1985; o Cibse Guide D (2000) amongs o he s) ha uppeak
a ic is he mos s essed a ic ha is p oduced in a high- ise building. Thus, acco ding o
ha we unde ake he analysis o ou algo i hm unde such condi ions.
In gene al e ms, he main decision he con olle has o ake is o de e mine which ca om he
EGCS (one speci ic li om he ele a o g oup) mus be assigned o a call. Tha is, once a
passenge wan s o a el om a loo o ano he di e en loo in a building, and he passenge
makes a hall call o an ele a o by p essing a landing call bu on ins alled a he loo and
loca ed nea he ca s o he ele a o g oup, he EGCS mus iden i y he ele a o in he g oup
ha is mos sui able o se e he passenge . Thus, he p oblem o be sol ed is o selec an
ele a o o each hall call ha is issued. This p oblem is called as he ca -call alloca ion
p oblem.
The op imiza ion o such p oblem is mainly based on he minimiza ion o he a e age jou ney
ime (AJT), which is he ime in seconds ha a passenge spends a elling o a des ina ion loo
measu ed om he ins an o call egis a ion o he ins an passenge s eps on o he des ina ion
loo . AJT ime consis s o he a e age a el ime (ATT), which consis s o he ca a el ime
om he o igin loo o he des ina ion plus he a e age wai ing ime (AWT), which consis s o
he ime in seconds ha a passenge wai s o se ice measu ed om he ins an a passenge
egis e s a call o he ins an he passenge en e s an ele a o ca s. See Ba ney e al. (1985) and
he Cibse Guide D on T anspo a ion sys ems in buildings (2000), which p obably cons i u e
some o he mos ecognized handbooks in e ical anspo a ion. These pa ame e s a e also
discussed in he simula ion sui e ha is p oposed in Co és e al. (2006). Recen ly, some au ho s
(Tyni e al. 2006 o Hasan e al. 2012) a e conside ing he ene gy consump ion o he sys em as
objec i e unc ion o low a ic si ua ions and basing i s analysis on he use o he ene gy
gene a ion by he coun e weigh . I is called he ene gy p oblem o he e ical anspo a ion
sys em.
In his pape , we ocused on he ca -call alloca ion p oblem is NP-Ha d independen ly o he
c i e ion. Thus, mos o he app oaches a e ocused on so compu ing echniques. In ac ,
e ical anspo a ion has become a majo ield o applica ion o so compu ing app oaches
such as uzzy logic, neu al ne wo ks, gene ic algo i hms, e c. (see sec ion 2). All o hem a e
echniques capable o p o iding be e solu ions han adi ional con olle s implemen ing
dispa ch expe ules ha make use o simple IF-ELSE logical command se s.
The es o he pape ollows wi h he p esen a ion is o ganized as ollows: sec ion 2 desc ibes
ela ed wo k unde aken o sol e he ca -call alloca ion p oblem in ele a o g oup con ol
sys ems; he PSO algo i hm ha we implemen ed is desc ibed in sec ion 3; he compu e
simula ions a ending o he a e age jou ney ime and compu a ional ime o PSO algo i hm
and i s compa ison wi h gene ic algo i hm and abu sea ch app oaches a e deal in sec ion 3;
and inally main conclusions a e discussed in sec ion 4.
2. Rela ed wo k
As i has been p e iously add essed in sec ion 1, he implemen a ion o e icien con ol
sys ems in ele a o g oups is absolu ely equi ed o all buildings. Such ype o sys ems a e
called Ele a o G oup Con ol Sys ems (EGCS), and al hough his is a young ield o esea ch
(acco dingly wi h he ecen e olu ion o he elec onic) i can be ound ele an e e ences in
he scien i ic li e a u e.
Mos o mode n EGCSs implemen complex algo i hms based on di e en so compu ing
echniques. Gene ic algo i hms appea ed as one o he i s a emp s o ackle wi h such
p oblem, and since hen ha e been widely used p o iding good and aluable esul s. Examples
o ha a e Co és e al. (2003) and Co és e al. (2004), whe e he au ho s implemen ed a bina y
gene ic encoding o easible solu ions ha has been widely ollowed by se e al au ho s since
hen. The algo i hm was compa ed o adi ional indus y collec i e con olle s ( ha a e no
based on so compu ing app oaches) p o iding signi ican imp o emen s. The compa ison was
unde aken using simula ion ARENA so wa e. The main p oblem o hese app oaches was he
equi ed compu a ion ime equi ed o he mic ochips o he con olle s. A e ha , gene ic
app oaches con inued being applied. I was he case o Tyni e al. (2006) ha de eloped a bi-
objec i e gene ic algo i hm ha op imized he wai ing imes and he ene gy consump ion o
he KONE Co po a ion. La e , Hi asawa e al. (2008) applied ano he gene ic app oach o a
speci ic ype o ele a o s ha a e called double-deck whe e wo cages a e connec ed in a sha
ha e been de eloped o he ising demand o mo e e icien anspo o passenge s in high- ise
buildings. He e, he au ho s de elop a g aph-based e olu iona y me hod named gene ic ne wo k
p og amming ha in oduce a ious node unc ions ha can be easily execu ed by an e icien
ule-based g oup supe iso y con ol ha is op imized in an e olu iona y way. Mo e ecen ly,
Bola e al. (2010) de eloped a gene ic algo i hm based on he p e ious wo k o Co és e al.
(2004) p o iding a no el way o compu e he a e age wai ing ime o passenge s ha allowed a
compu a ionally e ec i e e alua ion, and o e coming ha limi a ion.
Tabu sea ch has a ac ed less a en ion han gene ic algo i hms, al hough ecen ly wo
algo i hms based on de e minis ic and p obabilis ic app oaches ha e been p esen ed o deal
wi h he p oblem. Bola e al., (2011) ha e p o ided a de e minis ic and p obabilis ic app oach
o he abu sea ch algo i hm ha allows ou pe o m he esul s p o ided by he equi alen
gene ic implemen a ion.
Li e al. (2007) has ied immune sys ems based on a wo-le el con ol s uc u e. One s uc u e
is he locally op imal assignmen o a hall call pe o med by a con en ional collec i e
algo i hm; he o he is he globally op imal assignmen o all hall calls, which is execu ed
pe iodically by a i icial immune algo i hm. This ep esen s one o he e y ew app oaches
based on immune sys ems o deal wi h e ical anspo a ion p oblems.
Con ol me hodologies ha e been applied o EGCS oo. I is he case o neu al ne wo ks ha
we e e y en husias ically ied in he o igins o he discipline (as i was he case o Im ak e al.,
2001). Mo e ecen ly, he same au ho (Im ak, 2008) has p esen ed an e olu ion o his i s
app oach and has compa ed i wi h indus y con en ional app oaches. The pape shows how he
EGCS can p edic he nex s opping loo s o s op by conside ing wha has been lea n by
p ocessing he changes in passenge se ice demand pa e n. Echa a ia e al. (2009) ha e
de eloped a eed- o wa d neu al ne wo k based con ol algo i hm has been de eloped ha can
app oxima e ele a o call pa e ns by lea ning o associa e ime o day wi h speci ic call
loca ions. A b ie compa ison o he me hod allows an icipa ing sui abili y when compa ing o
uzzy app oaches, al hough u u e esea ch is equi ed o gua an ee such claim. Mo e ecen ly,
Du sun (2010) has implemen ed a neu al ne wo k o con ol he EGCS o cases o ex e nal
bu ons. I co esponds o he case o e ical anspo a ion zoning app oaches. The app oach
p o ided ad an ages o con en ional s a ic zoning app oaches.
Al hough uzzy logic was also conside ed since a long ime ago (see Kim e al., 1998 o one o
he i s a emp s o deal wi h uzzy con olle s), nowadays, he ex ensi e use o uzzy logic is
being implemen ed in an en husias ic way. I is he case o Jamaludin e al. (2010) ha ha e
p esen ed a uzzy con olle ha ins ead o depending hea ily on he p edic ed passenge a ic
pa e n o adap a ion, he uzzy logic g oup con olle adjus s i sel o sui he sys em’s
en i onmen h ough a sel - uning scheme. Resul s a e simula ed and compa ed o con en ional
app oaches showing a signi ican imp o emen . La e , Co es e al. (2011) ha e de eloped a
uzzy con olle o o ecas he a ic pa e ns in he e ical anspo a ion sys em. The
con olle allows o iden i y whe he sys ems is unde an uppeak, downpeak, lunchpeak o
in e loo pa e n. Recen ly, Chen e al. (2012) ha e p esen ed a uzzy logic app oach o con ol
he EGCS sel - uned by a gene ic algo i hm o maximize se ice quali y and managing a wide
se o a iables such as he numbe o ele a o s, a ic low, di ec ion, conges ion, p io i y o
loo , and p e e ence o passenge s, amongs o he s.
He e we p esen a pa icle swa m op imiza ion (PSO) algo i hm based on a hall call alloca ion
s a egy o de ine he solu ion encoding and compu a ionally e ec i e ca -call alloca ion quali y
es ima ion. I cons i u es a no el applica ion o PSO o a new ele an indus y sec o .
App oaches based on PSO a e sca ce in he li e a u e. Li (2010) PSO algo i hm was an
excep ion. Li p esen ed a PSO algo i hm ha is used o op imize ypical zoning ele a o
p oblems whe e he loo sec o ha is going o be se ed by each ele a o ca has o be s a ed.
The app oach shows good esul s when compa ing o o he echniques.
Zoning (o sec o ing) app oach o e ical anspo a ion is a echnique used o se e
skysc ape s du ing he uppeak a ic. I di ides he building in o di e en sec o s and assign
one (o a g oup o ) ca o each sec o . When he ca lea es he passenge s in he loo s assigned
o i s zone, he ca go down o collec mo e passenge s a elling o he loo s o he zone. This
echnique is only applied o uppeak condi ions because i wo sens he EGCS pe o mance
du ing o he pa e ns (downpeak, lunchpeak o in e loo ). I also equi es he ins alla ion o
ex e nal bu on box wi h all he loo des ina ion numbe in he hall o he ca . Thus, i ela es o
o he e ical anspo a ion philosophy o managemen . This is called he zoning p oblem a
di e ence om he dispa ching p oblem (o ca -call alloca ion p oblem). When using zoning
app oaches, one o he main pa ame e s o be op imised is he ound ip ime (RTT), which
measu es he ime equi ed o ake a passenge om he g ound loo , go up o he highes loo
and come back o he g ound loo . Hence, RTT o mulas can only be applied o up a ic
whe e passenge s en e o he lobby and a e des ined o uppe loo s.
The e o e, due o hese easons, we canno compa e ou algo i hm ha is con igu ed o a
adi ional bu on box wi h only up and down bu ons and ha do no conside zoning app oach
o dispa ching ca s, bu a global EGCS whe e all he ca s se e all he loo s o he building.
Howe e , he good esul s p o ided by PSO o he zoning p oblem, lead us o y wi h he
app oach o he ca -call alloca ion p oblem p o iding ou s anding esul s as i is shown in he
esul sec ion. To ha e benchma ks capable o assessing ou p oposal, we compa e i agains
iden ical objec i e unc ion implemen a ions and he same building con igu a ion o gene ic
algo i hms and abu sea ch app oaches. In his line, esul s a e p o ided o high- ise buildings
om 10 o 24 loo s, and se e al ca con igu a ions om 2 o 6 ca s.
3. The pa icle swa m op imiza ion algo i hm o EGCS
In eal buildings, passenge s a i e andomly o di e en loo s, e en a he same ime, wishing
o be anspo ed om a loo o o he one. In addi ion, buildings show speci ic mo emen o
passenge s ha can de e mine he low pa e n in he building. Fou main pa e ns a e
adi ionally ca alogued: (i) uppeak a ic when a la ge han a e age numbe o up landing
calls a e p oduced ( ypically because he building’s wo ke s a i ing o s a wo k); downpeak
a ic when a la ge han a e age numbe o down landing calls akes place (because building’s
popula ion wishes o go back home a e he wo king day); (iii) lunchpeak a ic ha akes
place in he middle o he day, and i is due o he appea ance o up and down peaks; and inally
(i ) in e loo a ic co esponding o he es o he day. The la e phenomenon can be
cha ac e ised o a low demand (usually a ound 4% o he popula ion) in bo h di ec ions.
The ele a o g oup con ol sys ems de e mine which ca o he g oup should se e a hall call. I
he hall call is alloca ed o he mos app op ia e ca , he passenge s’ a el and wai ing imes a e
educed. We de elop he e a PSO algo i hm ha ou pe o ms o he so compu ing
implemen a ions such as gene ic algo i hm o abu sea ch. Nex we de elop he solu ion
encoding o he cha ac e iza ion o p oposed solu ions ha is desc ibed in nex subsec ion 3.1;
he way o assess he ca -call alloca ion, i.e., he e alua ion o he candida e solu ions (wha is
called as i ness); and he de ailed desc ip ion o he algo i hm and i s lowcha .
3.1. Ca -call alloca ion solu ion encoding
Hall calls a e encoded using a 2×(Numbe o loo s – 1) elemen s a ay. The i s hal o he
a ay co esponds o upwa ds landing calls, and he second hal co esponds o downwa ds
landing calls. A speci ic ca om he ele a o g oup mus be alloca ed o each eques ed hall
call, and a solu ion o he ca -call alloca ion p oblem consis s o he alloca ion o a speci ic ca
o he ele a o g oup o all he hall calls being eques ed in he building. The e o e, he
dimension o he solu ion becomes equal o numbe o eques s in he hall call a ay (‘ones’ in
Figu e 1.a), as no ele a o needs o be assigned o loo s wi hou eques s. So, he dimension o
he solu ion encoding can change acco ding o he di e en hall call eques con igu a ions.
Figu e 1 (a) and (b) p o ides an example o ca -call alloca ion solu ion encoding o a building
wi h 10 loo s and 3 ele a o s. Figu e 1 (a) ela es o he hall call eques ed a he loo s. I is
ep esen ed by an a ay o 18 elemen s, 9 o upwa ds landing calls, and 9 o downwa ds
landing calls. A numbe equal o ‘1’ means ha he e is a call, and a ‘0’ means ha he e a e no
calls a ha loo . Figu e 1 (b) ela es o he solu ion encoding o he ca -call alloca ion
p oblem. In he example, Figu e 1 (a) shows ha he equi ed solu ion encoding will ha e a
dimension equal o 13; such dimension co esponds o he eques ed hall calls. A solu ion
consis s o he alloca ion o a ca o he ele a o g oup (composed o h ee ca s) o a hall call.
The solu ion encoding in he example shows he numbe o he ca ha is assigned. Hence, each
elemen can ge a alue be ween 1 and 3 since we ha e 3 ele a o s.

Upwa ds Landing Calls Downwa ds Landing Calls
F1 F2 F3 F4 F5 F6 F7 F8 F9 F2 F3 F4 F5 F6 F7 F8 F9 F10
1 0 1 1 0 1 1 1 0 1 1 1 1 0 1 1 1 0
(a)
F1 F3 F4 F6 F7 F8 F2 F3 F4 F5 F7 F8 F9
3 1 2 3 3 1 2 1 2 3 2 1 3
(b)
Figu e 1. (a) An example o hall call encoding o a building o 10 loo s. (b) Co esponding
solu ion encoding o an ele a o g oup wi h 3 ca s.
3.2. Ca -call alloca ion i ness e alua ion
To e alua e he quali y o a ca -call alloca ion by he EGCS is, we e alua e he i ness o such
alloca ion. As any possible alloca ion mus be e alua ed, compu a ionally e ec i e is o
eno mous ele ance. The e o e, he me hod o e alua e he i ness unc ion is a e y impo an
issue. In addi ion, he i ness has o p o ide an adequa e alue o he pe o mance o he ca -call
alloca ion. One o he mos ime-implemen ed dispa che s in he ele a o indus y is he THV
one (see Co és e al., 2003). THV algo i hm was implemen ed a UMIST (Uni e si y o
Manches e Ins i u e o Science and Technology), and assigns he hall call o he nea es li in
he adequa e ip di ec ion (some imes appea s e e ed as nea es call algo i hm). I is easy o
be implemen ed bu does no p o ide high quali y es ima ions o he p oposed alloca ion.
Ano he common implemen a ion is due o he es ima ed ime o a i al (e en a he same ime,
wan ing) algo i hm, which unde akes an es ima ion o he equi ed ime since he landing call
is issued un il he ca a i es (Ba ney e al., 1985). ETA includes di e en le els o p io i y: (i)
long wai ing calls; (ii) high ac i i y loo s; (iii) p io i y le els; and (i ) emaining calls. Those
ca s a ending calls wi h p io i y le el one o h ee do no s op a landing calls and only se e
ca calls equi ing a special AJT calcula ion. ETA algo i hm p o ides a sui able beha io du ing
uppeak a ic, a medium le el se ice o in e loo , and a bad le el o se ice du ing
lunchpeak and specially downpeak. These app oaches we e es ed in Co es e al. 2004, Bola
e al. 2010 and i was app ecia ed ha he Bola e al. (2011) i ness e alua ion p oposal
ou pe o med he o he p e iously desc ibed app oaches bo h in quali y o solu ions and
compu a ional speed. So hen, we op ed o ollow he me hodology desc ibed in Bola e al.
(2011) as an easy- o-implemen and as - o-compu e echnique, p o iding he be e index o
pe o mance o he sys em in a sui able ime o esponse. I is desc ibed nex .
Gi en he ollowing pa ame e s:
- Ψ
1
: g ound loo le el
- Ψ
2
: highes down hall call le el
- Ψ
3
: numbe o down hall calls be ween Ψ
1
and Ψ
2
.
- Ψ
4
: highes up hall call le el
- Ψ
5
: numbe o up hall calls be ween Ψ
1
and Ψ
4
.
- Ψ
6
: lowes down hall call le el
- : doo opening and closing ime
-
p
: passenge ans e ime
- Hc : Highes ca ip ime
- Lc : lowes ca ip ime
The solu ion i ness, , is calcula ed depending on he ype o passenge s’ mo emen s. So, a
i ness alue is i s ly calcula ed o each ca , i, in he g oup by conside ing ou di e en cases
ha a e shown in igu e 2. No e ha each o mula ela es in which loo he passenge is aken
and in o which loo he passenge is ans e ed. Hence, as a gene al de ini ion o Ψ, i shows
he loo s whe e passenge s a e ge ing on and o .
CASE 2
Only up hall calls
uppeak a ic pa e n
CASE 3
Only down hall calls
downpeak a ic pa e n
CASE 4
Down and up hall calls
lunchpeak and in e loo
a ic pa e ns
(
)
( )
2 1
3 1
Ψ − Ψ 
=
 
+ Ψ − Ψ
 
 
i
p
(
)
( )
4 1
5 1
Ψ − Ψ 
=
 
+ Ψ − Ψ
 
 
i
p
(
)
(
)
( ) ( )
( )
4 1 2 4
2 6 3 5 1
 
Ψ − Ψ + Ψ − Ψ +
 
=
 
Ψ − Ψ + Ψ + Ψ − Ψ
 
i
p
0
=
i
CASE 1
No hall calls
Figu e 1. Fi ness es ima ion depending on he a ic pa e n
Finally he g oup i ness is calcula ed, see equa ion (1), and he inal i ness is e alua ed o he
p oposed alloca ion, see equa ion (2). Tha is, he o al i ness o he sys em is calcula ed as a
wo-pa unc ion whe e he i s pa collec s he conce n ela ed o passenge s wai ing in he
halls, and he second pa conce n ela ed o an unbalanced pe o mance in he ca s o he
g oup. Tha is, he second pa ies o le el he use o each ca . Le ’s conside he ollowing
example: ca no.1 is wo king du ing 100 seconds and ca no.2 du ing 11 seconds, so bo h ca s
would be wo king du ing 111 seconds bu in a much unle elled manne . On he o he hand i
ca no.1 wo ks du ing 57 seconds and ca no.2 du ing 54 seconds, he wo ca s wo k du ing 111
seconds bu in a much mo e le elled manne . This ac ion will ha e epe cussion on he li e
cycle o he ca s. Al hough k
1
and k
2
a e design pa ame e s, and a ia ions and a disc e ional
c i e ion can be admi ed, we selec hem equal o 1.5 and 2 acco dingly wi h Bola e al.
(2011).
1
being he numbe o ca s in he g oup
n
i
i
g oup
, n
n
=
=
∑
(1)
(
)
1 2g oup
k · k · Hc Lc
= + −
(2)
3.3. Pa icle swa m op imiza ion algo i hm
Pa icle swa m op imiza ion, p esen ed in Kennedy e al. (1995), is a popula ion based
s ochas ic op imiza ion echnique inspi ed by social beha iou o bi d locking o ish
schooling.
PSO is a popula ion based echnique. The sys em is ini ialized wi h a popula ion o pa icles
ha co espond o ini ial solu ions o he p oblem. These pa icles mo e in he sea ch space
sea ching o op imal i ness alue.
Following Coelho (2010), PSO algo i hm de ines pa icle as a po en ial solu ion ep esen ed by
an s-dimensional ec o , whe e s is he numbe o op imiza ion a iables. The swa m concep
ela es o an appa en ly diso ganized popula ion o mo ing pa icles ha ends o clus e
oge he while each pa icle seems o be mo ing in a andom di ec ion. The pa icle bes
posi ion is calcula ed o a pa icle mo ing h ough he sea ch space, i compa es he i ness
alue a he cu en posi ion o he bes i ness alue i has e e a ained a any ime up o
cu en ime. Then, he global bes ela es o he bes posi ion among all indi idual bes
posi ion; and he eloci y o he pa icle ligh ep esen s he eloci y o he pa icle in he
physical analogy, aking in o ha all he pa icles eloci ies and posi ions a e upda ed a e each
i e a ion.
We implemen ed a PSO algo i hm ha is desc ibed by he lowcha in Figu e 3. The s eps o
he lowcha a e numbe ed o easy e e ing. S eps 1 o 4 ini ialize a iables ha change in he
main loop in a sui able way be o e en e ing in o he main loop (s ep 5). In s ep 1, each pa icle
is “ h own” o some andom place in he sea ch space: each pa icle is gi en a andom posi ion.
Then, each pa icle ge s a andom eloci y (s ep 2). Since a pa icle has a posi ion and a gi en
eloci y, i s nex posi ion can be calcula ed in he i s i e a ion o he main loop ollowing he
ules and calcula ions de ailed in s eps 8 and 9 la e . In s ep 3, he i ness alue o each
pa icle’s ini ial posi ion is calcula ed. The bes o hese ini ial posi ions is assigned o he
a iable “global bes ” (s ep 4). S eps 5 o 14 cons i u e he main loop o he algo i hm. In s ep 5,
i is decided whe he ano he i e a ion o he loop is unde aken o no by checking he numbe
o i e a ions. Once he maximum numbe o i e a ions is eached, he p ocedu e ge s ou o he
loop and p oceeds o s ep 15. S ep 6 makes su e ha all he pa icles a e p ocessed. In s ep 7,
we ge he nex unp ocessed pa icle, and p oceed wi h i .
Le us call his pa icle he “cu en ” pa icle. In s ep 8, a new eloci y is se o he cu en
pa icle using nex equa ion (3):
g wl w w
jj
∆
+
∆
+
=
+231211
,
(3)
whe e
1+j
is he new eloci y ec o ,
j
is he cu en eloci y ec o ,
1
w
,
2
w
, and
3
w
a e
cons an scala s (weigh s),
1
and
2
a e ec o s whose elemen s a e andom alues be ween
0and
1
, l
∆
is he di e ence ec o be ween he bes and cu en posi ion o he pa icle, and
g
∆
is he di e ence ec o be ween he global bes posi ion and he pa icles cu en posi ion.
This equa ion explains ha he new eloci y o he pa icle is calcula ed using h ee alues: he
cu en eloci y, he dis ance o he pa icle o i s bes posi ion, and i s dis ance o he global bes
posi ion. Weigh s
1
w
,
2
w
, and
3
w
allow us o se ela i e impo ance o he h ee e ms in
de e mining new eloci y. A e se e al es s, we concluded ha no pa o equa ion (3) should
be p io i ized o ano he in o de o ge he bes pe o mance o he algo i hm. So hen, alues
we e se o 0.34, 0.33, and 0.33 o make hem oughly equal, and o make hem sum up o 1.
In s ep 9, he cu en posi ion oge he wi h he new eloci y de e mines he new posi ion (each
i e a ion was assumed as a uni ime). In some occasions, he new calcula ed posi ion could no
be alid, e.g. hey a e ou o he bounds o he sea ch space. This ac is checked in s ep 10, and
i he posi ion is no alid a andom alid posi ion is assigned o he pa icle. And he new
eloci y alue is e-upda ed acco ding o his new posi ion (s ep 11). S eps 12 o 14 a e o
bookkeeping and p epa a ion o he nex i e a ion. S ep 12 calcula es he i ness alue o he
new posi ion. I necessa y, he pa icle’s bes posi ion and global bes posi ion a e also upda ed
(s eps 13 and 14). When he maximum numbe o i e a ions is eached, he p ocedu e ge s ou o
he main loop and mo es o nex s ep 15. The global bes e ains he esul o he op imiza ion.
A e es ing se e al alues, we ealized ha a maximum numbe o i e a ions equal o 30 we e
enough o gua an ee con e gence wi hou penalising compu a ional imes. In a simila line, a
numbe o 30 pa icles gua an eed a sui able mapping o he al e na i e solu ions. Inc easing he
i e a ions and he numbe o pa icles did no epo signi ican imp o emen s meanwhile he
equi ed ime o un he algo i hm inc eased in alues ou o eal applicabili y in he ele a o
indus y.
5. Conclusion
We ha e p esen ed a no el applica ion o PSO algo i hm o op imize he ca -call alloca ion
s a egy o he con olle in Ele a o G oup Con ol Sys ems. PSO algo i hms had been ied
in e ical anspo a ion p oblems o deal wi h sec o ing p oblems, which is a echnique used
o se e skysc ape s du ing he uppeak a ic di iding he building in o di e en sec o s and
assigning one (o a g oup o ) ca o each sec o . Ou PSO implemen a ion o sol e he ca -call
alloca ion p oblem, also known as he dispa ching p oblem, cons i u es a no el y in he
ele a o scien i ic li e a u e.
Ou implemen a ion is based on a hall call alloca ion s a egy o de ine he solu ion encoding
and includes compu a ionally e ec i e ca -call alloca ion quali y es ima ion. Resul s we e
p o ided o high- ise buildings om 10 o 24 loo s, and se e al ca con igu a ions om 2 o
6 ca s. The algo i hm was success ully applied o all he case s udies ou pe o ming o he so
compu ing echniques such as gene ic algo i hms and abu sea ch. Resul s we e be e
a ending o he a e age jou ney ime ( ha includes he wai ing plus a el imes) as well as
a ending o he algo i hm compu a ional imes. E en mo e he obus ness o he me hod
(measu ed as he s anda d de ia ion) was also p o en. Hence, i was shown ha PSO go
be e esul s in a as e , cheape way compa ed wi h he o he me hods. Ano he eason ha
made PSO mo e a ac i e wi h espec o o he echniques is i s capabili y o be adjus ed wi h
e y ew pa ame e s, which adds addi ional obus ness.
Howe e , esul s ob ained wi h a CPU can be limi ed wi h espec he eal indus y ope a ion.
In p ac ice, eal implemen a ion should be ins alled in speci ic mic ochips, which would
equi e ligh e implemen a ions. Bounding his ac , he eal implemen a ion o he PSO
algo i hm in he indus y appea s o be possible. Fo example, in eal cases an al e na i e can
be calcula ing he i ness no e e y ime bu in a selec i e manne , o s opping he algo i hm
a e a lowe numbe o i e a ions. O cou se all hese decisions a e e y dependen on he
compu a ion speed o he elec onic mic ochips ins alled by he company in he con olle ,
and could a ec he quali y o he implemen ed algo i hm.
Cu en ly, ou u he esea ch ocuses on global con olle s capable o iden i ying a ic
pa e n aking pa in he building and launching he co esponding specialised algo i hm.
This global con olle in ends o inco po a e he conside a ion o ene gy op imiza ion o low
demand pa e ns such as in e loo a ic pa e n ha can p oduce a signi ican educ ion o
ene gy consump ion wi hou wo sening e y much he quali y o se ice indexes.
Acknowledgemen s
The Spanish au ho acknowledges he inancial suppo gi en by he Counselling o
Inno a ion, Science and Business o Andalusia, h ough i s Excellence P ojec s P og amme
(p ojec e . P07-TEP-02832), and he Spanish Resea ch Agency dependen on he
Depa men o Science and Inno a ion, h ough i s DPI P og amme (p ojec e . DPI2010-
15352).

Re e ences
[1] G. Ba ney, S. Dos San os (1985)
Ele a o a ic analysis design and con ol
, Pe e
Pe eg inus L d, Lond es.
[2] B. Bola . P. Co és, E. Yalcin, M. Alis e isci (2010) Op imal ca dispa ching o ele a o
g oups using gene ic algo i hms,
In e na ional Jou nal o In elligen Au oma ion and So
Compu ing
, 16(1), pp. 89-99.
[3] B. Bola , P. Co és (2011) Gene ic and abu sea ch app oaches o op imizing he hall call-
ca alloca ion p oblem in ele a o g oup sys em,
Applied So Compu ing
, 11(2), pp.1792-
1800.
[4] Cha e ed Ins i u ion o Building Se ices Enginee s. CIBSE Guide D (2010)
T anspo a ion sys ems in buildings
, London.
[5] T-C. Chen, Y-Y. Hsu, Y-J. Huang (2012). Op imizing he In elligen Ele a o G oup
Con ol Sys em by Using Gene ic Algo i hm.
Ad anced Science Le e s
9(1), pp. 957-962
[6] L.d.S. Coelho (2010) Gaussian quan um-beha ed pa icle swa m op imiza ion
app oaches o cons ained enginee ing design p oblems,
Expe Sys ems wi h Applica ions
,
37(2), pp.1676-1683.
[7] P. Co es, J. La añe a, L. Onie a, (2003) A gene ic algo i hm o con olling ele a o
g oup sys ems. In:
A i icial Neu al Ne s P oblem Sol ing Me hods
, Pa II Vol. 2687, pp,
313-320.
[8] P. Co és, J. La añe a, L. Onie a (2004) Gene ic algo i hm o con olle s in ele a o
g oups: analysis and simula ion du ing luncpeak a ic,
Applied So Compu ing
, 4(2),
pp.159-174.
[9] P. Co és, J. Muñuzu i, L.Onie a (2006) Design and analysis o a ool o planning and
simula ing dynamic e ical anspo ,
Simula ion: T ansac ions o he Socie y o Modelling
and Simula ion In e na ional
82(4), pp. 255-274.
[10] P. Co és, J. Fe nández, J. Guadix and J. Muñuzu i (2012) Fuzzy logic based con olle
o peak a ic de ec ion in ele a o sys ems,
Jou nal on Compu a ional and Theo e ical
Nanoscience
9 (2), pp. 310-318.
[11] M. Du sun (2010). Es ima ion o passenge wai ing ime in ele a o sys ems wi h
a i icial neu al ne wo k.
In elligen Au oma ion and So Compu ing
16(1), pp. 101-110
[12] J. Echa a ia and C. M. F enz (2009). Imp o ing Ele a o Call Time Responsi eness ia
an A i icial Neu al Ne wo k Con ol Mechanism.
Sys ems, Applica ions and Technology
Con e ence
, 2009. LISAT '09. IEEE Long Island, pp. 1-3.
[13] M.Z. Hasan, R. Fink, M.R. Suyambu, M.K. Baska an (2012). Assessmen and
imp o emen o in elligen con olle s o ele a o ene gy e iciency.
IEEE In e na ional
Con e ence on Elec o/In o ma ion Technology
, EIT 2012; Indianapolis, (Code 91401).
[14] K. Hi asawa, T. Eguchi, J. Zhou, L. Yu, J. Hu, S. Ma kon (2008) A double-deck ele a o
g oup supe iso y con ol sys em using gene ic ne wo k p og amming, IEEE T ans. Sys .,
Man, Cybe n. C, Appl. Re ., 38(4), pp. 535–550.
[15] C.E. Im ak (2008) A i icial neu al ne wo ks applica ion in duplex/ iplex ele a o g oup
con ol sys em. Jou nal o Mechanical Enginee ing 54 (2), pp. 103-114.
[16] C.E. İm ak and G.C. Ba ney (2001) The Applica ion o neu al ne wo ks o li a ic
con ol, Ele a o Wo ld, 49(5), p. 82.
[17] J. Jamaludin, N.A. Rahim, W.P. Hew (2010) An ele a o g oup con ol sys em wi h a
sel - uning uzzy logic g oup con olle , IEEE T ansac ions on Indus ial Elec onics 57 (12),
pp. 4188-4198
[18] J. Kennedy, R.C. Ebe ha (1995) Pa icle swa m op imiza ion. In: P oc. IEEE In 'l.
Con . on Neu al Ne wo ks, Vol. IV, 1942–1948. Pisca away, NJ: IEEE Se ice Cen e .
[19] C. B. Kim, K. A. Seong, H. L. Kwang, and J. O. Kim (1998) Design and implemen a ion
o a uzzy ele a o g oup con ol sys em, IEEE T ans. Sys ., Man, Cybe n. A, Sys ., Humans,
28(3), pp. 277–287.
[20] Z. Li, (2010) PSO-Based eal ime scheduling o ele a o g oup supe iso y con ol
sys em, In elligen Au oma ion and So Compu ing, 16(1), pp.111-121.
[21] Z. Li,, H-Z.Tan, Y-N. Zhang, Z-Y. Mao (2007). Dynamic op imiza ion o ele a o g oup
con ol based on a i icial immune algo i hm o in e - loo peak a ic du ing lunch- ime,
Con ol Theo y and Applica ions 24(2), pp. 177-182.
[22] T. Tyni, J. Ylinen (2006) E olu iona y bi-objec i e op imisa ion in he ele a o ca
ou ing p oblem. Eu opean Jou nal o Ope a ional Resea ch 169, pp. 960–977