GENETIC ALGORITHM FOR CONTROLLERS IN ELEVATOR GROUPS:
ANALYSIS AND SIMULATION DURING LUNCHPEAK TRAFFIC
P. Co és1†, J. La añe a1 and L.Onie a1
1 Se ille Uni e si y
Ingenie ía O ganización.
Escuela Supe io Ingenie os, Camino de los Descub imien os s/n.
Se illa 41092. SPAIN
Tel. +34 95 448 72 05
Fax +34 95 448 73 29
† E-mail: [email p o ec ed]
† URL: h p://io.us.es/P.Co es/main.h m
Abs ac .- A gene ic algo i hm (GAHCA) is p oposed o con ol ele a o g oups o
p o essional buildings. The gene ic algo i hm is compa ed wi h he uni e sal con olle
algo i hm in indus y applica ions. In o he o do so an ARENA simula ion scena io has
been gene a ed du ing hea y lunchpeak a ic condi ions. The esul s allow us o a i m
ha ou gene ic algo i hm eaches a be e pe o mance a ending o he sys em wai ing
imes han adi ional duplex algo i hms.
Keywo ds.- e ical a ic, gene ic algo i hm, ele a o , con olle , simula ion,
lunchpeak.
1. INTRODUCTION
The p og essi e p ice inc ease in he u ban cen es o he la ge ci ies makes he
necessa y in ensi e g ound exploi a ion by means o he cons uc ion o high buildings.
Today he ins alla ion o synch onized ele a o g oups in p o essional use buildings
(o ices, hospi als o ho els) is an usual p ac ice.
The ele a o sys em esea ch is qui e ecen and has ollowed he echnology
de elopmen . The la e eigh ies and he nine ies decade can be conside ed as he s a
poin o he indus ial in es iga ion, especially in USA and Japan ([1], [2] and [3]).
A e ha he esea ch expe imen ed he impulse o he la ges mul ina ional companies
([4], [5], [6] and [7]). By he end o he nine ies he esea ch in e ical anspo a ion
was a eali y and he collabo a ions among he p i a e companies and he esea ch
cen es we e ein o ced, some examples a e he Sys ems Analysis Labo a o y in he
Helsinki Uni e si y o Technology wi h he KONE Co po a ion [8], he Kon ad-Zuse-
Zen um ü In o ma ions echnik o Be lin [9] o he Se ille Uni e si y wi h MAC
PUAR, S.A. [10].
In ele a o sys ems he use o he sys em wai ing ime is he p io i y objec i e o a ain
an e icien sys em pe o mance, a he same ime as ha ing a bounded maximum
wai ing ime. The sys em wai ing ime includes he wai ing ime o he li in he hall
plus he ip ime inside he li . Also, o he seconda y c i e ia a e used as he queue
sizes o he sys em ene ge ic consump ion.
The mo e gene al p oblem assumes he ollowing hypo hesis in he ele a o sys em
pe o mance. Each hall call is a ended by only one cabin. The maximum numbe o
passenge s being anspo ed in he cabin is bounded by i s capaci y. The li s can s op
a a loo only i i exis s a hall call o a cabin call in ha loo . The cabin calls a e
sequen ially se ed in acco dance wi h he li ip di ec ion. A li ca ying passenge s
canno change he ip di ec ion.
Usually, he con olle implemen s dispa ch ules ha make use o an IF-ELSE logical
commands se . Among hese dispa ch ules, a simple li g oup supe iso y con ol
sys em, sui able o g oups o wo o h ee in no e y high ise buildings, is simula ed
in he compu e -aided design sui e LSD (Li Simula ion and Design), implemen ed a
UMIST (Uni e si y o Manches e Ins i u e o Science and Technology), unde he
designa ion o he THV algo i hm [11]. This algo i hm collec s he mos common ules
in duplex o iplex algo i hms. The THV algo i hm assigns he hall call o he nea es
li in he adequa e ip di ec ion (see appendix 1 o pseudocode).
Recen ly, mo e ad anced me hods ha e gained be e pe o mance. So, he Op imal
Rou ing algo i hm, he Dynamically Adap i e Call Alloca ion (DACA) and he Adap i e
Call Alloca ion (ACA) [12] a e all based on Dynamic P og amming. Also, p e ious
esea ch ela ed o So Compu ing echniques in ele a o sys ems has been esponsible
o impo an ad ances.
Fo example, algo i hms based on lea ning ha e been de eloped wi h success. The
con olle Neu os-I [13] o Fuji ec is a neu al ne wo k whe e he g oup ele a o s a e
and he li s s a e a e inpu s o he neu al ne wo k. The ne wo k has a p e ious lea ning
and subsequen adap i e au o- une online lea ning. Also, in he amewo k o he
lea ning, Rein o cemen Lea ning algo i hms [14] ha e shown an accu a e beha iou . I
consis s o a semi-Ma ko ian p ocess and uses an agen - eam whe e each agen con ols
one li . Unde hese condi ions wo a chi ec u es a e used: a pa allel a chi ec u e whe e
he agen s sha e he ne wo k (RLp, Pa allel Rein o cemen Lea ning) and a decen alised
a chi ec u e whe e each agen ha e i s own ne wo k (RLd, Decen alized Rein o cemen
Lea ning).
Fuzzy Logic has been p o ed as a aluable al e na i e when e alua ing a la ge amoun
o c i e ia in a lexible manne . The uzzy ele a o g oup con ol sys em [15] and he
Fuzzy Ele a o G oup Con olle wi h Linea Con ex Adap a ion [16] a e some
examples whe e di e se c i e ia a e used as he HCWTi (Hall Call Wai ing Time o he
i-li ), he maxHCWTi (maximum Hall Call Wai ing Time), he CVi (capaci y o
co e abili y o nex calls o he i-li ), and he minimum dis ance be ween new calls
and he las calls alloca ed GDi (Ga he ing Deg ee). Also in his line, gene ic algo i hms
[17] and [18] ha e been used wi h success o adjus he con ol se ings (a se o c i e ia)
in o de o gi e obus ness o he ele a o g oup con ol sys em, wi hin a se o g ea
a ie y o con ol pa ame e s. These wo ks allow adjus ing he con ol se ings
acco ding o indi idual loo u iliza ion si ua ions making use o a combina ion o ca
and loo a ibu es.
Also e olu iona y sys ems ha e e ealed success ul capaci ies in o de o maximize he
e iciency o he ele a o sys em call alloca ion. Gene ic algo i hms [19] and [20] ha e
been designed wi hin a disc e e e en simula ion ying o p edic he op imal decisions
o he ca dispa ch. Bo h a e sho -pape s wi h a non-wide explana ion o he me hods
and wi h an addi ional di icul y when ying o iden i y he c i e ion used o assessing
he quali y o he solu ions (by means o a pe o mance index). Howe e he au ho s
s a e he alida ion and success o he implemen a ion by he ep esen a ion o di e se
igu es and g aphics. Also, in his pape we ha e de eloped a gene ic algo i hm o
maximize he call alloca ion e iciency and o educe he o e all sys em wai ing ime.
He e, we p opose a gene ic algo i hm based on a hall call alloca ion s a egy (GAHCA)
o iden i y he ch omosomes o he popula ion indi iduals and we compa e ou p oposal
wi h con en ional duplex con olle s o he indus y in a disc e e e en s simula ion
scena io.
As he ele a o sys ems include unce ain y due o he u u e beha iou o he
passenge s is unknown, op imisa ion app oaches a e no o ally sui able. Ins ead o his,
he simula ion becomes a key ac o o demons a e he alida ion and accu acy o he
me hods and echniques as p e ious s ep o he physical implemen a ion (see [21] o a
wide pe spec i e).
The es o he pape ollows wi h he second sec ion dealing wi h he simula ion model
de ini ion o speci y he accu a e ele a o sys em pe o mance acco ding o he ules
p e iously s a ed. The hi d sec ion s a es he gene ic algo i hm cha ac e is ics. The
ou h sec ion shows he main esul s o he simula ions and he compa ison be ween
ou algo i hm and he adi ional duplex algo i hm. Finally, we highligh he main
conclusions in he inal sec ion.
2. SIMULATION MODEL
We ha e made use o he ARENA .5.0 so wa e o simula e he possible e en se .
ARENA is a powe ul in e ac i e isual modelling sys em ha makes use o he
SIMAN language. The ini ial model consis s o an anima ion zone and a module logical
zone ha can be di ided in o one con olle zone, one passenge zone and wo ele a o
zones o each o he cabins. The con olle , passenge and ele a o zones a e he
esponsible o he IF-ELSE ules ha manage he g oup ele a o sys em. The
op imisa ion algo i hm ( ha we will see in sec ion 3) is called in he passenge zone o
he call alloca ion.
2.1. Anima ion Zone
This zone is de ined by he A i e and Depa modules, which egula e he a i als
and depa u es o he passenge s a he sys em.
The A i e modules include he passenge a i al a e in he loo , he passenge
a i al ime (loaded in o he Time_A i al a ibu e), he passenge o igin loo
(loaded in o he O igin a ibu e) and he passenge des ina ion loo (loaded in o he
Des ina ion a ibu e).
The Depa modules ca y ou he inc ease o one uni in he loo depa u e coun e .
Also hey include he Time_Sys em as a ally bu e ing he passenge sys em wai ing
imes, as well as wo queues de ined by loo (one o passenge s going down and one
o passenge s going up, a excep ion o he g ound loo and he highes loo whe e
only one queue exis s).
We a e a aching one ideoclip o each o he simula ed algo i hms (THV and
GAHCA). The ideoclips include he simula ion unde he a ic and building
condi ions o sec ion 4.1. Mo eo e , he g aphical anima ion zone can be obse ed in
ideoclip1 and 2.
"Video clip 1. Gene i
c
algo i hm.wm "
"Video clip 2. THV
duplex al
g
o i hm.wm
2.2. Con olle Zone
One en i y has been c ea ed by li o a el a ound he logical zone. When he
passenge s come in o he li , he passenge s a e joined o he con olle en i y
shaping one only en i y a he same ime as holding all he pa icula indi idual en i ies
a ibu es.
2.3 Passenge Zone
The passenge zone consis s o he alloca ion o he UpDown a ibu e (1 i he
passenge goes up and 2 o he wise) ha is s a ed as unc ion o he O igin and
Des ina ion a ibu es. So, he passenge is sen o he wai ing queue i i exis s.
O he wise he hall call alloca ion p ocedu e is done by means o he co esponden
op imisa ion algo i hm (ou gene ic algo i hm by he case). The nex igu e 1 ep esen s
he con olle and passenge zone ARENA modules.
Figu e 1. Con olle and Passenge Zone ARENA modules
2.4. Ele a o Zone
When he li a i es a a loo he subsequen ac ions mus be checked and done i
necessa y: li wai s o calls, passenge s lea es he li , passenge s come in o he li ,
li alloca ion in case o ull capaci y, cabin call alloca ion and call e alua ion.
When he li a i es a a loo , he s a e o he li is e alua ed. I he li s a e is
se o ze o, he li is s opped and will ha e access o he Wai ing_ o _Calls
submodule. I he li is no s opped and i is ca ying passenge s, i inpu s in o he
Lea ing_ he_Li submodule. I i is no ca ying passenge s, i inpu s in o he
Taking_Passenge s submodule a e a Delay o simula e he opening doo s ime
(we use he delay a iable Time_Doo s(2.5 seconds).
When he li a i es a a loo , he A i al_E alua ion submodule p esen s h ee
op ions: he li con inues up, he li con inues down o he li s s a s he decele a ion
p ocess (p epa ing o s op). We use he LDX (T anspo e ID, uni
numbe ) as an ARENA p op ie a y a iable allowing o know he loo in which he
li is. We load his da a in he a iable Le el.
A e upda ing Le el, i he li is going up and he li is in he g ound loo , he li is
sen o he i s loo ; i he li is going down and he li is in he highes loo , he li
is sen o he las bu one loo . O he wise he simula ion model checks i he li has
s opped in he loo ha Le el indica es ( his da a can be checked by means o he
a iable Las _Visi ed_Floo ), in his case he li is sen up o down depending
on he ip di ec ion. O he wise he li s ops a he loo i he e exis s a cabin call o a
hall call and he capaci y is no ull. I he li is ull capaci y and i does no exis cabin
calls, he li is sen up o down depending on he ip di ec ion. Nex igu e 2 depic s
he main ARENA modules and submodules o he ele a o zone.
Figu e 2. Ele a o zone ARENA module

3. GENETIC ALGORITHM FOR THE CONTROLLER
We p opose a gene ic algo i hm ha makes use o a hall call alloca ion s a egy
(GAHCA) o pe o m he ele a o g oup con olle (see appendix 2 o pseudocode).
Fo each ime, , he hall calls and he cabin calls o he sys em a e e alua ed, alloca ing
he hall calls o one speci ic li . Each ime, , he se o hall calls a e ealloca ed
allowing he subsequen modi ica ion i he sys em pe o mance imp o es. Each ime
he se o decisions is aken managing all he a ailable in o ma ion (planning o he
long e m) bu only ca ying he immedia e ac ion ou o each li o he g oup: s op,
upwa ds o downwa ds displacemen .
So, each ime, , he simula ion model makes a call o he con olle op imiza ion
module ( he gene ic algo i hm) ha e u ns he o e all call alloca ion. The gene ic
algo i hm is de ined by he ollowing cha ac e is ics.
3.1. Indi iduals and popula ion
Two a ays o size [2·Numbe _o _Floo s-2] de ine he indi idual ch omosome.
Each o he a ays de ines he sys em s a e o each one o he li s. The a ay is
di ided in o wo pa s; he i s e e s o he up a ic and he second one o he down
a ic.
The i s Numbe _o _Floo s-1 in ege s co espond o he hall calls in he upwa d
di ec ion om he g ound loo o he highes loo . The second
Numbe _o _Floo s-1 in ege s co espond o he hall calls in he downwa d
di ec ion om he highes loo o he g ound loo . Figu e 3 depic s he ch omosome
indi iduals:
UP------------------------------>|<------------------------------DOWN
F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11F12F11F10 F9 F8 F7 F6 F5 F4 F3 F2
s a e(i_es )
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 [1x22]
Figu e 3. Indi idual ch omosome o a wel e loo s case building co esponding o one speci ic
ele a o o he g oup and i s associa ed physical bu on box
The a ay holds he in o ma ion e e ing o he hall calls by means o a bina y
codi ica ion. The bi 0 indica es no hall call a he loo , and he bi 1 indica es an
exis ing hall call a he loo .
The popula ion size is a majo ac o in he e ec i eness o gene ic algo i hm. I has
been p o ed [22] ha ela i ely small popula ions allow eaching success ul solu ions
wi h li le compu a ion e o . Ou expe imen s show ha inc easing he popula ion size
beyond 20, al hough inc easing he compu a ional e o , is no ewa ded by a
co esponding inc ease o pe o mance. So, we ha e main ained a popula ion size o 20
indi iduals in ou es s, al hough in eal implemen a ions his popula ion size could be
educed o en in o de o gain in compu a ional speed (wi h li le loss o e iciency).
3.2. Fi ness
We ha e used an app oxima e unc ion (in seconds) o es ima e he indi idual i ness.
The i ness unc ion e u ns he expec ed ime in which he ele a o g oup would se e
he en i e alloca ed hall calls and cabin calls. Ob iously, i will be es ima ion because o
he incapabili y o p edic ing he passenge u u e beha iou . The passenge a i al o
he loo is andom and hei des ina ions a e unknown.
The i ness es ima ion p ocedu e depends on he ele a o s a e (going up, down o
s opped). Howe e in e e y case i can be calcula ed by means o ou peak alues ha
we will no e as P1, P2, P3 y P4.
E e y ime, he p ocedu e has in accoun he o e all alloca ed hall calls s a ing each
new hall call o be alloca ed. Figu e 4 shows he op ions depending on he up o down
a ic.
The ele a o is s opped o going up
 P1. Cu en loo .
 P2. Highes loo o ake passenge s up. In igu e 4: displacemen a.
 P3. Lowes loo o ake passenge s down. In igu e 4: displacemen b.
 P4. The highes loo among he loo s lowe han P1 o ake passenge s up, always
P4<P1. In igu e 4: displacemen c.
Fi ness = [(P2-P1)+(P2-P3)+(P4-P3)]×[es ima ed in e loo ip ime]
I includes he maximum known upwa d ip plus he maximum known downwa d ip
plus he subsequen maximum known no -se ed upwa d ip in he i s up a ic
because P4<P1. We ha e o no e ha he ma hema ical exp ession do no include he
passenge des ina ion ips because we unknown i un il he passenge s come in o he
cabin.
The ele a o is going down
 P1. Cu en loo
 P2. Lowes loo o ake passenge s down. In igu e 4: displacemen a.
 P3. Highes loo o ake passenge s up. In igu e 4: displacemen b.
 P4. The lowes loo among he loo s highe han P1 o ake passenge s down,
always P4>P1. In igu e 4: displacemen c.
Fi ness = [(P1-P2)+(P3-P2)+(P3-P4)]×[es ima ed in e loo ip ime]
I includes he maximum known downwa d ip plus he maximum known upwa d ip
plus he subsequen maximum known no -se ed downwa d ip in he i s down a ic
because o P4>P1.
Figu e 4. Possible ele a o s eams o es ima e he i ness
Addi ionally, we ha e o conside a se ies o delays o es ima e he o al p ocess imes.
All hem a e associa ed o he cabins and include he decele a ion p ocess in he ele a o
a el speed, he opening doo s delay, he passenge incoming/ou coming p ocess, he
closing doo s delay and he accele a ion p ocess in he ele a o a el speed un il aking
he c uise speed. Usual alues a e 2 seconds o e e y delays excep ing o he
incoming/ou coming ime ( aking 5 seconds o his delay). Mo eo e , we a e aking 5
seconds as es ima ed in e loo ip ime a c uise speed.
3.3. Ope a o s
The gene ic ope a o s used a e c osso e and mu a ion. We ha e used an uni o m
c osso e ope a o ha andomly selec s wo indi iduals (pa en s) om he popula ion
and gene a es he o sp ing by c ossing he indi idual genes. The o sp ing inhe i s an
exac copy o hose genes ha a e equal in he pa en s’ ch omosome and, in o he case;
i inhe i s each gene wi h p obabili y o 50%. Al hough he pa en s’ selec ion is andom,
he algo i hm includes an inces p e en ion con ol when pa en s di e in less han a
gene pai . The mu a ion ope a o eplaces a hall call alloca ion om he indi idual
ch omosome by changing he genes om 01 o 10 o ice e se. The selec ion o he
indi idual is andom.
Tes s we e ca ied ou wi h di e en p obabili ies o applying c osso e and mu a ion o
he selec pa en s. Wi h c osso e , i was ound ha a ying he p obabili y om 50% o
100% had li le e ec on pe o mance, wi h a alue o 85-90% being ma ginally op imal
o he es s ca ied ou . A alue o 85% is used in he main uns. Fo mu a ion, alues
be ween 5% and 15% we e seen o be gi ing be e esul s han ypically smalle alues.
A alue o 15% is used in he main uns in o de o en ich he gene ic a ie y o he
popula ion. Howe e , i is o be no ed ha GAHCA is obus in he sense ha he
solu ions o he es p oblems we e achie ed on he whole wi h a wide ange o
pa ame e alues, and wi h no ine- uning equi ed o achie e e iciency.
3.4. Replacemen ule and gene a ions
We p opose he use o a hype geome ic unc ion allowing mo e p obabili y o
eplacemen o indi iduals wi h wo se i ness and less p obabili y o eplacemen o
indi iduals wi h be e i ness. So, he indi idual in anking posi ion-i, ha e a
eplacemen p obabili y equal o q(1-q)i, being q he eplacemen p obabili y o he
wo s indi idual. We ob ained he be e pe o mances se ing a alue o q be ween 55-
65%. The main es s a e un wi h a alue o 60%.
Addi ionally o he eplacemen ule, we inco po a e an indi idual duplici y con ol in
he popula ion gene a ion.
The numbe o gene a ions (o i e a ions) o he gene ic algo i hm can be a c i ical
pa ame e when we y o each e iciency o he solu ion and sho ime execu ion.
Gene ic algo i hms a e i e a i e and he e o e hey can ake e y much ime o
execu ion. In eal cases he numbe o gene a ions will be bounded by he exigencies o
he eal implemen a ion. Howe e he ad an age o gene ic algo i hms is ha hey can
be s opped a any ime ha ing he be e solu ion a he momen . We expe imen ed wi h
di e se pa ame e s o he numbe o gene a ions: simila alues we e ob ained o he
in e al be ween 50 and 100 i e a ions and he inc eases on he quali y o he solu ions
we e mode a ed be ween 20 and 50. In any case, a leas 20 i e a ions should be done.
4. SIMULATION RESULTS
4.1. Da a o he es s: building and lunchpeak a ic
We ha e es ed he algo i hms in a wel e loo s building. The e a e 30 wo ke s in each
o he building loo s excep ing he 7 h loo ( he adminis a ion depa men wi h 60
wo ke s) and he 12 h loo ( he manage depa men wi h 15 wo ke s). The e a e wo 20
pe sons capaci y ele a o s in he hall.
The in e loo a el p obabili ies a e de ined wi hin a lunchpeak a ic si ua ion:
F om he g ound loo :
 To he 7 h loo : 15%
 To he 12 h loo : 4%
 To he es o he loo s: 9%
F om o he loo s:
 To he g ound loo : 95%
 To he es o he loo s: 5%
The nex igu e 5 depic s he a i al a e du ing lunchpeak a ic. Mos o he wo ke s
go ou o lunch du ing he in e al [14:00,15:00] hou s, e u ning o he building du ing
[15:20,16:00] hou s. We ha e aken hese da a om di ec eal li e inspec ion in a such
case building.
STATE OF OCCUPATION OF THE ELEVATORS (GAHCA)
STATE OF OCCUPATION OF THE ELEVATORS (THV)
0: S op
1: Up
2: Down
Figu e 9. Analysis o ele a o occupa ion
5. CONCLUSIONS
We ha e p oposed a gene ic algo i hm (GAHCA) o con ol he ele a o g oup in a
p o essional building. The esul s allow us o a i m ha ou gene ic algo i hm eaches a
be e pe o mance a ending o he sys em wai ing imes and queue sizes han
adi ional con olle s in indus y applica ions as THV algo i hm. The educ ion o
wai ing imes has been almos he 25% a he same ime as ge ing a signi ican
educ ion o he hall down queues. In his si ua ion he passenge s a e supposed o
expe imen a sys em ime educ ion om 3min15sec o 2min30sec. The analysis has
been done unde hea y lunchpeak a ic condi ions.
The esul s ob ained in he pape allow us o a i m ha gene ic algo i hms, in gene al,
and ou GAHCA in pa icula , a e aluable ools wi h a g ea po en ial in he con ol o
ele a o sys ems. Howe e , he implemen a ion o such ype o algo i hms in eal
con olle s has o be done ca e ully in o de o main ain bounded he esponse ime o
he algo i hm. Gene ic algo i hms a e i e a i e and he e o e hey can ake e y much
ime o execu ion when a long popula ion and a g ea numbe o i e a ions a e used. The
elec ion o hese pa ame e s has o be selec ed a ending no so much o he algo i hm
accu acy bu o he a ailable ime o ip o he ele a o be ween di e en e en s, ha is
he ime necessa y o alloca e a hall call (i can be he ip ime be ween a numbe o
loo s de e mined, e.g. no mo e han wo). In eal cases an al e na i e can be s opping
he algo i hm p e iously o each he nex e en , which would occu a e a known ime
in e al. O cou se all hese decisions a e e y dependan on he compu a ion speed o
he elec onic mic ochips ins alled by he company.

Acknowledgemen s
This pape has been ca ied ou in collabo a ion wi h MAC PUAR, S.A. (MP). MP has been suppo ing
ou esea ch on ele a o sys ems since 2000. Addi ionally, he au ho s acknowledge he inancial suppo
gi en by he Minis e io de Ciencia y Tecnologia, in i s Indus ial P oduc ion and Design P og amme
(p ojec e . DPI2002-01264), Spain.
Appendix 1. THV pseudocode
N = numbe o loo s in he building
Read he sys em cu en s a e
d = Dis ance (call, ca ) = |call loo – ele a o loo |
IF ele a o is homing o he call loo wi h he same ip di ec ion o he hall call
Fi ness Func ion = N+1-d
ELSE IF ele a o is homing o he call loo wi h ip di ec ion di e en om he hall call
Fi ness Func ion = N-d
ELSE IF ele a o has jus lea ed he loo o he hall call
Fi ness Func ion = 1
ELSE ( he ele a o is s opped)
Fi ness Func ion = N-d
Ca Alloca ion = Bes Fi ness Func ion
ARENA assigna ion
Appendix 2. GAHCA pseudocode
Read he sys em cu en s a e
Gene a e he popula ion
Calcula e he i ness popula ion
O dina e he i ness popula ion
IF Popula ion size = 20 THEN
i = 0
WHILE i < 50
p = Rnd
IF p < 0.85 THEN ‘C osso e ope a o ’
Inces =1
WHILE inces =1
Randomly selec ion o pa en s
Pa en s inces p e en ion
IF No inces THEN
C osso e -> o sp ing
END IF
END WHILE
ELSE ‘Mu a ion ope a o ’
Randomly selec ion o pa en
Mu a ion -> o sp ing
END ELSE
Indi iduals duplici y con ol
IF No duplici y THEN
E alua ion o he indi idual i ness
Selec ion o indi idual o eplacemen
New indi idual -> o sp ing
Modi ica ion o he popula ion i ness able
i = i + 1
END IF
END WHILE
END IF
Solu ion = Bes i ness indi idual
ARENA assigna ion
Re e ences
[1] MacDonald, C. Robe and E. Ab ego. Coinciden call op imiza ion in a ele a o
dispa ching sys em, Wes inghouse Elec ic Co p. U.S. Pa en No. 4 782 921,
1988.
[2] Thanga elu and Kandasamy. Queue based ele a o dispa ching sys em using
peak pe iod a ic p edic ion, O is Ele a o Company. U.S. Pa en No. 4 838
384, 1989.
[3] Thanga elu and Kandasamy. "A i icial in elligence", based lea ning sys em
p edic ing "peak-pe iod" imes o ele a o dispa ching, O is Ele a o Company,
U.S. Pa en No. 5 241 142, 1993.
[4] Kameli, N. and Nade . Floo popula ion de ec ion o an ele a o sys em, O is
Ele a o Company. U.S. Pa en No. 5 511 635, 1996.
[5] Kameli, N., Nade , Collins and M. James. Ele a o downpeak sec o ing, O is
Ele a o Company. U.S. Pa en No. 5 480 006, 1996.
[6] Kim, C. and O. Jeong. G oup managemen con ol me hod o ele a o sys em
employing a ic low es ima ion by uzzy logic using a iable alue
p e e ences and decisional p io i ies, LG Indus ial Sys ems Co., L d. U.S.
Pa en No. 5 679 932, 1997.
[7] Bahja , S. Zuhai and J. Bi a . Au oma ed selec ion o high a ic in ensi y
algo i hms o up-peak pe iod, O is Ele a o Company. U.S. Pa en No. 5 168
133, 1992.
[8] Siikonen, M-L.. Ele a o g oup con ol wi h a i icial in elligence, Helsinki
Uni e si y o Technology, Sys ems Analysis Labo a o y, Resea ch Repo s A67
(1997).
[9] Haup meie , D., S.O. K umke and J. Rambau. The online dial-a- ide p oblem
unde easonable load, P ep in SC 99-08, Kon ad-Zuse-Zen um ü
In o ma ions echnik Be lin (1999).
[10] La añe a, J. and Co es, P. Op imización Dinámica en Sis emas de T á ico
Ve ical. Escuela Supe io de Ingenie os. Ingenie ía de O ganización. Se ille
Uni e si y. Technical Repo IO-01MP
[11] Ba ney, G.C. and S.M. dos San os. Ele a o T a ic Analysis, Design and
Con ol (Pe e Pe eg inus L d, 2nd edi ion, London, 1985).
[12] Siikonen, M-L.. Planning and con ol models o ele a o s in high- ise buildings,
Ph.D. Thesis, Helsinki Uni e si y o Technology, 1997.
[13] Sasaki, K., S. Ma kon and M. Makagawa. Ele a o G oup Supe iso y Con ol
Sys em Using Neu al Ne wo ks, Ele a o Wo ld (1996).
[14] C i es, R.H. and A.G. Ba o. Imp o ing ele a o Pe o mance Using
Rein o cemen Lea ning, Ad ances in Neu al In o ma ion P ocessing Sys ems 8.
MIT P ess (1996).
[15] Kim, C., K.A. Seong and H. Lee-kwang. Design and implemen a ion o a uzzy
ele a o g oup con ol sys em, in: P oceedings o he IEEE T ansac ions on
sys ems, man and Cybe ne ics (1998), ol. 28, No. 3, 277-287.
[16] Gudwin, R., F. Gomide and M.A. Ne o. A Fuzzy Ele a o G oup Con olle wi h
Linea Con ex Adap a ion, in: P oceedings o FUZZ-IEEE98, WCCI’98 - IEEE
Wo ld Cong ess on Compu a ional In elligence, Ancho age, Alaska, USA (1998)
481-486.
[17] Fujino, A., T. Tobi a, K. Segawa, K. Yoneda and A. Togawa. An Ele a o G oup
Con ol Sys em wi h Floo -A ibu e Con ol Me ed and Sys em Op imiza ion
using Gene ic Algo i hms, IEEE T ansac ions on Indus ial Elec onics Vol. 44 no.
4 (1997) 546-552.
[18] Tobi a, T. A. Fujino, K. Segawa, K. Yoneda and Y. Ichikawa. A Pa ame e
Tuning Me ed o an Ele a o G oup Con ol Sys em using a Gene ic Algo i hm,
Elec ical Enginee ing in Japan, Vol. 124, No. 1 (1998) 55-64.
[19] Gudwin, R. and F. Gomide. Gene ic Algo i hms and Disc e e E en Sys ems: An
Applica ion, in: P oceedings o The Fi s IEEE Con e ence on E olu iona y
Compu a ion, IEEE Wo ld Cong ess on Compu a ional In elligence (1994), ol II,
742-745.
[20] Alande , J.T., J. Ylinen and T. Tyni. Ele a o G oup Con ol Using Dis ibu ed
Gene ic Algo i hm, in: P oceedings o he In e na ional Con e ence. Sp inge -
Ve lag, Vienna, Aus ia (1995), 400-403.
[21] Siikonen, M-L.. Ele a o a ic simula ion, Simula ion Vol. 61 No. 4 (1993) 257-
267.
[22] Goldbe g, D.E. Sizing popula ions o se ial and pa allel gene ic algo i hms, in
P oceedings o he Thi d In e na ional Con e ence in Gene ic Algo i hms, (J.D.
Scha e , ed.) (1989), 70-79.