UNIVERSIDAD DE LAS PALMAS DE GRAN CANARIA
AGENCY SERVICES FOR THE MANAGEMENT
OF DISTRIBUTED ENERGY NETWORKS USING
PARALLEL AUCTION MARKETS
by
Ignacio Jos´
e L´
opez Rod ´
ıguez
A hesis submi ed o he
Uni e si y Ins i u e o
In elligen Sys ems and Nume ic Applica ions in Enginee ing (SIANI)
Uni e si y o Las Palmas de G an Cana ia
o he deg ee o
DOCTOR OF PHILOSOPHY IN COMPUTER SCIENCE
Sep embe o 2015
AGENCY SERVICES FOR THE MANAGEMENT OF
DISTRIBUTED ENERGY NETWORKS USING PARALLEL
AUCTION MARKETS
Uni e si y Ins i u e o
In elligen Sys ems and Nume ic Applica ions in Enginee ing (SIANI)
Uni e si y o Las Palmas de G an Cana ia
A hesis p esen ed by
Ignacio Jos´
e L´
opez Rod ´
ıguez
Di ec ed by
D . F ancisco Ma io He n´
andez Teje a
The PhD s uden The di ec o
Las Palmas de G an Cana ia, Sep embe o 2015
A mis pad es y mis he manos;
a ese odo del que me sien o
pa e has a los huesos.
A mis ´
ıas, Ma ´
ıa Dolo es y
Ma ´
ıa Vic o ia.

Abs ac
Facing g owing ene gy demand, he exhaus ion o ossil uels, and he e ec
o g eenhouse gases equi es mo ing owa ds a new model o elec ical g id.
Awa e o his si ua ion, he US and Eu ope’s go e nmen s a e wo king on he
de elopmen o he Sma G id, which is de ined as a dis ibu ed, eac i e and
in elligen g id ha will allow modula ing he demand and use o a ailable
powe gene a ion dynamically, hus acili a ing he in eg a ion o enewable
ene gy sou ces sa ely and e icien ly. One o he mos impo an aims o
he Sma G id is o inc ease he pa icipa ion o use s in he managemen
sys em o he ne wo k, who, by using sma local de ices, a e en isioned o
be able o con igu e hei consump ion habi s acco ding o ene gy p ices and
o con ibu e in he gene a ing ace .
The eno mous dependence o mode n socie y on ene gy supply p o okes he
need o a g adual ansi ion ha con e s he Sma G id in o a long- e m
p ocess. The ini ial s age o his ansi ion is suppo ed by small dis ibu ed
ene gy ne wo ks ha a e designed o be sel -managed a eas o he dis ibu ion
ne wo k composed o modula loads and dis ibu ed ene gy sou ces. These
ne wo ks ep esen con olled en i onmen s o he inclusion o enewable
ene gy sou ces and he ins alla ion o dis ibu ed and eac i e managemen
sys ems such as hose en isioned in he Sma G id. The Chap e s 1 and 2
a e de o ed o desc ibing he elec ical g id and he Sma G id, including
hei p esen o m, mission and challenges.
The cha ac e is ics o in elligen agen s make hem a pa icula ly sui able
echnology o managing en i onmen s in a dis ibu ed, eac i e, in elligen
and pa icipa i e manne . Acco dingly, he e a e many oices ad oca ing he
implemen a ion o he managemen sys em o he Sma G id as a mul i-agen
sys em. Howe e , hough in elligen agen s ha e been a p omising echno-
logy o he las i een yea s, he u h is ha , a om achie ing his, he
echnology has ailed o es ablish i sel as a p ac ical solu ion in many ech-
nological se ings ha demand i s unc ionali y, as he Sma G id now does.
Among hese a e compu a ional g ids, P2P ne wo ks and i ual o ganiza-
ions, which ha e op ed o mo e p ac ical solu ions, elega ing he in elligen
agen s o he academic sphe e.
Awa e o his ba ie o en y, his hesis designs and implemen s an a chi ec-
onic model called Agency Se ices (Chap e 3). This aims o acili a e he
easy and ealis ic in eg a ion o so wa e agen s in o he i ual en i onmen s
ha ha e a isen as esul o he ecen ad ances in in o ma ion and commu-
nica ion echnology, being especially sui able o implemen ing elec onic
ma ke s and managemen sys ems based on coo dina ion and nego ia ion ac-
ions, such as hose expec ed in he Sma G id. The Agency Se ices model
is hea ily based on he Cloud Compu ing pa adigm, which has p o en o be
he ype o solu ion ha use s adop in p ac ice. Also, he new model akes
impo an lessons om solu ions ha ha e succeed in simila en i onmen s,
such as compu a ional g ids and P2P ne wo ks.
This hesis also p o ides a de ailed e iew o he mos impo an algo i hms
o he li e a u e o he implemen a ion o ene gy ma ke s in dis ibu ed en-
e gy ne wo ks (Chap e 4). The e iew is pa icula ly ocused on agen -based
solu ions because hey a e he unique app oach o empowe use s and imple-
men eac i e and dis ibu ed solu ions. The e iew classi ies and s udies he
li e a u e acco ding o he p esen needs o ene gy ma ke s, highligh ing he
mos comple e p oposals and discussing he subjec s on which wo k emains
o be done. Finally, an algo i hm based on e e se pa allel auc ions has been
selec ed as he mos sui able o achie ing dis ibu ed, lexible and eac i e
managemen sys ems based on au onomous en i ies.
The combina ion o bo h he a chi ec onic solu ion and he ene gy ma ke ’s
algo i hm is e alua ed in a no el co-simula ion in as uc u e specially de-
signed o his pu pose, which combines he bes solu ions o bo h wo lds,
mul i-agen sys ems and he elec ical g id (Chap e 5). The expe imen al
e alua ion is based on demand- esponse p og ams, which a e en iched wi h
a new concep ual model based on c i ical loads and nega i e loads ha opens
he doo o he implemen a ion o ma ke mechanisms (Chap e 6). Demand-
esponse p og ams a e chosen in pa icula because hey a e one o he mos
immedia e and ealis ic miles ones on he oad o he Sma G id. The i ues
and bene i s o he Agency Se ices model a e p o en bo h quan i a i ely
and quali a i ely. Speci ically, he Agency Se ices model has been shown o
acili a e he ins alla ion o dis ibu ed agen -based solu ions in es ic ed en-
i onmen s, o simpli y he echnical equi emen s o he clien ’s acili y, and
o imp o e he eliabili y o he sys em. Fu he mo e, in he pa icula case o
ene gy managemen , i has been p o en ha ene gy balancing is success ully
accomplished h ough nego ia ions be ween so wa e agen s.
As he expe imen al e alua ion e eals, he e ec i eness o pa allel auc ions
alls d ama ically as he dis ibu ion o buye s be ween auc ions becomes
less uni o m. This ac , which has no been add essed in he li e a u e so a ,
may u n pa allel auc ions in o a useless managemen sys em. To sol e his
p oblem, his hesis designs and implemen s a no el solu ion ha manages o
dis ibu e buye s p ope ly (Chap e 7). The mechanism no iceably imp o es
he e ec i eness o pa allel auc ions and is guided by a se o ules ha p e-
se es ma ke compe ence. Fu he mo e, i is ailo ed o he needs o la ge
and highly dis ibu ed en i onmen s, such as he Sma G id and compu a-
ional g ids.
The solu ion p esen ed in his documen is ully complian wi h he main
ene gy s anda ds. In pa icula , he Agency Se ices model is complian wi h
he Ene gy In e ope a ion model de ined by OASIS. As o DR p og ams,
hey a e implemen ed using he OpenADR s anda d, which is he mos widely
adop ed solu ion by selle s in his ma e . Fu he mo e, i is no ewo hy ha
all wo k de eloped in he cou se o his hesis espec he p inciples o he
ep oducible esea ch mo emen , so ha i can be e alua ed and e i ied by
o he sea che s.

Lis o Figu es
1.1 O e iew o he s uc u e o he elec ical g id. . . . . . . . . . . 5
1.2 Gene al p ope ies o he p ima y, seconda y and e ia y con ols. 10
2.1 Basic scheme o mic o-g ids. . . . . . . . . . . . . . . . . . . . . 25
2.2 A chi ec u e based on he concep ene gy cell o he CRISP p ojec . 26
2.3 Con ol le els o he mic o-g id en i onmen . . . . . . . . . . . . 28
2.4 Common oles ha play so wa e agen s in ene gy cells. . . . . . . 29
2.5 Pa ies in e ac ing using ende s and ansac ions in he EI s anda d. 33
2.6 Example o in e ac ions be ween VTN and VEN nodes. . . . . . . 34
2.7 P o iles o he OpenADR 2.0 speci ica ion. . . . . . . . . . . . . . 36
2.8 Tasks and in e ac ions assigned o he sma local de ices in dis-
ibu ed con ol schemes o he Sma G id. . . . . . . . . . . . . 38
2.9 Disallowed in e ac ions be ween local nodes in ene gy cells and
mic o-g ids.............................. 39
2.10 Main ypes o P2P ne wo ks. . . . . . . . . . . . . . . . . . . . . 42
2.11 S uc u al pa e n o he Gnu ella ne wo k. . . . . . . . . . . . . . 43
3.1 Gene al scheme o he Agency Se ices model. . . . . . . . . . . 55
3.2 B oke agen wo king as p oxy o mul iple local agen s. . . . . . . 57
3.3 Simpli ied e sion o he Agency Se ices model. . . . . . . . . . 58
3.4 In e ac ions o an ASPEM node. . . . . . . . . . . . . . . . . . . 67
3.5 Co espondence be ween he a chi ec u e o he EI s anda d and
he Agency Se ices model. . . . . . . . . . . . . . . . . . . . . 68
xiii
LIST OF FIGURES
4.1 Linea piece-wise unc ions ha de ine he beha io o he gene a-
ion and consump ion o ene gy. . . . . . . . . . . . . . . . . . . 75
4.2 Time slo s o ene gy ma ke s exp essed as a bina y ee. . . . . . . 82
5.1 Componen s o he simula ion in as uc u e. . . . . . . . . . . . 100
5.2 UML2 componen s diag am o he so wa e modules ha make up
he simula ion in as uc u e. . . . . . . . . . . . . . . . . . . . . 101
5.3 Fo m o egis e a new scena io. . . . . . . . . . . . . . . . . . . 107
5.4 UML2 sequence diag am o he ini ia ion s age o he simula ion
li ecycle. .............................. 118
5.5 UML2 sequence diag am o he execu ion s age o he simula ion
li ecycle. .............................. 119
5.6 UML2 sequence diag am o he comple ion s age o he simula ion
li ecycle. .............................. 119
6.1 Demand cu e co esponding o 1s Augus , 2000 in he IEEE 13-
node when no signal is applied. . . . . . . . . . . . . . . . . . . . 125
6.2 Demand cu e co esponding o 1s Augus , 2000 in IEEE 13-node
when a del a signal o 30.000 kW is applied. . . . . . . . . . . . 128
6.3 Demand cu e co esponding o 1s Augus , 2000 in IEEE 13-node
when a del a signal o 70.000 kW is applied. . . . . . . . . . . . 130
6.4 Linea piece-wise unc ion used by he bidde s in auc ion ma ke s. 133
6.5 A ec ed pa o he demand cu e o he pa allel auc ion ma ke
(wi hou s a ing p ices) when a mode a e signal is o de ed. . . . . 135
6.6 Demand cu e o he pa allel auc ion ma ke (wi hou s a ing p ices)
when a high signal is o de ed. . . . . . . . . . . . . . . . . . . . 136
6.7 A ec ed pa o he demand cu e o he pa allel auc ion ma ke
when he du a ion o he ma ke is se o 60 minu es. . . . . . . . 138
7.1 O e iew o he HUDP mechanism. . . . . . . . . . . . . . . . . 150
7.2 UML2 sequence diag am o HUDP. . . . . . . . . . . . . . . . . 153
7.3 Buye s associa ed wi h a posi ion o he lis o selle s acco ding o
hei iden i ica ion oken. . . . . . . . . . . . . . . . . . . . . . . 155
xi
LIST OF FIGURES
7.4 O e -concen a ion o buye s is limi ed by he on ie o each g oup
o selle s. .............................. 159
7.5 Linea piece-wise unc ion. . . . . . . . . . . . . . . . . . . . . . 161
7.6 Buye s lis s o he auc ions. . . . . . . . . . . . . . . . . . . . . . 162
7.7 Example o ex a buye s accumula ed o e he auc ions. . . . . . . 162
7.8 Example o shi ing he s a ing posi ion o buye s acco ding o he
su pluselemen s. .......................... 163
7.9 His og am o he co e ed auc ions in each scena io. . . . . . . . . 170
7.10 Pe cen age o co e ed auc ions depending on he a io Rca. . . . . 171
7.11 E ec o he cons an C on he pe o mance o he HUDP mechanism.172
7.12 E olu ion o he numbe o emp y auc ions wi h he inc easing o C .173
7.13 E olu ion o he numbe o canceled auc ions wi h he inc easing
o C . ................................ 173
7.14 E ec o he cons an C on he pe o mance o he HUDP mech-
anism when playe s adop sha ed s a egies. . . . . . . . . . . . . 175
C.1 A qui ec u a b´
asica de una mic o ed. . . . . . . . . . . . . . . . 208
C.2 A qui ec u a basada en el concep o celda de ene g´
ıa del p oyec o
Eu opeoCRISP............................ 209
C.3 Ta eas e in e acciones que se asocian a los disposi i os de con ol
locales en el ma co del sis ema de ges i´
on del Sma G id. . . . . . 214
C.4 Tipos p incipales de opolog´
ıa de las edes P2P. . . . . . . . . . . 215
C.5 Esquema gene al del modelo de Se icios de Agencia. . . . . . . 220
C.6 Esquema gene al de la e si´
on simpli icada del modelo de Se i-
ciosdeAgencia............................ 222
C.7 In e acciones de un nodo ASPEM. . . . . . . . . . . . . . . . . . 225
C.8 Componen es de la in aes uc u a de simulaci´
on.......... 233
C.9 Funci´
on lineal a ozos que en ´
ıan los p oduc o es a los consumi-
do es pa a ep esen a sus o e as. . . . . . . . . . . . . . . . . . 236
C.10 Cu as de demanda cuando se aplica una se˜
nal OpenADR de ipo
mode a e. .............................. 237
C.11 Cu as de demanda cuando se aplica una se˜
nal OpenADR de ipo
ha d.................................. 238
x
LIST OF FIGURES
C.12 Cu as de demanda cuando se aplica una se˜
nal OpenADR de ipo
del a que o dena una educci´
on de 30.000 kW. . . . . . . . . . . . 240
C.13 Comp ado es asociados a una posici´
on de la lis a de endedo es. . 242
C.14 La concen aci´
on de los comp ado es es ´
a limi ada po la on e as
de los g upos de subas as. . . . . . . . . . . . . . . . . . . . . . . 244
C.15 Funci´
onlineala ozos........................ 245
C.16 Ejemplo de comp ado es ex a acumulados en cada subas a. . . . 246
C.17 Ejemplo de comp ado es que son easignados a o as subas as. . . 246
C.18 His og ama de la can idad de o e a cubie a cuando se usa el me-
canismo HUDP y los pa icipan es adop an es a egias. . . . . . . 251
x i
Lis o Tables
4.1 P ope ies o desc ibing he wo ks ha aim o implemen he SDM
exchange model a medium o la ge DENs. . . . . . . . . . . . . 91
4.2 Desc ip ion o he wo ks acco ding o he cha ac e is ics o ene gy
ma ke s................................ 92
4.3 Desc ip ion o he wo ks acco ding o hei s anda d-o ien a ion
and he capaci y o ep oduce expe imen s based on hem. . . . . . 93
5.1 Desc ip ion o he laye o REST ul web se ices p o ided by he
applica ion ha ep esen s he sys em ope a o . . . . . . . . . . . 108
5.2 Desc ip ion o he laye o REST ul web se ices p o ided by he
applica ion ha ep esen s ASPEMs. . . . . . . . . . . . . . . . . 110
6.1 Consump ion o he ASPEMs pe OpenADR le el. . . . . . . . . 124
6.2 Consump ion o he ASPEMs pe ype o load. . . . . . . . . . . 124
6.3 Numbe o nodes associa ed o each ype o schedule. . . . . . . . 125
6.4 Consump ion o he ASPEMs pe le el when a del a signal o 30.000
kW isapplied. ........................... 129
6.5 Numbe o signals pe ype when a del a signal o 30.000 kW is
applied. ............................... 129
6.6 Amoun o load pe ype when a del a signal o 75.000 kW is applied.129
6.7 Numbe o signals pe ype when a educ ion o 75.000 kW is applied.131
6.8 Desc ip ion o he simula ed scena ios using pa allel auc ion ma ke s.134
6.9 Da a co esponding o he simula ion o pa allel auc ion ma ke s
when signals o ype mode a e and high a e applied. . . . . . . . . 139
x ii

LIST OF TABLES
7.1 Desc ip ion o he scena ios used in he expe imen s. . . . . . . . 168
7.2 Scena io #1: Resul s when s a ing p ices a e no used. . . . . . . 168
7.3 Scena io #1: Resul s when using s a ing p ices and mul iple bid-
dingle els. ............................. 168
7.4 Scena io #6: Resul s when no using s a ing p ices. . . . . . . . . 169
7.5 Scena io #6: Resul s when using s a ing p ices and mul iple bid-
dingle els. ............................. 169
7.6 Pe o mance o he scena io wi h a io Rcaequal o 3when he
cons an C a ies. ......................... 174
8.1 Me a desc ip ion o he a icle [LRHT15], de eloped as pa o his
disse a ion, and published in he jou nal Applied Ene gy. . . . . . 189
8.2 Ranking o he jou nal Applied Ene gy. .............. 189
8.3 Me a desc ip ion o he a icle [LRHTHC15], de eloped as pa o
his disse a ion, and published in he jou nal Expe Sys ems wi h
Applica ions. ............................ 190
8.4 Ranking o he jou nal Expe Sys ems wi h Applica ions. . . . . . 190
C.1 Desc ipci´
on de los algo ´
ımos m´
as des acados del es ado del a e
pa a la ges i´
on de en o nos de ene g´
ıa dis ibuida usando mecanis-
mosdeme cado. .......................... 229
C.2 Consumo de los ASPEM po cada ni el de consumo p opio de las
se˜
nales de ipo simple deOpenADR................. 234
C.3 Pe il del me cado de subas as. . . . . . . . . . . . . . . . . . . . 235
C.4 Da os co espondien es a las simulaciones de los me cados de su-
bas as pa alelas cuando se aplican las se˜
nales mode a e yha d. . . 239
C.5 Desc ipci´
on de los escena ios usados en los expe imen os del me-
canismoHUDP............................ 247
C.6 HUDP: Escena io #1: Resul ados pa a los casos en los que no se
usan p ecios de en ada. . . . . . . . . . . . . . . . . . . . . . . . 248
C.7 HUDP: Escena io #1: Resul ados pa a los casos en los que se usan
p ecios de en ada y pujas en o ma de unciones lineas a ozos. . 249
C.8 HUDP: Escena io #6: Resul ados pa a los casos en los que no se
usan p ecios de en ada. . . . . . . . . . . . . . . . . . . . . . . . 249
x iii
LIST OF TABLES
C.9 HUDP: Escena io #6: Resul ados pa a los casos en los que se usan
p ecios de en ada y pujas en o ma de unciones lineas a ozos. . 250
xix
¡Es oy has a los huesos de u cue po!. . . ¡De u ca ne de
homb e que no aguan a los iempos!. . . ¡Ni aguan a el sol
de es ´
ıo!. . . ¡Ni los ´
ıos de diciemb e!. . . ¡Pa a es o c i´
e yo
mis pechos, du os como el pede nal!. . . ¡Pa a es o c i´
e yo
mi boca, esca como la pa ´
ıa!. . . ¡Pa a es o e di yo dos
hijos, que ni el anda de la caballe ´
ıa ni el mal ai e en la
noche supie on aguan a !
“La amilia de Pascual Dua e”, Camilo Jos´
e Cela.
Su nomb e amon ona pasado en mis ojos.
“Es e P ima e a”, Robe o A l .
CHAPTER
1
The elec ical g id
The p oli ic in en o Thomas Edison in oduced he i s elec ic powe sys em
in New Yo k Ci y in 1882. The Edison Illumina ing Company ope a ed wi h di ec
cu en a 110 ol s and ini ially supplied ligh o 59 cus ome s in he Wall S ee
a ea. Unawa e o he impo an social bene i s ha would a ise la e , he main goal
o Thomas Edison was o c ea e a new p o i able business. By he end o 1880s,
Edison had sold he pa en o gene a ing and ansmi ing elec ici y, and many
ci ies o US and Eu ope had se up many simila small cen al s a ions capable o
supplying ew ci y blocks. Howe e , due o he use o di ec cu en , he ange
o hese i s gene a ing s a ions was limi ed o a couple o kilome e s. In 1888,
Nikola Tesla, a o me employee o Edison, ecei ed a pa en o he induc ion mo-
o , which would enable he high- ol age ansmission o al e na ing cu en o e
long dis ances wi h low losses. In 1896, Nikola Tesla, by hen wo king o Wes ing-
house Elec ic & Manu ac u ing Co, u ned his idea in o eali y by cons uc ing a
hyd oelec ic s a ion in he Niaga a Falls ha was capable o ansmi ing signi i-
can amoun s o powe o Bu alo, New Yo k, mo e han 32 kilome e s away. In he
end, his inno a ion would es ablish he ounda ion o an elec ical g id designed
o cen alized gene a ion and dis ibu ed loads, a concep ha has endu ed o he
p esen day.
1
1. THE ELECTRICAL GRID
es ablished by he sys em s anda d, which is 50 Hz in Eu ope and 60 Hz in he
US. In mo e de ail, i p oduc ion exceeds demand, he o a ional mo emen o
he gene a o speeds up, hus inc easing he equency. Con e sely, i he e is a
empo a y p oduc ion de iciency, he gene a o ’s o a ional speed slows down and
equency dec eases. I is impo an o no e ha he p ima y con ol does no e u n
equency o no mal, bu only s abilizes i .
When he ene gy imbalance canno be sol ed by small co ec ions o indi idual
engines, i is necessa y o pu in o ac ion a coo dina ed plan ha in ol es gene-
a o s om mul iple powe plan s. This kind o ac ion belongs o he ealm o
he seconda y con ol le el. I usually akes ew minu es and aims o es o e he
minu e- o-minu e balance by se ing he equency o he s anda d alue o he g id
a e i has been no malized by he p ima y con ol.
To acili a e he con ol mission, some na ional g ids a e di ided in o con ol
a eas. Each a ea is go e ned by he igu e o Balancing Au ho i y (BA), which
is in cha ge o ensu ing he ope a ion wi hin an a ea h ough he de elopmen o
esou ce plans, main aining he balance be ween load and gene a ion, con olling
ansmission lows and ol ages, and ensu ing ha equency is held wi hin no mal
limi s. When balancing a eas canno mee he demand using hei own esou ces,
o he e is su plus o p oduc ion ha can be consumed in o he egions, BA can
manage exchanges wi h neighbo ing con ol a eas. The numbe o BAs, as well as
he egions hey un, is de e mined by he sys em ope a o . In he US he e a e o e
130 balancing au ho i ies, al hough o e he las pas se e al decades, mo i a ed
by economies o scale, hey a e g adually ge ing la ge as a esul o he union o
some o hem. The common ule in Eu ope is ha each na ional g id wo ks as a
unique con ol a ea ha can exchange ene gy wi h connec ed neighbo ing coun-
ies. On he o he hand, co ec ing he de ia ions o balancing a eas is based on
he alue A ea Con ol E o (ACE), es ima ed in MW. ACE is a measu e o e o
in he sys em equency ha helps o iden i y di e ences be ween he ac ual and
he scheduled ne powe low wi hin a con ol a ea. A posi i e alue o ACE means
ha gene a ion wi hin he a ea exceeds he load by mo e han he expec ed alue.
In his case, he gene a ion wi hin he con ol a ea mus be educed. Con e sely,
nega i e ACE means local gene a ion mus be inc eased.
8

1.4 Ope a ion
The elimina ion o ACE equi es he coo dina ed ac ion o mul iple gene a o s
o mul iple powe plan s. These kind o ac ions can only be un wi h an o e iew
o he con ol a ea. This in o ma ion gap is co e ed wi h an ad anced sys em called
Supe iso y Con ol and Da a Acquisi ion (SCADA). Making use o de ices and
senso s deployed ac oss he ne wo k, SCADA con inuously collec s in o ma ion on
he g id s a us. In pa icula , wi h a ypical pe iodici y o ou seconds, i collec s
da a on he sys em equency, he gene a o s and ne eal in e change be ween he
sys em and adjacen sys ems. Using his in o ma ion, he Au oma ic Gene a ion
Con ol (AGC) subsys em is esponsible o se ing ACE nex o ze o. The AGC
so wa e is un on he a ea con ol cen e o de e mine he mos eliable, s able and
economical solu ion, which basically consis s o he esou ces ha mus pa icipa e
in he es o a ion, as well as he se poin s o which de ices mus ope a e. Gene-
a o s equipped wi h AGC de ices a e in o med o he new se poin s h ough he
communica ion ne wo k, a e which hey p oceed o eadjus hei con igu a ion.
To achie e a eac i e con ol sys em, gene a o s ha pa icipa e in AGC ac ions
gene ally ha e as esponse imes and lexible p oduc ion le els. In case he ca-
paci y o he AGC subsys em is no su icien o co e he imbalance, he di ec
ac ion o human ope a o s may be necessa y so ha he sys em ope a o can phone
he gene a ion ope a o s and ask o a change in he ou pu o he powe s a ion.
When esou ce planning o demand es ima es ail so much ha ac ions o he
seconda y con ol a e no su icien o es o e he s abili y o he a ea, he e ia y
con ol comes in o play. I is in ended p ima ily o add ess con ingencies ha
equi e solu ions ha las om 15 minu es o se e al hou s. Common ac ions o he
e ia y con ol a e enabling ese es, escheduling ne in e changes o he con ol
a ea, and shedding pa s o he load when necessa y. This con ol le el is no
de ined in he same way a ound he wo ld, so some imes his is unde s ood o be
pa o he seconda y con ol. Te ia y con ol can ce ainly be seen as a so o
long- e m seconda y con ol, bu i is clea ly di e en ia ed by he ype o esou ces
i uses and he du a ion o i s ac ion.
So, as depic ed in Figu e 1.2, ene gy imbalances a e add essed g adually in
h ee le els, which a e mainly cha ac e ized by he pe iod in which hey un, and
he size o he imbalance hey can sol e: om he i s le el, which is a ge ed o
9
1. THE ELECTRICAL GRID
small imbalances ha can be co ec ed quickly, o he hi d le el, which handles
la ge imbalances ha may ake hou s o be co ec ed.
P ima y
con ol
as and au onomous,
powe s a ion con ex
Seconds
Minu es
balancing a ea con ex ,
SCADA sys em,
se e al MWs
Seconda y
con ol
Te ia y
con ol Hou s
backup gene a o s and load shedding,
hund eds o MWs
human ope a o s,
Figu e 1.2: Gene al p ope ies o he p ima y, seconda y and e ia y con ols.
Once he ma ke o he sys em ope a o has assigned gene a ing plans o powe
s a ions, he la e a e esponsible o de e mining how hei gene a ing uni s mus
mee he demand a all imes o he day. Tha is, he powe s a ion has o se which
uni s a e going o be used, a which pe iods o he day hey mus ope a e, and which
mus be hei p oduc ion le el a each momen . This ask is always app oached as
sho - e m op imiza ion algo i hms ha , beyond he essen ial pu pose o supplying
he equi ed powe , a e ocused o minimize o e all uel cos s while sa is ying he
cons ain s imposed by he sys em. Speci ically, sho - e m scheduling is ackled
h ough wo p ocesses:
Uni commi men (UC): This de e mines he ime poin s a which each gene-
a ing uni mus be s a ed up and shu down, as well as he amoun o ene gy
i should p oduce when is on-line. Uni commi men is ypically done one day
ahead.
Economic dispa ch: This de e mines he se -poin s o each o he on-line gene-
a ing uni s in o de o mee he exis en load a minimum cos . The economic
dispa ch op imiza ion algo i hm is ypically un e e y 5 o 10 minu es.
The e iciency o hese p ocesses is c ucial o la ge u ili ies: a educ ion o less
han 1% can esul in sa ings o millions o dolla s a yea .
10
1.5 Demand- esponse p og ams
1.5 Demand- esponse p og ams
Ra he han inc easing gene a ion capaci y, dec easing demand is he mos desi-
able measu e when balancing ene gy. As desc ibed in Sec ion 1.7, a signi ican
amoun o he in es men on he gene a ion, ansmission and dis ibu ion o ene gy
is d i en by he need o co e he demand in high peak imes, which in p ac ice
means less han 1% o he demand du ing he yea . Also, a oiding using gene a-
ion capaci y, especially ha ins alled o co e demand peaks, implies conside able
sa ings on ene gy sou ces ha a e expensi e and ine icien by na u e. Awa e o
hese condi ions, he elec ical g id has de eloped mechanisms o comba ise in
demand. They a e mainly wo g oups o means:
Demand- esponse (DR): This is an ac ion by which end-nodes a e encou aged
o make sho - e m educ ions in esponse o p ice signals, o as a esul o
bila e al con ac s. A e ecei ing a DR signal, ypical ac ions a e u ning o
banks o ligh ing, adjus ing HVAC le els o shedding pa o he demand o
indus ial p ocesses.
Demand side managemen (DSM): This in ol es measu es in ended o im-
p o e ene gy e iciency, which a e mainly ela ed o he unc ioning o con-
sume de ices.
DR p og ams a e one o he mos p omising mechanisms o e icien ene gy
balance. Howe e , due o he lack o mode niza ion o he elec ical g id, i s appli-
ca ion is s ill a he limi ed: i is mainly ocused on disca ding load om ac o ies
and la ge acili ies unde con ac s p e iously ag eed wi h he sys em ope a o . In
pa icula , he acili ies in ol ed, in exchange o a paymen , access o disca d pa
o hei demand unde ce ain ci cums ances and in speci ic ime pe iods. This ype
o solu ion lacks lexibili y.
Wi h he aim o b inging DR p og ams o all cus ome s, and hus ob aining
all he bene i s hey can eally o e , he communi y has long been wo king on
DR s anda ds and de ices. The OpenADR s anda d [OAD] is he mos ad anced
p oposal, o which mos popula selle s a e al eady o e ing p oduc s.
11
1. THE ELECTRICAL GRID
1.6 Libe aliza ion and ene gy ma ke s
Since he beginning o he la ge-scale gene a ion o elec ical ene gy in he la e
nine een h cen u y, up un il less han wo decades ago, he powe supply has always
been ea ed as a na u al monopoly: as a se ice ha should be p o ided by he go-
e nmen h ough u ili ies. The elec ici y sec o , due o he complexi y o bo h i s
in as uc u e and ope a ion, has always been unde s ood as a ield in which i is no
easy o success ully apply ma ke p inciples. In addi ion, he eno mous impo ance
ha elec ici y has acqui ed o economic and social de elopmen o coun ies led
go e nmen s o ake on i s managemen . Howe e , his iew o he elec ical g id,
pa ly because compe i ion is limi ed o null, b ings impo an d awbacks such as
high cos s, lack o inno a ion, and ine iciency. In con as , he heo e ical bene i s
commonly associa ed wi h he opening o elec ici y ma ke s a e: lowe p ices and
ope a ing cos s, imp o emen o he sys em e iciency and se ice quali y, os e ing
inno a ion, encou agemen o he use o clean ene gy solu ions, and inc emen o
he a ay o ene gy p oduc s a ailable o consume s. Wi h hese goals in mind, mos
de eloped coun ies ha e begun libe alizing he elec ici y sec o , as well as c ea -
ing and opening elec ici y ma ke s, which is a complex ask due o he magni ude
o he sys em, long-es ablished adi ions, and he need o con inuously p o ide a
eliable se ice.
In pa icula , he libe aliza ion o he sec o is ocused on he de ini ion and
implemen a ion o a amewo k ha sepa a es he ac i i ies ha can success ully
ope a e in compe i ion and hose which, by na u e, mus emain as na u al mono-
polies. In gene al, he guidelines ha he eme ging libe aliza ion p ocess ollows
a e:
a) De egula ing gene a ion and supply ac i i ies, and allowing he en y o p i a e
agen s in hese s ages. The idea is ha gene a o s sell ene gy in a wholesale
ma ke , and bo h e ail companies and la ge indus ial cus ome s buy i in o -
de o o e supply o cus ome s in he domes ic and comme cial sec o s a
egula ed p ices.
b) T ansmission and dis ibu ion ne wo ks emain as a na u al monopoly. The
managemen o he ne wo k is delega ed o an independen ope a o ha mus
gua an ee i s impa ial and non-disc imina o y use o gene a o s and e aile s.
12
1.6 Libe aliza ion and ene gy ma ke s
This scheme aims a b eaking he s ong e ical in eg a ion ypical o he s a e-
owned elec ical g ids, in which a single company usually akes p esence in all
s ages o he ene gy deli e y: gene a ion, ansmission, dis ibu ion and supply.
By con as , i is eplaced by a model ha pu sues o in oduce compe i ion in he
gene a ion and supply ac i i ies. Acco dingly, he sec o es uc u ing is o mally
guided by wo key lines:
i. Ve ical unbundling: A oiding he simul aneous p esence o agen s in mul iple
s ages o he gene a ion and deli e y p ocess. Tha is, o ensu e and ein o ce
compe i ion by a oiding ha a playe can be a cus ome o i sel a a sub-
sequen s age. Ac ually, i i is no p ope ly egula ed, an agen , in he ole o
e aile , may buy he same ene gy ha i p oduces as gene a o , hus possibly
dis o ing p ices.
ii. Ho izon al sepa a ion: Reduce he ho izon al concen a ion and gua an ee he
o e in he gene a ion and supply ac i i ies by s imula ing he en y o agen s
a each le el.
This model is in ended o c ea e compe i ion h ough wholesale and e ail ma -
ke s. In he o me case, e aile s and la ge indus ial consume s buy elec ici y
di ec ly om gene a o s; while in he la e case, e aile s w ap elec ici y om he
wholesale ma ke in comme cial p oduc s ha e-sell o end-use cus ome s. Due o
he c i ical na u e o elec ici y o socie y, a i s a his la e s age a e egula ed.
The main ading mechanism in e ail ma ke s a e long- e m con ac s be ween
e aile s and end-use cus ome s. Speci ically, he cus ome con ac s an elec ici y
p oduc ha , acco ding o he p o ile o he demand, gua an ees a ixed p ice o
he ene gy. In his way, end-use cus ome s a e p o ec ed om he high ola ili y
o ene gy p ices. On he o he hand, wholesale ma ke s a e based on ules and
mechanisms ypical o commodi y ma ke s. In his case, he ac ha elec ici y
canno be s o ed on a la ge scale along wi h he impossibili y o p o iding i a he
ins an cus ome s demand i , makes i necessa y o wo k wi h es ima es and hold
ma ke s in ad ance. Ac ually, his ype o ma ke in ol es mos o he wholesale
ac i i y. The e, a cen al au ho i y ini ia es a ma ke in which gene a o s and e ai-
le s, acco ding o demand es ima es, nego ia e a speci ic olume o elec ici y o
speci ic u u e blocks o ime. Acco ding o he du a ion o hese pe iods o ime,
h ee ypes o spo ma ke s a e held:
13

1. THE ELECTRICAL GRID
Day-ahead: This schedules he p oduc ion and consump ion o he nex day.
This ma ke is o ganized on an hou ly basis: playe s submi o e s o pu cha-
sing and selling ene gy o blocks o hou s o he day-ahead. Gene a o s make
o e s o he hou s hey ha e a ailable capaci y, while e aile s submi bids
o pu chasing ene gy in acco dance o he es ima ed demand. Ma ke playe s
commonly pu in 24 bids pe day.
Hou -ahead: This adjus s he de ia ions on he supply and consump ion wi h
ega d o ha ini ially scheduled in he day-ahead ma ke . Se e al ma ke
sessions a e p og ammed pe day, which a e in ended o co ec he de ia ions
o speci ic ime pe iods. The playe s a e usually he same as in he day-ahead
ma ke . This ma ke is also known as In a-day ma ke o Adjus men ma ke .
Real- ime: This ac s as he las economic le el o achie ing he balance be ween
supply and demand. I is no based on he same ma ke mechanisms ha he
day- and hou -ahead ma ke s use. Ins ead, in he eal- ime ma ke , gene a o s
and consume s submi bids ha speci y he p ices hey equi e o a y hei
supply o demand o a speci ic olume in a sho pe iod o ime. This ma ke
is also known as Balancing ma ke .
A he closing ime o day- and hou -ahead ma ke s, a e collec ing all o e s
and bids, a cen al au ho i y p oceeds o clea he p ice o each block o ime.
Clea ing algo i hms o elec ici y ma ke s a e mainly based on he ma ginal cos
o gene a ion. In e ms o ading olume, he day-ahead ma ke is he p incipal
mechanism o scheduling he ene gy dispa ch o each day. Nex , hou -ahead and
eal- ime ma ke s a e used espec i ely o co ec de ia ions om he ini ial plan
and o balance he supply and demand minu e o minu e. Fu he mo e, coun ies
ha ha e made mos p og ess in he libe aliza ion p ocess a e inc easingly in eg a -
ing ma ke s o ancilla y se ices. The aim o hese is o ade capaci y and unc ions
ha sys em ope a o s use o co e unplanned imbalances and inciden s.
Wi h he onse o elec ici y ma ke s, has also eme ged he igu e o Ma ke
Ope a o (MO), which is he en i y esponsible o he managemen o he ma -
ke s. The MO is commonly ela ed o he ISO, and is complemen ed wi h he
Ene gy Regula o (ER). This second ole is es ablished by he go e nmen and i s
p incipal mission is o ensu e ha ma ke ope a ion occu s in compliance wi h he
14
1.7 On oad o obsolescence
go e nmen ’s egula ion, paying special a en ion o he pa s o he sec o ha e-
main na u al monopolies, such as he ansmission and dis ibu ion ne wo ks. In
mos cases, he MO is in cha ge o he managemen o he day-ahead and in a-day
ma ke s, while he balancing ( eal- ime) and ancilla y se ices ma ke s a e deleg-
a ed o he ER.
1.7 On oad o obsolescence
The exis ing g id shows clea signs o obsolescence: he echnological basis, he
means o con ol, and he in as uc u es o gene a ing and deli e ing elec ici y
ha e emained unchanged o decades. This lack o inno a ion con as s wi h he
con ex , which con e sely has become pa icula ly ola ile. The huge inc ease in
demand, he con inuous luc ua ion o p ices and he ine i able deple ion o ossil
uels a e all challenges ha elec ical g ids a ound he wo ld ha e o ace in he me-
dium e m. Acco ding o es ima es, he coming decades will be c ucial o o e come
a h ea ening ho izon [IEA14, EU14, EU13]:
Exponen ial inc ease in demand. Be ween 2008 and 2035, wo ld ma ke ed
ene gy consump ion will inc ease by 53%. Al hough much o his g ow h is
associa ed wi h eme ging economies, de eloped coun ies will also expe ience
a high ise in ene gy consump ion. In pa icula , in Eu ope ene gy demand will
inc ease by 60% om now o he yea 2030.
Deple ion o ossil uels. Mos o he supply o mee he new demand will be
based on ossil uels. As a esul , i is es ima ed ha mos o he con en ional
oil ese es will be deple ed by 2030.
Lack o p ice con ol. Mos coun ies need o impo ossil uels. In addi ion,
much o he ossil uel esou ces a e con olled by a small g oup o p odu-
cing coun ies, which a e mainly ep esen ed by he O ganiza ion o he Pe -
oleum Expo ing Coun ies (OPEC) and he Gas Expo ing Coun ies Fo um
(GECF). The lack o local sou ces and he ac ha main exis ing p oduce s
a e li ing unde uns able go e nmen s leads o lack o con ol o p ices. The
Eu opean case is pa adigma ic: i he con inen is no able o inc ease i s
ene gy p oduc ion, 70% o demand will ha e o be me wi h ex e nal ene gy
15
1. THE ELECTRICAL GRID
sou ces o e he nex 20 yea s. As o he US, i will ha e o inc ease he
p oduc ion o c ude oil by 13% by 2019 in o de o comba he ise o p ices.
Global wa ming: Clima e change is a ac . The In e go e nmen al Panel on
Clima e Change (IPCC) has e idence ha g eenhouse gases ha e so a caused
a global empe a u e ise o 0′6deg ees cen ig ade. Mo eo e , he IPCC es-
ima es ha in he e en o con inuing abuse o ossil uels, his empe a u e
will inc ease be ween 1′4and 5′8deg ees du ing he wen y- i s cen u y.
Also, cha ac e is ic ac o s o he g id, such as he low e iciency o gene a ing
mechanisms, o losses due o he ansmission o ene gy o e long dis ances, mus
be e ised and co ec ed. Fo example, in ela ion o he amoun o uel used, he
exis ing gene a ing uni s only use be ween 25% and 40% o he ene gy gene a ed.
The es o he ene gy is dissipa ed in o m o hea du ing he p ocess, which is
di icul o anspo . The e iciency a io can be imp o ed i gene a ing uni s a e
nea he place o consump ion, so ha hea ejec ion is applied igh he e. In his
case, he e iciency a io may each 70%. As o he losses due o ansmission and
dis ibu ion o ene gy, he EIA es ima es ha hey each 7% in he US and 6% in
he EU.
The Augus 14, 2003, he no heas e n US and sou he n Canada su e ed one
o he wo s blackou s in his o y. I in ol ed some 50 million cus ome s, and i s
economic cos was es ima ed a be ween 7 and 10 billion dolla s. The causes we e
he combina ion o o e loading and ailu es in he con ol algo i hm. Fa om
being a single e en , blackou s wi h signi ican consequences occu e e y yea . The
ise in demand and ha ing o con ol an inc easingly complex sys em make ailu es
ine i able. Howe e , he lack o eac i i y o he g id and i s monoli hic s uc u e
causes small blackou s o become olling blackou s, hus ex ending consequences
h oughou he ne wo k.
The mos common cause o ou ages a e empo a y peaks o demand, which
a e usually ocused on a small se o summe and win e days, when as a esul o
unusual empe a u es he e is a massi e use o ai condi ioning o hea ing sys ems.
These sho pe iods o ime may ha e a equency o once e e y i e yea s o e en
once e e y en yea s, which makes ese e capaci y ex emely expensi e. A he
p esen ime, be ween 25% and 50% o he elec ici y bills o mos coun ies goes
o inance he in as uc u e in cha ge o co e ing usual peaks, which can mean
16
1.7 On oad o obsolescence
ac i i y ha occu s du ing less han 1 pe cen o yea . Mo eo e , in he cu en
ci cums ances, his scena io can only ge wo se: acco ding o he IEA, peak load
will inc ease un il 2050 by 28% in he OECD coun ies o EU, 15% in OECD
coun ies o US, and 200% in China. This may be alle ia ed i i is possible o shed
dynamically speci ic pa s o he load du ing especially demanding ime pe iods.
Howe e , in p ac ice his op ion is usually only a anged wi h indus ial and la ge
comme cial cus ome s. The e o e, he concep o eliabili y o he elec ical g id
is highly ine icien , as i gene ally assumes esponsibili y o supply e e y load,
ega dless o i s ype and impo ance. The excep ions a e he zones wi h essen ial
public se ices, such as hospi als and police s a ions, which a e especially p o ec ed
so ha hey a e usually he las o be a ec ed.
Acco dingly, he exis ing elec ical g id is becoming obsole e:
The In as uc u e, he echnological basis and he con ol scheme o he g id
ha e s ayed he same o he pas 60 yea s, wi h he esul ha hey a e no
p epa ed o deal wi h he ac i i y and complexi y ha he expec ed demand
will en ail.
Gene a ing uni s a e highly ine icien , and he ansmission o ene gy implies
signi ican losses.
The con ol sys em is highly cen alized, so i losses eac i i y and abili y o
ac as i g ows.
The eliabili y and quali y o se ice o he g id a e i ually based on he
concep o all o no hing. The need o supply ou s anding peaks o demand
cha ges he g id wi h excessi e cos s.
In iew o he ac s and o ecas s, he communi y ag ees ha he exis ing elec-
ical g id is no eady o mee he upcoming challenges, so ha he en i e concep-
ion o i , om he s uc u e o he echnologies used, mus be e ised.
17
2. DISTRIBUTED ENERGY NETWORKS
sys em ope a o , mic o-g ids beha e like any o he poin o he ne wo k, so hey a e
conside ed an e ec i e means o he anspa en and g adual in eg a ion o DER
de ices in o he g id, including enewable ene gy sou ces. Figu e 2.1 depic s he
scheme adi ionally used in he li e a u e o illus a e he mic o-g ids. As shown,
his is composed o eede s ha can be de o ed o speci ic ypes o loads. Fo in-
s ance, in Figu e 2.1, he wo uppe ones con ain loads ha , co espondingly, can
be disca ded o adjus ed o a speci ic le el o consump ion; while he bo om eede
is de o ed o c i ical loads ha he con ol sys em s i es o p ese e. By sepa a ing
he de ices acco ding o hei ype, non-c i ical loads can be disconnec ed apidly
in case o an eme gency o lack o supply. This ac ion is commonly pe o med
by he Sepa a ion De ice (SD), which is o icially esponsible o acili a ing he
ansi ion o isola ed ope a ion. Speci ically, a mic o-g id is in islanded mode when
i is isola ed om he main g id so ha i emains ope a ional and unc ional as an
au onomous en i y. Fu he mo e, all communica ions be ween he main g id and
he mic o-g id a e pe o med by a de ice called Poin o Common Coupling (PCC).
No e ha , depending on he con ex , he scena io depic ed in Figu e 2.1, ins ead
o being composed o end-use uni s, may be based on acili ies ha ope a e as
con ollable end-nodes.
As o beha io , mic o-g ids a e said o be good ci izens, which means hey a e
en i ies ha , by de ini ion, pose no isk o he ne wo k and do no add complexi y
o i s managemen ei he . In addi ion, o mo e ad anced phases, i is expec ed o
implemen he beha io model ci izen, by which he mic o-g id will also p o ide
ancilla y se ices o he main g id, by ei he injec ing ene gy when necessa y, o
limi ing i s demand when eques ed.
A a highe le el o abs ac ion, a mic o-g id is a speci ic implemen a ion o
an enclosed, au onomous a ea o he elec ical g id. In [PLSW06], he Na ional
Sandia Labo a o ies uses he concep o cell o desc ibe a simila s uc u e. In his
case, cells a e de ined as se o dis ibu ed ene gy esou ces ha a e simple enough
o be managed by a single en i y based on local p inciples. Fu he mo e, he con ol
sys em is supplemen ed by a so wa e agen ha is esponsible o in e ac ing wi h
he neighbo ing cells. Acco ding o he na u e o in e ac ions, wo o ganiza ion
models a e possible:
24

2.2 Dis ibu ed Ene gy Ne wo ks
Figu e 2.1: Basic scheme o mic o-g ids.
Glob: A ne wo k composed o cells ha , ollowing hei own in e es s, nego i-
a e ene gy exchanges among hemsel es. The con ol agen is esponsible o
nego ia ing he pu chase and sale o ene gy.
Co-op: A ne wo k composed o cells ha , besides ha ing all cha ac e is ics
o Glob cells, a e also able o coope a e wi h each o he in o de o achie e
collec i e goals.
In u n, a cell o ype Co-op, due o i s capaci y o coo dina e wi h o he cells
and pu sue common goals, can pa icipa e as an in e nal elemen in o he Co-op
and Glob cells, hus making i possible o c ea e composi e s uc u es. F om a
p ac ical s andpoin , using Co-op cells is he mos easible app oach o ackling he
de elopmen o he Sma G id, since i allows he de ini ion o goals ela ed o he
eliabili y and quali y o he ene gy supply.
On he o he hand, he EU CRISP p ojec [ECN06] uses he e m ene gy cell
[ARP+02] o e e o enclosed, sel -managed a eas o he dis ibu ion ne wo k.
In his wo k, one o he mos ep esen a i e cha ac e is ics o ene gy cells is ha
hey can be g ouped so ha he union o wo o mo e ene gy cells can make a
new cell, hus se ing up a s uc u e capable o scaling ho izon ally and e ically
(Figu e 2.2). A key di e ence be ween ene gy cells and mic o-g ids is ha uni s
o he o me can be o he cells, while mic o-g ids a e in ended o be composed o
gene a ion and consump ion en i ies.
25
2. DISTRIBUTED ENERGY NETWORKS
Agg ega o
Uni Uni Uni
Cell Z
Agg ega o
Uni Uni Uni Uni Uni Uni
Cell X Cell Y
Cell XY
Cell XYZ
Agg ega o Agg ega o
Agg ega o
Figu e 2.2: A chi ec u e based on he concep ene gy cell o he CRISP p ojec .
As an ini ial s ep o implemen he ene gy cell concep in o he elec ical g id,
he CRISP p ojec p oposes [CRI02] wo hie a chical le els:
Le el 1: A cell ha is made up o de ices belonging o one o mo e eede s o
he dis ibu ion ne wo k. The bounda ies o his ype o cells a e subs a ions.
Le el 2: A cell ha a ises om g ouping Le el 1 cells ha a e connec ed o
he same medium ol age ans o me .
2.2.2 Ene gy Managemen Sys em
As pa o i s daily ope a ion, dis ibu ed ene gy en i onmen s, ei he hey a e
mic o-g ids o ene gy cells, mus mee economic, hea load, en i onmen al and
legisla i e cons ain s. The e o e, apa om he as elec ical con ol sys ems,
hese en i onmen s equi e an in elligen global con ol sys em. This is called
Ene gy Managemen Sys em (EMS) and p ima ily aims o op imize cell’s ene gy
cos h ough planning, coo dina ing and supe ising he ac i i y o all esou ces
[KSLK03]. The EMS wo ks in he seconda y con ol sys em making sho - e m
plans based on ac o s such as: condi ions imposed by he main g id, speci ic ea-
u es o gene a ing uni s, amoun o load ha can be modula ed and shed, amoun
26
2.2 Dis ibu ed Ene gy Ne wo ks
o ene gy ha can be s o ed, ene gy p ices, cu en legisla ion, demand es ima es
and wea he o ecas s.
I should be no ed ha he EMS does no necessa ily imply he p esence o a
physical de ice. The EMS is p ima ily a concep ha may be implemen ed using
he simples me hod, such as he hand-con ol, o he mos mode n and sophis ic-
a ed ones, such as dis ibu ed sys ems based on concep s and echniques belonging
o he a i icial in elligence ield. In any case, building an EMS is ecognized as a
complex ask. In p ac ice, he EMS is mos ly implemen ed as a cen alized module
ha is pa o a hie a chical con ol s uc u e wi h h ee le els [DH05]:
1. Dis ibu ion Ne wo k Ope a o (DNO) and Ma ke Ope a o (MO): The DNO
is a managemen sys em esponsible o he ope a ion o he medium o low
ol age a ea ha he mic o-g id is connec ed o. Thus, he a ea o ac ion o he
DNO can span mul iple mic o-g ids and u ili y g ids. On he o he hand, he
MO is esponsible o he economic ope a ion o one o mo e mic o-g ids.
2. Mic o-G id Cen al Con olle (MGCC): A e ecei ing in o ma ion om he
DNO and MO, as well as om in e nal senso s and componen s, he MGCC
de elops ac ion plans and sends commands o he con ollable uni s.
3. Local Con olle (LC): Each con ollable uni o he mic o-g id has associ-
a ed wi h a LC ha is in cha ge o moni o ing i s ac i i y and applying he
commands sen by he MGCC.
In his scheme (Figu e 2.3), he EMS wo ks as an embedded module o he
MGCC de o ed o he schedule o he local uni s ope a ion. In pa icula , he EMS
is commonly p oposed as a non-linea op imiza ion p oblem [HTV+04, HAIM07]
ha includes a iables e e encing o economic ac o s and echnical cha ac e is-
ics.
Howe e , he a o emen ioned solu ion is conside ed nei he e icien no scal-
able o medium o la ge dis ibu ed en i onmen s because:
i. The compu a ional cos o inding a solu ion inc eases exponen ially wi h he
size o he model, so ha i can easily esul in a NP-Ha d p oblem.
ii. S ochas ic and nonlinea a iables ypical o ene gy uni s a e di icul o model,
so hey ha e o be simpli ied o omi ed.
27
2. DISTRIBUTED ENERGY NETWORKS
Figu e 2.3: Con ol le els o he mic o-g id en i onmen .
iii. The sys em has low eac i i y, since any change in he en i onmen demands
es a ing he op imiza ion p ocess.
i . The sys em p o ides a low le el o au onomy o use s, who a e limi ed o
exp essing hei in en ions h ough p ices o u ili y unc ions.
E en hough he e a e also wo ks based on neu al ne wo ks [CPPS06] and uzzy
logic [KDAP12, LSM09], adop ing a cen alized app oach o implemen ing he
EMS is conside ed unsui able because i g an s nei he au onomy o use s no he
eac i i y and lexibili y equi ed by he Sma G id. On he con a y, dis ibu ed
con ol solu ions i be e wi h an en i onmen like he Sma G id, which aims
o be bi-di ec ional, dis ibu ed, in elligen and eac i e. In esponse o his de-
mand, he EU CRISP p ojec pu s o wa d he Supply and Demand Ma ching
(SDM) managemen model [KCKA04], whe eby en i ies owning gene a ion and
consump ion esou ces can dynamically ba gain exchanges o ene gy blocks so
ha he ne wo k is con inuously balanced. The SDM model s ands ou o p o id-
ing au onomy o p oduce s, unlike echniques such as DSM (Demand Side Man-
agemen ) and DRR (Demand Response Resou ces), in which only au ho i y nodes
and consume s ha e capaci y o ac .
In essence, he SDM model p oposes he c ea ion o mic o-ene gy ma ke s in
dis ibu ed ene gy con ex s such as cells and mic o-g ids. On a smalle scale, hey
emula e he mechanics o wholesale ene gy ma ke s: h ough nego ia ions each
28
2.2 Dis ibu ed Ene gy Ne wo ks
node decides he amoun o ene gy i p oduces and consumes, and o how long he
ac ion is ca ied ou . Mic o-ene gy ma ke s a e concei ed as being highly eac i e,
ins an ia ed on demand, and wi h a sho ime ho izon (usually sho e han 15
minu es).
The SDM model, as well as many o he solu ions ela ed o he Sma G id, e-
qui e placing an au onomous piece o so wa e a each node, which has he mission
o : (i) ep esen ing he in e es s o use s in mic o-ene gy ma ke s; and (ii) co-
o dina ing wi h o he nodes in o de o mee collec i e goals. In elligen agen s
a e accep ed as he mos sui able echnology o add ess his challenge. Howe e ,
he lack o s able s anda ds o a no iceable pe iod o ime and he need o make
assump ions abou u u e scena ios, ha e esul ed ha many s udies ha p opose
so wa e agen s o he con ol o he Sma G id do no sha e a common ocabu-
la y. In o de o p oceed wi h ou s udy, his wo k assumes he p esence o he basic
componen s desc ibed below, which a e usually ound in he li e a u e (Figu e 2.4):
Local Agen
Local Agen Local Agen
Local Agen
Cen al Agen
Agg ega o
Agen
Figu e 2.4: Common oles ha play so wa e agen s in ene gy cells.
Local agen : This ep esen s p oduc ion, consump ion and s o age en i ies
h oughou he managemen p ocess. The main asks o local agen s a e o ne-
go ia e on behal o cus ome s, sending commands o he local de ices, mon-
i o ing hei ac i i y, and sending upda ed in o ma ion o he au ho i y nodes.
29

2. DISTRIBUTED ENERGY NETWORKS
Local agen s a e usually planned o be un in he con ol de ice o he cus-
ome ’s acili y.
Cen al agen : I is a sys em agen ha , depending o he le el o decen aliza-
ion o he solu ion, is in cha ge o supe ising and/o con olling he ope a ion
o he mic o-g id o ene gy cell. This ole is essen ially he agen -based e -
sion o ha o cen al con olle in he adi ional app oach.
Agg ega o agen : I manages he in e ac ions o he cell wi h he main g id
and o he ex e nal en i ies, which in u n can be o he cells o mic o-g ids.
The pa icipa ion o he cen al agen in he clea ing p ocess a ies acco ding o
he decen aliza ion le el o he solu ion. Th ee main asks a e dis inguished along
he s a e o he a :
Managemen : The cen al agen is esponsible o de eloping and con olling
he ac ions plans o all he en i ies. These plans a e ca ied ou acco ding o
he in o ma ion sen by he local agen s ega ding bo h he s a e o he uni s
unde con ol and he use s’ p e e ences.
Supe ision: The cen al agen moni o s he ac i i y o he local agen s, which
in his case a e he en i ies esponsible o d awing up he ac ions plans. The
cen al agen may e use o in e cede on bo h he plans and objec i es in o de
o ensu e ha condi ions ela ed o he e iciency, secu i y and eliabili y a e
me .
Se ices: The ac i i y o he cen al agen is limi ed o p o iding suppo
h ough unc ions and da a se ices. In his ega d, FIPA p o ocols [FIP96]
de ine se o se ices in ended o acili a e ypical asks in mul i-agen sys-
ems, such as loca ing and egis e ing agen s. Also, in o de o make eliable
plans, local agen s will likely equi e da a se ices such as demand es ima es
and wea he o ecas s. As o he business logic, hey a e also necessa y unc-
ions ha con ol he ma ke ’s li e-cycle.
Fu he mo e, ega dless o he ype o app oach, he cen al agen is commonly
p oposed o eco d he ac i i y o he sys em’s componen s, and con i m ha local
agen s beha e in acco dance wi h he ag eed plans and goals.
En i onmen s in which he cen al agen pe o ms he clea ing p ocess a e es-
sen ially cen alized solu ions [HDT+05, OJ05, KWK05, FTNY08], so ha he
30
2.3 S anda ds o he ene gy managemen
unc ions o he cen al agen a e p ac ically iden ical o hose o he Mic o-g id
Cen al Con olle (MGCC, [HTV+04]), which is a physical de ice designed o
ake on he en i e con ol o he mic o-g id. When he cen al agen wo ks as su-
pe iso [DH04, DH05, A n00], he solu ion gains in decen aliza ion, as i a ises
om he in e ac ion o local agen s. Howe e , in his case he ou come s ill e-
qui es he app o al o he cen al agen , which may be p og ammed o look ou
o pa ame e s such as g id s abili y, powe quali y, supply secu i y and e iciency.
Finally, when he asks assigned o he cen al agen (when necessa y) a e limi ed
o p o iding ancilla y se ices, he solu ion can be conside ed ully decen alized
[AB00, RPT07, PFR09, LKG05, Jia06, BCG+98, PLSW06].
Much o he li e a u e ends o include he agg ega o agen as a subsys em o
he cen al agen . This wo k ep esen s hese wo igu es sepa a ely hough because
hey ac ually wo k in well-di e en ia ed unc ional a eas ha , due o hei comple-
xi y, equi e indi idual analysis. As men ioned be o e, he agg ega o is in cha ge
o managing he in e ac ions o he cell wi h he con ex . The agg ega o may e-
cei e ins uc ions om he sys em ope a o , sends in o ma ion o i abou he local
de ices, and manages he exchange o ene gy wi h he main g id and su ounding
cells. The p esence o he agg ega o agen is common in en i onmen s in which
local agen s a e able o coo dina e and coope a e wi h each o he . In e nally, he
agg ega o communica es all he in o ma ion and ins uc ions o he cen al agen ,
which is esponsible o p ocessing hem.
To sum up, he EMS, when implemen ed in a dis ibu ed manne , is a mul i-
agen sys em in which so wa e agen s, ep esen ing local nodes, in e ac and co-
o dina e be ween hem in o de o balance he sys em and accomplish bo h pa icu-
la and collec i e goals. This ype o implemen a ion shows ha , indeed, he EMS
is mo e concep ual han physical, since he managemen he e a ises as esul o he
communica ion and coo dina ion o independen so wa e agen s.
2.3 S anda ds o he ene gy managemen
The mos impo an e o o s anda dize he Sma G id comes om he Na ional
Ins i u e o S anda ds and Technology (NIST). This aims o guide he de elop-
men o a amewo k ha includes s anda ds o sys ems, de ices and p ocedu es
31
2. DISTRIBUTED ENERGY NETWORKS
[NIS12]. In o de o suppo NIST in his ask, he Sma G id In e ope abili y
Panel (SGIP) was es ablished in la e 2009, which, in collabo a ion wi h ex e nal
o ganiza ions, aims o de ine equi emen s o essen ial communica ion p o ocols
and o he common speci ica ions.
As a i s s ep in he ace o he Sma G id, SGIP has iden i ied key a eas o
which s anda ds should be de eloped. Fu he mo e, a small g oup o hese s and-
a ds was conside ed highly impo an , being classi ied as P io i y Ac ion Plans
(PAP). The PAP09 is speci ically de o ed o he de elopmen o DR p og ams,
hus ecognizing he impo ance o his a ea in he sho - e m u u e o he Sma
G id. Much o he wo k de o ed o his plan has ocused on he de ini ion o he
OpenADR s anda d, which is suppo ed by he in o ma ion and communica ion
model desc ibed in he Ene gy In e ope a ion s anda d de eloped by he collabo -
a ing o ganiza ion Ad anced Open S anda ds o he In o ma ion Socie y (OASIS).
The basic concep s o bo h s anda ds a e b ie ly desc ibed below.
2.3.1 The Ene gy In e ope a ion s anda d
The goal o he Ene gy In e ope a ion (EI) s anda d om OASIS is o de ine mes-
sages o communica ing p ices, eliabili y and eme gency condi ions. Fo mally,
he s anda d is said o desc ibe “an in o ma ion and communica ion model o co-
o dina e ene gy supply, ansmission, dis ibu ion, and use, including powe and
ancilla y se ices, be ween any wo pa ies, such as ene gy supplie s and cus o-
me s, ma ke s and se ice p o ide s” [OASa]. I is impo an o highligh ha ,
in he a chi ec u e de ined in he EI s anda d: (i) in e ac ions a e always possible
be ween any pai o ac o s; and (ii) an ac o can pa icipa e in many in e ac ions a
he same ime. The s anda d adop s a se ices-o ien ed app oach and is agnos ic
in ela ion o he echnology used o ca y he messages. As o he local de ices,
acili ies mus be p o ided wi h communica ion in e aces such as ha desc ibed
in [Hol09]. Speci ically, he poin o communica ion whe eby nodes o e and con-
sume se ices is he ESI.
The in o ma ion and communica ion model de ined in he EI s anda d is in en-
ded o acili a e collabo a ion in ene gy use. Collabo a i e Ene gy s ands o he
managemen o ene gy using coope a i e mechanisms. In addi ion, when he e a e
32
2.3 S anda ds o he ene gy managemen
ma ke in e ac ions, he managemen model is e e ed o as T ansac i e Ene gy.
In his scheme, pa ies buy and sell ene gy using ende s ha , i accep ed, esul in
ansac ions (Figu e 2.5). In a ansac ion, a pa y can ake on he ole o buye o
selle . No mally, a gene a o will be on he selle ’s side o he ansac ion, and an
end-use cus ome on he buye ’s one; al hough no hing p e en s hem om swap-
ping hese oles. As o he nego ia ion p ocess, he ende ha ini ia es he ans-
ac ion can be sen by any o he pa ies.
Figu e 2.5: Pa ies in e ac ing using ende s and ansac ions in he EI s anda d.
Apa om he T ansac i e Ene gy model, he EI s anda d also de ines a s uc-
u al model o in e ac ions ypical o DR p og ams, which consis s o e en -based
dispa ch o esou ces. The model is p incipally based on he de ini ion o wo
oles: Vi ual Top Node (VTN) and Vi ual End Node (VEN). A VTN can in e ac
simul aneously wi h many VENs, while VENs a e no allowed o in e ac di ec ly
among hemsel es. As in any in e ac ion o he EI s anda d, pa ies may pa icip-
a e in many in e ac ions concu en ly. In his case, a node may implemen bo h
in e aces, playing he ole o VTN in some in e ac ions, and he ole o VEN in
o he s.
In he common use case, VTNs a e in ended o be au ho i a i e nodes, such as
he DSO o he Mic o-g id Ope a o , while VENs a e in ended o ep esen gene-
a ion and cu ailmen esou ces. Thus, VTN nodes usually send DR signals and
eques s o in o ma ion o VENs. The nodes ha implemen bo h in e aces a e
usually agg ega o s.
Figu e 2.6 illus a es how he combina ion o pai wise in e ac ions o VTNs and
VENs enables he implemen a ion o complex s uc u es. The g aph could model
a DR e en ini ia ed by he sys em ope a o , which in his case is ep esen ed by
he node A. Ini ially, he e en is sen o he i s -le el nodes B and C, which wo k
33
2. DISTRIBUTED ENERGY NETWORKS
hus aising hei p ice and making hem mo e di icul o main ain. This condi ion
is clea ly undesi able, since hese de ices a e in ended o be ins alled massi ely.
In pa icula , h ee impo an da a se ices ha e cons an ly been p oposed o he
co ec unc ioning o he Sma G id:
Demand es ima es: These a e necessa y in o de ha he sys em ope a o
knows when cu ailmen e en s a e necessa y, and likewise nodes know how
much demand hey mus shed o shi . To supply his ype o se ice im-
plies ha ing da abases ha con ain in o ma ion abou each use ’s consump ion
h oughou he yea .
Wea he o ecas s: Consump ion depends highly on ac o s such as a mo-
sphe ic empe a u e. In win e , wa e -hea e s and HVAC sys ems ep esen
an impo an sou ce o consump ion, while in summe ai condi ioning sys-
ems a e mos impo an . The e o e, o ob ain accu a e demand es ima es, i is
necessa y o ha e accu a e wea he o ecas s.
Ene gy p ices: The demand o ene gy, as well as he eac ion o DR e en s,
may depend on he ene gy p ices. I i is possible, use s may be willing o
con igu e hei consump ion le el acco ding o p ice le els.
Mul i-agen sys ems a e a di icul ma e . Al hough in elligen agen s a e con-
inuously being p oposed o he implemen a ion o dis ibu ed managemen sys-
ems, he u h is ha , in p ac ice, de elope s and esea ches end o choose mo e
p ac ical solu ions o eal cases. Ac ually, despi e he high numbe o esea ch
s udies ha p opose so wa e agen s o he managemen o dis ibu ed i ual en-
i onmen s such as g id compu ing and P2P ne wo ks, a he p esen ime solu ions
based on in elligen agen s a e no widely adop ed. In ac , he p ac ical applica ion
o agency heo y is mainly ocused on he domain o p ocesses and se e applica-
ions, while solu ions which connec agen s wi h use s a e limi ed (see 3.1.2). This
lack o success is pa ly due o he complexi y ha a ises om solu ions based
on agen s, and he absence o models adap ed o he eal habi s o use s, who a e
inc easingly demanding anspa en and simple solu ions ha a oid echnological
de ails.
40

2.5 Lessons lea ned om simila ields
2.5 Lessons lea ned om simila ields
2.5.1 Pee - o-pee ne wo ks
In pa icula , Pee - o-pee (P2P) ne wo ks [ATS04] aim o acili a e he exchange
o esou ces be ween pee s ha , in heo y, can be conside ed equals in e ms o
unc ionali y. Thei implemen a ion has adi ionally been ocused on exchanging
iles, Naps e ,Gnu ella and eDonkey being he bes -known cases. All hese ne -
wo ks include ancilla y se ices ha acili a e p ocesses such as he in e connec-
ion o pee s, he sea ch o esou ces and he classi ica ion o con en s.
One ac o ha has p o en decisi e in he success o P2P ne wo ks is he opo-
logy (Figu e 2.10). In his ega d, he e a e h ee main op ions [Sch01]:
Cen alized: This is he simples scheme. Pee s connec o cen alized se e s
in o de o access special unc ionali ies. Among hem a e he sea ch o e-
sou ces, and use egis a ion. No e ha esou ces ( iles in mos o he cases)
a e s ill exchanged di ec ly be ween pee s. The mos ep esen a i e example
o his ype o ne wo k is Naps e .
Decen alized: All se ices, including he egis a ion o new pee s and he
sea ch o con en s, can be ca ied ou in each node. The implemen a ion o
hese asks is pe o med by sending eques messages o he closes neighbo s.
F om he e, messages a e ecu si ely p opaga ed un il eaching a maximum
i e a ion dep h. The mos ep esen a i e ne wo k using his opology is Gnu-
ella.
Hyb id: This opology uses special nodes, called supe nodes, ha , o a limi-
ed sec ion o he ne wo k, wo k ou as en y poin s o use s, indexing all hei
con en s, and p ocessing sea ch eques s. Supe nodes a e connec ed be ween
hemsel es so ha hey can exchange in o ma ion abou he ne wo k and i s
con en s. No e ha , by sha ing his in o ma ion, supe nodes a e able o sea ch
o con en s in he whole ne wo k. eDonkey is he mos bes -known imple-
men a ion o his ype o ne wo ks.
The main d awback o he cen alized opology is ha ing a single poin o ai-
lu e, hus being oo ulne able o a acks and p osecu ion. As a case in poin , i
ook only one day o he au ho i ies o shu down Naps e . Al e na i ely, he e a e
41
2. DISTRIBUTED ENERGY NETWORKS
Cen alized Decen alized Hyb id
Figu e 2.10: Main ypes o P2P ne wo ks.
decen alized ne wo ks. Howe e , he lack o a ull index o a ailable esou ces in
he la e ype has p o en o make he sea ching p ocess ine ec i e. In addi ion,
hese ypes o ne wo ks ha e p o en o be ha d o scale and main ain. In o de o
o e come hese d awbacks, in ne wo ks such as Gnu ella, nodes ha e eme ged ha
a e able o handle la ge numbe s o connec ions and ake on special unc ionali -
ies on behal o o he nodes, such as inding esou ces. In p ac ice, his app oach
makes Gnu ella esemble hyb id ne wo ks, since special nodes beha e much like
supe nodes. As a ma e o ac , as shown in Figu e 2.11, he ac ual opology o
Gnu ella is simila o a hyb id one, hus ein o cing he hesis o he la e app oach.
Ac ually, he hyb id opology is he one mos widely used in p ac ice, ha ing many
success ul implemen a ions, and also ha ing p o en o be he mos e icien o ex-
changing esou ces. I s success elies la gely on he assump ion ha all nodes a e
no equal: hey a e no ac ually pee s, since in p ac ice hey ha e di e en cha ac e-
is ics, including compu ing powe , bandwid h and quali y o se ice. The e o e, i
is na u al ha , in o de o imp o e he o e all sys em pe o mance, he e a e some
nodes ha ha e o ake mo e esponsibili ies han o he s.
As o he Sma G id, since he OASIS and NIST s anda ds lea e he doo
open o he ins alla ion o nodes wi h di e en p o iles, i is ad isable o s udy he
bene i s ha may a ise om he ins alla ion o nodes which a e mo e powe ul han
hose en isioned so a .
42
2.5 Lessons lea ned om simila ields
Figu e 2.11: S uc u al pa e n o he Gnu ella ne wo k.
2.5.2 G id compu ing
The aim o a compu a ional g id is o c ea e he image o a powe ul compu e
h ough he in e connec ion o he e ogeneous in e connec ed sys ems [FKT01].
This goal is e y simila o ha pu sued by he Sma G id, which s i es o build
a la ge gene a ion sys em om he join p oduc ion o dis ibu ed, small ene gy
esou ces.
The i s compu a ional g ids we e ad-hoc solu ions implemen ed om sc a ch.
As a esul , hese we e di icul o eplica e in o he a ge en i onmen s. The se-
cond gene a ion was cha ac e ized by he c ea ion o amewo ks and ools ha
acili a ed he implemen a ion o compu a ional g ids, as well as an applica ion eco-
sys em a ound hem. Among he bes -known amewo ks we e Legion [GWTLT97],
Condo [TTL05] and G idbus Toolki [FK97]. Howe e , ea ly e sions o hese
amewo ks we e monoli hic, ha d o scale and wi h li le capaci y o connec o
ex e nal middlewa e laye s. As a esul , many island g ids eme ged in he US wi h
a chi ec u es ha mus be de ined as oo speci ic as hey we e p incipally in ended
o inding p ac ical solu ions, neglec ing impo an ea u es such as scalabili y and
in e ope abili y. As a ma e o ac , in e e ence o he oughness o he solu ions,
his s age is commonly desc ibed as “big i ons and a pipes” [GDR04].
The hi d gene a ion o compu a ional g ids was bo n emb acing he se ices
o ien a ion [Fos05]. This eplaced he concep o esou ce wi h ha o se ice,
wi h he esul ha nodes o he g id ac ually o e and consume se ices ha mask
esou ces. Also, in o de o inc ease in e ope abili y, his new gene a ion o com-
43
2. DISTRIBUTED ENERGY NETWORKS
pu a ional g ids p omo es he adop ion o open s anda ds. In his ega d, he Globus
Toolki eam and IBM con ibu ed o he c ea ion o Open G id Se ices A chi ec-
u e (OGSA), which is conside ed he de ac o s anda d o he implemen a ion o
compu a ional g ids [FKT04].
In addi ion, as occu s in he Sma G id, he e is a school o hough ha ou s
he i ues o using ma ke -based mechanisms as managemen sys em. This ap-
p oach is known as g id economy [BAV05], and p oposes o swi ch o a model in
which clien s a e au onomous en i ies who y o de end hei own in e es s and
goals h ough nego ia ion sys ems. This app oach can be de ined as use -cen ic,
while he adi ional model ha looks o imp o ing he global e iciency o he
sys em is de ined as sys em-cen ic.
Compa ed o o he echnological ields ha ace he challenge o sha ing and
coo dina ing dis ibu ed esou ces, he g id compu ing communi y has been p aised
o i s abili y o achie e alid solu ions. Al hough hese solu ions a e o en de-
sc ibed as being igid, i is also ue ha a di ec app oach has p o en o be e ec i e
o achie ing ope a i e sys ems.
The g id compu ing communi y has also s udied he bene i o in eg a ing so -
wa e agen s h oughou he a chi ec u e [FJK04]. These a e mainly p oposed o
p o ide lexibili y and au oma e he managemen asks, and ac on behal o use s
in ma ke -based en i onmen s. Howe e , he u h is ha he p esence o so wa e
agen s in eal sys ems is sca ce, possibly because his ype o solu ion adds a new
le el o complexi y in so wa e de elopmen , equi ing knowledge o he ield o
a i icial in elligence.
In conclusion, he mos impo an lessons lea ned om he g id compu ing ex-
pe ience a e o:
i. Emb ace a se ices-o ien ed app oach;
ii. In ensi y e o s dedica ed o he de ini ion and adop ion o s anda ds;
iii. Reach a comp omise o e he need o ind p ac ical solu ions; and
i . De o e mo e esea ch e o s o achie ing solu ions based on so wa e agen s.
44
2.5 Lessons lea ned om simila ields
2.5.3 Vi ual o ganiza ions
In he business wo ld, en e p ises a e also expe iencing challenges ha , in essence,
a e simila o hose add essed by he compu a ional g ids and he Sma G id. In
pa icula , due o he g owing end owa ds specializa ion, business oppo uni ies
a e inc easingly ul illed by empo a y coali ions o en e p ises ha coope a e and
sha e knowledge, esou ces and compe ences. This ype o coali ion is known as
Vi ual En e p ise (VE); a concep ha usually a ises when indi idual en e p ises
do no ha e he esou ces o achie e a speci ic goal ac ing on hei own, o o do
so p o i ably [MFPF01]. In o de o ake ad an age o his ision and lea n om
he expe ience gained in his ield so a , he second gene a ion o compu a ional
g ids s a ed using he concep o Vi ual O ganiza ion (VO), which aims o apply
he concep o VE in en i onmen s ha a e essen ially echnological. Speci ic-
ally, VOs a e de ined as empo a y coali ions o dis ibu ed en i ies ha collabo a e
and sha e esou ces o mee global and indi idual goals making in ensi e use o
new in o ma ion and communica ion echnologies [NT07]. This app oach a ose in
esponse o en i onmen s ha a e inc easingly changing, agile and dis ibu ed, in
which pa ne s look o alliances ha help hem o achie e new goals, inc ease hei
compe i i eness and educe isks. Socie ies ha a e classi ied as VOs commonly
sha e he ollowing p ope ies:
They a e speci ically c ea ed o mee ing a empo al business oppo uni y.
They ha e a s ong dependence on ICT.
They do no equi e he pa ne s o be colloca ed in o de o ca y ou he
assigned asks.
They a e capable o adap ing hei s uc u es o he needs o he con ex .
They make an in ensi e use o coope a ion mechanisms in o de o achie e he
de ined goals.
They a e composed o au onomous en i ies ha , besides pu suing global goals,
s i e o mee hei own goals.
Dis ibu ed ene gy ne wo ks, pa icula ly when he managemen sys em is based
on a dis ibu ed mechanism, mee hese p ope ies: hei ac i i y is en isioned as
being suppo ed by mul iple au onomous uni s ha coope a e in o de o gua an-
ee global goals (such as eliabili y and secu i y o supply) and pa icula goals
45

2. DISTRIBUTED ENERGY NETWORKS
(such as exchanging ene gy in a p o i able way) by using he la es in o ma ion and
communica ion echnologies. The e o e, dis ibu ed ene gy en i onmen s can be
concep ually conside ed VOs, hus being in good posi ion o lea n om impo an
unde akings in his ield in ecen yea s. Howe e , in di e ence o he a en ion
paid o g id compu ing, he li e a u e seems o ha e igno ed his impo an example.
Speci ically, he expe ience o VOs wa ns us ha he au oma ion o he en i e
li e cycle o a VO is a complex ask, which in p ac ice equi es speci ic solu ions,
and usually he in e en ion o human ope a o s. This expe ience he e o e shows
ha , despi e he ema kable p og ess in compu e echnology, he c ea ion o dis-
ibu ed i ual en i onmen s inhabi ed by au onomous en i ies is a di icul ask
which a p esen equi es he implemen a ion o ad-hoc solu ions, e en he supe -
ision o human ac o s. In pa icula , he main challenge aced by VOs is he
implemen a ion o he c ea ion s age [CM06], which mus accomplish asks such
as: oppo uni y iden i ica ion, ac ion plan designing, sui able pa ne s selec ion and
asks assigna ion. In addi ion, he Sma G id poses ypical challenges o open and
eac i e en i onmen s, such as communica ion and coo dina ion be ween he e o-
geneous agen s, and he implemen a ion o us mechanisms ha help o a oid he
isk ha he p esence o agen s wi h unknown epu a ion c ea es.
To o e come he complexi y o he c ea ion s age, VO esea ches ha e p o-
posed c ea ing a specialized en i onmen called Vi ual B eeding En i onmen (VBE),
which is a s able limi ed clus e composed o well-known and capable pa ne s ha
main ain long- e m ela ionships [CMA03]. A VBE imposes on he pa ne s he
use o common echnological in as uc u es, on ologies, communica ion p o o-
cols and social con en ions. Fu he mo e, a VBE au ho i y ce i ies he skills o
each pa ne , hus p o ing ha i is sui able o being pa o VOs in he u u e.
All hese condi ions a e in ended o con igu e a sa e, eliable and no malized en-
i onmen ha acili a es he dynamic ins alla ion o VOs. As o i s d awbacks, i
mus be no ed ha a VBE is a semi-closed en i onmen ha , o some ex en , lacks
lexibili y and es ic s pa icipa ion.
NIST and OASIS s anda ds co e some o he ea u es equi ed o VBEs. They
de ine he a chi ec u e, he communica ion p o ocols and he echnologies o be
used, including all issues ela ed o he secu i y o he ne wo k. Howe e , in o de
o achie e a ully ope a i e sys em based on au onomous, sel -in e es ed agen s,
46
2.5 Lessons lea ned om simila ields
i is s ill necessa y o es ablish he nego ia ion algo i hms and social con en ions
h ough which agen s mus beha e unde no mal and excep ional si ua ions. In his
ega d, VBE expe ience shows ha he mo e de ined and limi ed he con ex is, he
easie i is o implemen an e ec i e solu ion. He e, he challenge o he Sma
G id communi y is o es ablish a well-de ined amewo k ha also p ese es he
au onomy o so wa e agen s and p omo es he pa icipa ion.
47
En mi la go a o con el ma ap end´
ı que lo m´
as na u al
del mundo son los cambios.
“La ob a”, Adol o Bioy Casa es.
Qui´
en sabe. A m´
ı me pa ece que los peces ya no quie en
sali de la pece a, casi nunca ocan el id io con la na iz
[. . . ] Ches o hab´
ıa hablado de pece as con un abique
m´
o il que en un momen o dado pod´
ıa saca se sin que el
pez habi uado al compa imen o se decidie a jam´
as a
pasa al o o lado. Llega has a un pun o del agua, gi a ,
ol e se, sin sabe que ya no hay obs ´
aculo, que bas a ´
ıa
segui a anzando.
“Rayuela”, Julio Co ´
aza .
CHAPTER
3
Agency Se ices
As desc ibed in Chap e 2, he ene gy managemen o DENs is o en en isioned
in he o m o a mul i-agen sys em. Unde his app oach, end-nodes a e ep esen-
ed by so wa e agen s in local ene gy ma ke s in which hey plan and conduc he
ac ion o he p oduc ion and consump ion uni s. The o igin o his app oach is
based on he heo e ical p ope ies o in elligen agen s, which a e o mally des-
c ibed as en i ies capable o p o iding au onomy, in elligence and eac i i y in
dis ibu ed en i onmen s. Howe e , he p oposals based on his idea ha e gi en
li le a en ion o he ac ha in elligen agen s, con a y o he eno mous expec-
a ions buil up abou hem o mo e han en yea s now, ha e ac ually had li le
p ac ical impac on echnological a eas ha also seemed sui able o hem. Some
ema kable examples o hese a eas a e compu a ional g ids, P2P ne wo ks and he
mul iple ypes o i ual socie ies c ea ed a ound In e ne .
In his ligh , his chap e i s discusses he limi ed success o in elligen agen s
in he p ac ical ield. To his end, he ques ion “Whe e a e all he in elligen
agen s?”, which was ecen ly pu o he communi y by an au ho i y on he sub-
jec , is used as s a ing poin . The u h is ha he complexi y o ypical a i icial
in elligence solu ions, oge he wi h he lack o knowledge on he subjec , poses
insu moun able ba ie s o eams acing mul idisciplina y challenges. As a ma -
e o ac , au ho s usually p opose solu ions in which cus ome s mus pe o m
49
3. AGENCY SERVICES
To shed mo e ligh on he pu pose o he AS model, an example based on eBay
is illus a ed. In pa icula , eBay can de ine in e aces h ough which ex e nal so -
wa e agen s can bid on o e s, and ecei e in o ma ion abou s anding bids and
deadlines. In his case, ASPs au ho ized by eBay can be con ac ed by use s in
o de o e ec i ely au oma e hei pa icipa ion in he auc ions, hus sa ing hem
om ha ing o de elop hei own solu ions. ASPs also mean a gua an ee o eBay,
since hey may impose a minimum se o ules and social con en ions ha ensu e
he p ope ope a ion o he si e. This example, whe e he business si e is eBay and
he i ual en i onmen s a e auc ions, can be easily applied o mo e challenging
echnological ields whe e so wa e agen s a e called in o play an impo an ole,
such as g id compu ing and he Sma G id. In hese cases, in addi ion, a ole such
ha played by he local agen gains impo ance because i has o apply he ac ions
de e mined by he b oke agen on he local esou ces.
No e ha no hing p e en s he use om ha ing mo e han one local agen
assigned a he same ime. Thus, a local agen can au oma ically in o m he b oke
agen abou he s a e o he esou ces, meanwhile he use may use ano he local
agen ins alled in his/he mobile phone o bo h ecei ing in o ma ion om he
b oke and, i necessa y, sending i new di ec i es. The e o e, he b oke agen ,
apa om pa icipa ing in i ual socie ies, can also wo k as a p oxy agen able o
communica e local agen s wi h each o he (Figu e 3.2).
The main s ages h ough which clien s go in he AS model a e:
Regis a ion: The clien egis e s wi h an ASP ha has been p e iously ce i-
ied as eliable by he a ge business si e. The main aspec s o he ag eemen
a e he du a ion o he se ice, he p ocesses in which he clien wan s o pa i-
cipa e, and he con igu a ion o he b oke agen ha will ep esen he clien .
Con igu a ion: I necessa y, he clien ins alls local agen s in i s de ices and
con igu es he b oke ing se ice h ough so wa e applica ions.
Expi a ion: Once he con ac expi es, he ASP suspends he pa icipa ion o
he clien in he i ual socie ies o he business si e.
The use can se he p e e ences o he b oke ing se ice h ough web in e aces
o mobile applica ions, bo h p o ided by he ASP. This in o ma ion is o wa ded
o he b oke agen when i is ins an ia ed. Thus, in se ings wi h local esou ces,
56

3.3 Agency Se ices
local agen local agen
local agen
b oke agen
Figu e 3.2: B oke agen wo king as p oxy o mul iple local agen s.
he e is no need o embedded in e aces in o de ha he use communica es wi h
local agen s and esou ces.
Once he clien has con ac ed he se ices o an ASP, i s pa icipa ion consis s
o h ee main s ages:
Ini ia ion: The business si e in o ms he ASPs ha a new business p ocess has
been ini ia ed. Each ASP deploys a b oke agen o each clien . I necessa y,
he ASP au oma ically upda es he local agen s’ so wa e. The b oke agen
ells he local agen s a new nego ia ion p ocess has begun and, i any, he local
agen s epo on he s a us o he local esou ces.
Execu ion: B oke agen is egis e ed wi h he i ual en i onmen and, acco -
ding o he s a e o esou ces and he use p e e ences, i in e ac s wi h o he
b oke agen s de eloping ac ion plans, coo dina ion asks and nego ia ions.
Th oughou his p ocess, he b oke agen may pe iodically in o m he local
agen s abou i s pa icipa ion in he i ual en i onmen . Mo eo e , when he e
a e local esou ces, he b oke agen ansmi s he ac ions o be applied on
hem. Also, he local agen s in o m he b oke agen o new local e en s and
new di ec i es (de ined by he use ).
Close: The b oke agen eco ds he de ails o i s pa icipa ion in he i ual
socie y, and ells he local agen s ha he p ocess has been comple ed.
57
3. AGENCY SERVICES
I is possible o build a simple e sion o he AS model. Speci ically, he oles
o business si e and ASP can be joined in a single node (Figu e 3.3), so ha bo h
he i ual en i onmen s and he b oke agen s a e p o ided by he same en i y.
Al hough in doing so some le el o compe ence is los , his solu ion s ill main ains
use ul ea u es, since he ASP can s ill o e use s he possibili y o con igu ing
how he b oke agen mus beha e. Con inuing he example based on eBay, in
his simpli ied e sion o he AS model, eBay may p o ide so wa e agen s ha
use s con ac in o de o au oma e hei pa icipa ion in auc ions. In his case,
eBay would simul aneously wo k as business si e and ASP. In o de no o lose he
au onomy and independence ha in elligen agen s a e supposed o p o ide, eBay
may allow use s o con igu e he beha io o he so wa e agen s h ough di ec i es.
Fu he mo e, eBay may o e ad anced beha io s in exchange o mo e expensi e
a es.
Local
Agen
B oke
Agen
B oke
Agen
ASP / Business Si e
Local
Agen
Cloud
Clien
Local de ices
Vi ual
En i onmen
Figu e 3.3: Simpli ied e sion o he Agency Se ices model.
3.3.2 Technologies
The aim o his sec ion is o p o e he echnical easibili y o he p oposal. To
his end, ac ual echnologies ha add ess he majo challenges o he AS model
a e p esen ed. Howe e , i is no ed ha using o he echnologies is also possible.
The wo main challenges ha he AS model aces a e he abili y o: (i) conduc
58
3.3 Agency Se ices
asynch onous, bidi ec ional dialogues be ween emo e so wa e agen s; and (ii) dy-
namically deploy so wa e agen s in emo e de ices. Well-known echnologies ha
success ully sol e hese challenges a e:
The Ex ensible Messaging and P esence P o ocol (XMPP) o he communica-
ion be ween he b oke and local agen s. This is an ins an messaging p o ocol
[XMP] based on XML ha suppo s secu e communica ions. Al hough XMPP
is usually associa ed wi h applica ions such as Jabbe and Google Talk, i was
ac ually designed o communica ion be ween agen s, whe he hey a e hu-
man o so wa e. As a ma e o ac , he e a e al eady solu ions using XMPP
o encapsula e and send FIPA messages [GPA02]. As o he in as uc u e,
XMPP needs an ins an messaging se e , which can be ins alled in he ASP
in as uc u e.
Ja a Ne wo k Launching P o ocol (JNLP, [O a00]) o ans e ing and laun-
ching he local agen om he ASP. JNLP is a p o ocol o downloading and
launching emo e Ja a applica ions. I is a ma u e and widely used echnology
ha ensu es he la es a ailable e sion o he so wa e package is launched.
Fu he mo e, i uses digi al ce i ica es o gua an ee he au hen ici y and in e-
g i y o he applica ion.
In gene al, he use o ins an messaging p o ocols p o ides an easy and e ec i e
way o communica e emo e so wa e agen s (as he in e ac ion be ween b oke and
local agen s equi es), hus a oiding he need o using mo e complex mechanisms,
such as hose based on s a ic IP add esses and web se ices. In pa icula , web se -
ices a e no a easible echnology o his goal because i equi es ha one o he
wo nodes ins alls a web se e , and does no p o ide asynch onous, bidi ec ional
communica ions. On he o he hand, JNLP ensu es ha he clien can emo ely ins-
all and launch local agen s in a anspa en manne , hus conse ing he simplici y
ha cha ac e izes he cloud compu ing model. Fu he mo e, bo h XMPP and JNLP
consume ew esou ces so ha hey can be used in embedded sys ems and mode n
de ices, including mobile phones and able s.
The communica ion mechanism o he in e ac ion o b oke agen s wi h i ual
socie ies is de ined by he business si e and is anspa en o he cus ome s, as i is
an issue aken on by he ASP. The mechanism can be based on speci ic agen -based
59
3. AGENCY SERVICES
amewo ks. In any case, his poin does no ep esen a echnological isk as bo h
business si es and ASPs a e supposed o be echnological companies wi h su icien
knowledge and esou ces.
3.3.3 Bene i s and i ues
One o he majo goals o he AS model is o answe he challenges aced by in e-
lligen agen s in i s ques o become a mo e accessible echnology. In his ega d,
as desc ibed, when so wa e agen s (no jus speci ic unc ions o hem) a e o e ed
as cloud se ices:
i. The complexi y ha en ails de eloping so wa e agen s able o pa icipa e in
i ual en i onmen s is delega ed o hi d-pa y se ices p o ide s. In addi-
ion, he need o upda ing agen s in o de o imp o e hei pe o mance o o
adap hem o bo h he in e ace and social con en ions o he a ge i ual
en i onmen is esponsibili y o se ices p o ide s.
ii. Use s can con ac and in e ac wi h agen s using any de ice wi h In e ne
connec ion and hus pa icipa e in i ual socie ies wi hou hinde ing mobili y.
iii. Use s can pay o speci ic capabili ies, hus de e mining he scope and skills
ha b oke agen s can de elop in i ual socie ies.
Re isi ing he lessons lea ned om ields acing simila challenges (see Sec-
ion 2.5), i can be no ed ha he AS model eplica es many i ues o p e ious
success ul solu ions. In pa icula :
ASPs con ibu e o building i ual b eeding en i onmen s (see Sec ion 2.5.3,
page 45). As commen ed, a business si e e alua es he capabili ies o he ASPs
wan ing o pa icipa e in u u e business oppo uni ies. This condi ion ensu es
ha all so wa e agen s deployed by he ASPs mee beha io al condi ions and
sha e bo h common on ologies and communica ion echnologies. In addi ion,
ASPs do no su e om lack o pa icipa ion, which is a es ic ion adi io-
nally a ibu ed o he solu ions based on he VBE concep . On he con a y,
he ASP ole is designed o ins an ia e housands o nodes, being all o hem
conside ed alid. The e o e, ASPs help o no malize he en i onmen , while
p ese e he au onomy o he cus ome s and p omo e hei pa icipa ion.
60
3.3 Agency Se ices
As shown in Figu e 3.1, solu ions based on ASPs p o ide an a chi ec onic
s uc u e e y simila o ha o hyb id P2P ne wo ks (Figu e 2.10, page 42),
which a e cha ac e ized by he concep o supe node. In p ac ice, ASPs a e
supe nodes ha p incipally de elop ad anced b oke ing unc ions on behal o
o he nodes. In addi ion, hey can p o ide o he in e es ing beha io s such as
accessing hi d-pa y se ices o ob ain and p ocess in o ma ion ha may be
necessa y o he b oke agen s and he local nodes.
Many o he conclusions eached by he communi y de o ed o g id compu ing
a e ac ually pa o he agency se ices ounda ion. On one hand, he AS model
is buil on he p inciples o se ice o ien a ion and s anda dized communica-
ions, which is p ecisely he app oach adop ed by he la es g id compu ing
de elopmen amewo ks in o de o imp o e in e ope abili y. On he o he
hand, he AS model aims o achie e a comp omise be ween using ad anced
mechanisms based on in elligen agen s and delega ing he mos complex pa
o his echnology o specialized companies, as well as o conduc he p ocess
in con olled en i onmen s. This comp omise sha es many cha ac e is ics wi h
he p ac ical ision ha has b ough g id compu ing o achie e ope a ional
solu ions (see Sec ion 2.5.2, page 43).
Fu he mo e, in gene al, compa ed o adi ional mul i-agen sys ems, he AS
model o e s ad an ages in he ollowing aspec s:
Pa icipa ion: The ans e o he mos complex asks o he cloud, oge he
wi h he simplici y o e ed o he clien s, makes i easie o au oma e he cus-
ome s’ pa icipa ion in mode n i ual socie ies.
Scalabili y: Focusing he mos complex echnologies in companies ha a e
in ended o be powe ul and specialized, b ings ou ad an ages o economies
o scale, so ha he solu ion can g ow wi h li le e o .
Flexibili y: The clien can pa icipa e in mo e han one ype o i ual socie y
wi h no need o addi ional e o s. Fu he mo e, he clien can choose he
p o ide ha bes i s i s needs.
Reliabili y: The model pe mi s c i e ia o be es ablished o he ASPs so ha
he b oke agen s’ ac i i y does no endange he s abili y o he sys em due o
sel ish o an i-social beha io s.
61

3. AGENCY SERVICES
Compe i i eness: The na u e o in e ac ions, clea ly o ien ed o acili a e he
implemen a ion o compe i i e models, helps c ea e exchanges based on ma -
ke mechanisms.
3.3.4 In ela ion o in elligen agen s heo y
3.3.4.1 In elligen agen s and se ices-o ien a ion
In he wo ld o so wa e, se ices-o ien a ion p ac ically means web se ices, which
a e pieces o business logic accessible ia s anda d In e ne p o ocols. Thei aim
is ha emo e clien s can build obus and complex s uc u es based on loosely
coupled and he e ogeneous unc ionali ies. Al hough he echnology has been
widely accep ed o clien -se e communica ions, i s speci ica ion su e om cha-
ac e is ics ha limi In e ne op ions. The mos p ominen a e he need o know
in ad ance he de ini ion o se ices o in oke, he absence o seman ic in o ma-
ion and he use o non-pe sis en communica ions which a e always based on he
eques - esponse pa e n.
Acco ding o W3C speci ica ions, so wa e agen s a e a necessa y componen
o a icula e he web se ices in as uc u e [Bea04]:“so wa e agen s a e he u-
nning p og ams ha d i e web se ice, bo h o implemen and o access hem as
compu a ional esou ces ha ac on behal o a pe son o o ganisa ion”.
In line wi h his app oach, agen s ha e been p oposed o be pa o he busi-
ness logic o web se ices wi h he aim o p o iding in elligence and eac i i y o
hei beha io s, il e ing eques s and sea ching o sou ces o in o ma ion [CL07].
Fu he mo e, mechanisms ha e been p oposed in o de o agen s and se ices can
in e ac wi h each o he in a anspa en manne [GC04]. Thus, agen s may exploi
unc ionali ies o e ed by bo h o he agen s and se ices a ailable in he con ex .
Howe e , in p ac ice he absence o seman ic in o ma ion has signi ican ly li-
mi ed he applicabili y o hese lines o wo k. In esponse, he communi y has been
wo king since 2001 o ans o m In e ne in o a seman ic web by means o on o-
logies and adop ing s anda ds o he desc ip ion o esou ces. These echnologies
aim o enable so wa e agen s o eason abou p ope ies and unc ionali ies o web
esou ces and se ices [SBLH06]. In his subjec , agen s a e no mally p oposed o
o ches a ing se ices and sea ching hose ha sa is y he goals o he clien .
62
3.3 Agency Se ices
As shown, e o s ha ela e in elligen agen s and web se ices gene ally aim
o imp o e he unc ionali y and accessibili y o se ices. Howe e , he e a e no
p oposals in he opposi e di ec ion: assessing he success o web se ices solu ions
in o de o o e come he p oblems which hinde he popula iza ion o so wa e
agen s. The Agency Se ices model esponds o his no el ision: i uses se ice-
o ien a ion ideas o c ea e a new model ha acili a es use access o he in elligen
agen s echnology.
3.3.4.2 Agen s as in e media ies
In in elligen agen s heo y many e o s ha e been de o ed o de eloping he concep
o in e media y agen , also known as middle agen . The objec i e o his ype o
agen is o assis in communica ion asks in o de o acili a e exchanges be ween
eques e s and p o ide s. Reques e s a e agen s wi h objec i es hey wan o be
achie ed by o he agen s, whe eas p o ide s a e agen s ha ul ill objec i es on
behal o o he agen s. The p esence o middle agen s is especially aluable in dis-
ibu ed, open en i onmen s, whe e hey cons i u e a mechanism o o e come he
he e ogenei y be ween pa ne s. Al hough he e a e se e al oles o middle agen s,
h ee o hem a e mainly ecognized [KS01]:
Ma chmake : The unc ionali y o a ma chmake agen co esponds o ha
o he yellow pages. P o ide s egis e hei skills in he ma chmake agen .
Reques e s consul i in o de o iden i y hose p o ide s ha a e capable o
ul illing hei objec i es. I he ac i i y o he ma chmake is success ul, he
eques e and he p o ide hen en e in o a new dialogue.
Blackboa d: The blackboa d agen egis e s pe i ions co esponding o asks
o be done. Speci ically, eques e s send hei pe i ions o he blackboa d,
whe eas p o ide s ask his o pe i ions hey can ul ill. In addi ion, he black-
boa d agen is commonly p oposed o keep ack o he eques s and hei e-
spec i e answe s so ha o he agen s can easily ex ac in o ma ion la e .
B oke ing: The aim o he b oke agen s is ac on behal o he eques e s. I
nego ia es he eques e s’ pe i ions wi h he p o ide s, and inally conduc s he
esul s o he eques e . The e o e, in he models domina ed by a b oke , he e
is no di ec in e ac ion be ween eques e s and p o ide s.
63
3. AGENCY SERVICES
In he FIPA p o ocols he e is one specially dedica ed o he in e ac ions me-
dia ed by b oke agen s [FIP02a]. In sho , he agen ha ini ia es he in e ac ion
(ini ia o ) delega es he accomplishmen o a ask o a b oke . A e sending he
eques , he ini ia o plays no u he pa in he p ocess. In gene al, he FIPA spe-
ci ica ion p oposes nei he clien au onomy no he need o elie e he clien o he
echnical de ails.
The Agency Se ices model is clea ly based on he b oke ing ole. The no el y
o he model is ha i delega es his unc ionali y o he cloud so ha b oke agen s
a e hi ed and wo k as a cloud se ice. O cou se, he p oposal also de ails and
sol es all he issues ha a ise as a esul o ex e nalizing he b oke ing ole.
3.4 Agency Se ices o he Sma G id
3.4.1 ASPs o he Ene gy Managemen (ASPEMs)
The a chi ec u e p oposed o he Sma G id in he OASIS and NIST s anda ds
lea es he doo open o he ins alla ion o nodes wi h di e en p o iles. In he pa -
icula case o OpenADR, he i s e sion o he s anda d [PAG+09] conside ed:
(i)simple nodes, which au oma ically apply he DR signals hey ecei e; and (ii)sma
nodes, which usually wo k as agg ega o s o simple nodes, being able o p ocess
and ans o m he ecei ed signals. In his ega d, ou pu signals o sma nodes
a e usually designed o achie e he same esul as he inpu signals, bu espec ing
in e nal condi ions o he sub-sec ion managed by he node, including p e e ences
and equi emen s o cus ome s. Despi e hei ad anced beha io , sma nodes can-
no be conside ed as powe ul as supe nodes in P2P ne wo ks o ASPs in he AS
model. The e o e, gi en ha he AS model inhe i s many ad an ages o he cloud
compu ing pa adigm and eplica es he i ues o exis en success ul solu ions o
simila en i onmen s (such as P2P ne wo ks), i is ad isable o s udy he bene i s
ha may a ise om ins alling nodes wi h compa able cha ac e is ics o hose o
ASPs in dis ibu ed ene gy ne wo ks.
In esponse o his oppo uni y, his esea ch s udies he ins alla ion o Agency
Se ices P o ide o he Ene gy Managemen (ASPEMs), which is in oduced
64
3.4 Agency Se ices o he Sma G id
as a ype o node capable o adding in elligen managemen beha io s and p o id-
ing ad anced da a se ices o he elec ical g id’s cus ome s. Like agg ega o s,
ASPEMs a e nodes ha implemen bo h in e aces VTN and VEN simul aneously.
Wha highly dis inguishes an ASPEM o an agg ega o (o a sma node in he case
o DR a chi ec u es) is he manne in which incoming e en s a e p ocessed: ins ead
o edi ec ing he e en s di ec ly o he lea nodes, o dis ibu ing hem acco ding
o p ede ined c i e ia, ASPEMs a e in ended o p o iding ad anced unc ionali ies,
including he abili y o ins an ia e ene gy ma ke s. In his ega d, depending on he
au onomy o use s o de end hei in e es s, wo app oaches a e possible:
Dis ibu ed mechanism: The b oke agen s coo dina e o nego ia e be ween
hemsel es he signals hey will send o he cus ome s. This app oach allows
he ASPEMs o un in e nal ene gy ma ke s in which b oke agen s pa icipa e
acco ding o he p e e ences con igu ed by he cus ome s. As a esul o ne-
go ia ions, he inpu signal is ansla ed in o new se s o commands o be sen
h ough he VTN in e ace o clien s.
Cen alized mechanism: The ASPEM uns an in elligen algo i hm ha de-
cides which signals mus be sen o each cus ome . In making he decision,
he clea ing algo i hm can also conside he use s’ p e e ences. Unde his
app oach, di ec in e ac ion be ween b oke agen s may be unnecessa y.
Ce ainly, he mos no el app oach is ha based on dis ibu ed mechanisms,
since i allows esponding o he incoming signals using ma ke s.
One o he main goals o he Agency Se ices model is o simpli y he in as-
uc u e o he clien by delega ing he complex and ad anced beha io s o en i ies
in he cloud. Applying his condi ion o he Sma G id and he OpenADR s anda d
means ha end-nodes, which a e ypically use s’ acili ies, may adop he simples
p o ile o he s anda d (i.e., p o ile 2.0a), bu s ill enjoying pa o he ad an ages
o he mos sophis ica ed ones (i.e., p o iles 2.0b and 2.0c) hanks o he ac ion o
ASPEMs and b oke agen s. Fo ins ance, his capabili y would pe mi con e ing
signals o ype del a ( hose ha speci y he amoun o be cu ailed) in o se s o
signals o ype simple, which a e es ic ed o using he alues: no mal,mode a e,
high and c i ical (see Sec ion 2.3.2, page 34). This con e sion can be done so ha
65

Escucha. A iende. Vamos a ealiza jun os el esc u inio de
la esc i u a. Te ense˜
na ´
e el di ´
ıcil a e de la ciencia
esc ip u al que no es, como c ees, el a e de la lo aci´
on de
los asgos sino de la des lo aci´
on de los signos.
“Yo el Sup emo”, Augus o Roa Bas os.
CHAPTER
4
Re iew o he agen -based
algo i hms o DENs
managemen
The p e ious chap e in oduced an a chi ec onic solu ion aimed a acili a ing
he use o so wa e agen s in la ge dis ibu ed en i onmen s such as he Sma G id.
This chap e , in o de o p o ide a comple e solu ion, is de o ed o iden i ying
algo i hmic solu ions which use so wa e agen s o he managemen o DENs.
The ex ensi e li e a u e de o ed o sol ing simila p oblems leads us i s o s udy
he s a e-o - he-a , which gi es he oppo uni y o ha ness aluable insigh om
exis ing p oposals and, no leas , con ibu e o hei ealiza ion and imp o emen .
Howe e , a he p esen momen , he e a e no e iews dedica ed o he assessmen
o agen -based algo i hms o DENs managemen . The ex o his chap e ills his
gap. In pa icula , in o de o ully exploi he capaci y o he Agency Se ices
model, he ollowing s udy ocuses on esea ch ha acili a es he implemen a ion
o he SDM model (see Sec ion 2.2.2, page 28), so i is p incipally cen e ed on
wo ks based on ma ke mechanisms.
Fi s , his chap e cha ac e izes ene gy ma ke s by desc ibing all ea u es and
condi ions ha algo i hms designed o implemen e icien ene gy managemen sys-
73
4. REVIEW OF THE AGENT-BASED ALGORITHMS FOR DENS
MANAGEMENT
ems mus sa is y. Secondly, he ex p oceeds wi h he p esen a ion and e iew o
algo i hms ha , in p inciple, ha e been designed o implemen managemen sys-
ems based on he SDM concep . Wi h he aim o d awing use ul conclusions, he
algo i hms a e classi ied acco ding o he p ope ies equi ed by ene gy ma ke s.
This classi ica ion se es as s a ing poin o a discussion ha will p o ide unde s-
anding o hei sui abili y, ad an ages and weaknesses, paying pa icula a en ion
o hei comple eness and e iciency. This e iew o he s a e-o - he-a ends up
ema king on he mos p omising s udies and poin ing ou he aspec s ha need o
be ein o ced in u u e esea ch. In addi ion, as his wo k is closely ela ed o a i-
icial in elligence, his e iew includes a sec ion on ypical echniques used in his
esea ch ield which a e commonly p oposed o co e ing ancilla y asks.
4.1 Cha ac e is ics o ene gy ma ke s
In ene gy ma ke s he exchanged good is elec ical ene gy, and he pa icipan s
a e so wa e agen s ha make o e s o p oducing and consuming i on behal o
cus ome s. Despi e lessons lea ned in elec onic comme ce, ene gy ne wo ks ha e
special ea u es ha make ene gy ma ke s especially di icul o manage:
1. Supply and demand mus be balanced con inuously in o de o ensu e he
p ope ope a ion o he ne wo k.
2. Ene gy is no a good ha can be s o ed on a la ge scale, so all he ene gy ha
canno be consumed a he momen has o be disca ded.
3. Supply and demand depend on unce ain ac o s and he e o e i is necessa y
o wo k wi h o ecas s and es ima es.
4. Reac i i y o consume s and p oduce s is limi ed and slow.
In addi ion, he need o ensu e e icien use o esou ces and gua an ee ha use s
can sa is y hei demand wi hou excessi e isk- aking, make ha ene gy ma ke s
a e a he di icul o implemen . Speci ically, ully unc ioning ene gy ma ke s a e:
Mul i-uni : Consume s and p oduce s can nego ia e a a iable amoun o ene gy,
so mo e han one uni o he good can be exchanged. In his case, bids a e usua-
lly exp essed in o m o linea piece-wise unc ions [DJ03, SS01] (Figu e 4.1),
74
4.1 Cha ac e is ics o ene gy ma ke s
in which p ice inc eases in ela ion o he amoun ; while he clea ing p ice is
de e mined by he amoun o ene gy exchanged.
Mul i-i em: Ene gy ma ke s a e agmen ed in o ime slo s. In p ac ice, each
slo is a ma ke i em ha can be nego ia ed by he pa icipan s.
Combina o ial wi h complemen a y goods: This e e s o he suppo o bids
ha include g oups o i ems ha mus be accep ed o ejec ed as a whole. In
he case o ene gy ma ke s, i e e s o he possibili y o submi ing bids ha
span mul iple consecu i e ime slo s.
Combina o ial wi h supplemen a y goods: This e e s o he possibili y o ma-
king bids including se e al g oups o i ems, so ha only one o hem can be
accep ed. In ene gy ma ke s, his ea u e means ha a cus ome can de ine
mul iple pe iods o ime in which a speci ic amoun o load o gene a ion can
be accep ed.
gene a ion
p ice
demand
p ice
Figu e 4.1: Linea piece-wise unc ions ha de ine he beha io o he gene a ion and
consump ion o ene gy.
Suppo ing combina o ial bids, despi e he complexi y ha i en ails, is an im-
po an equisi e o ene gy ma ke s. Fo ins ance, i is usually necessa y ha gene-
a ion de ices such as oil-based engines ha e o be ac i e du ing a minimum pe iod
o ime o be p o i able and e icien . Likewise, some ypes o loads, such as wash-
ing machines and dishwashe s, may need o span hei ac i i y du ing mul iple
consecu i e ime slo s in o de o inish hei wo k. In hese cases, i submi ing
bids o complemen a y goods is no suppo ed, cus ome s ha e o send sepa a e
bids o each ime slo , which means he cus ome mus isk ha no all bids a e no
75
4. REVIEW OF THE AGENT-BASED ALGORITHMS FOR DENS
MANAGEMENT
accep ed o ejec ed join ly. On he o he hand, bids o supplemen a y i ems allow
ha de ices such as washing machines and wa e hea e s can de ine disjoin se s
o ime slo s in which hei demand can be supplied, hus allowing he sys em o
shi pa o he demand a peak imes, and gi ing lexibili y o he p o ide s. The
impo ance o combina o ial bids, howe e , is no e lec ed in he s a e o he a ,
which usually de ines ene gy ma ke s wi hou conside ing hese p ope ies.
Ma ke s ha mee all he abo e cha ac e is ics a e e y di icul o esol e. As
a ma e o ac , in medium o la ge sized en i onmen s, he sea ch o an op imal
solu ion is conside ed a NP-Ha d p oblem [RPH98, San02]. As a esul , in p ac ice,
solu ions a e based on simpli ied models, so ha heu is ics a e used o accele a e
he sea ch o solu ions, and, when possible, pa icula i ies o he case unde s udy
a e exploi ed.
Fu he mo e, he complexi y o ene gy ma ke s is inc eased by he p ope ies
ha a e expec ed om any clea ing algo i hm ha ope a es in he con ex o he
Sma G id, which should ha e:
Responsi eness: In dis ibu ed ene gy en i onmen s, planning is sho - e m,
using ime ho izons ha can ange om i een minu es o ew hou s, so he
algo i hm mus be able o quickly ind a solu ion.
Reliabili y: The solu ions p oposed by he algo i hm mus be able o mee
goals imposed by he SO, such as ene gy quali y and supply eliabili y.
Scalabili y: In con ex s ypical o he Sma G id, he numbe o nodes can
g ow signi ican ly, so i is impo an ha he pe o mance o he algo i hm, as
well as he quali y o he solu ions, scale success ully.
Au onomy: The solu ion p o ided by he algo i hm mus s i e o mee he p e-
e ences o use s. The sys em goals may con lic wi h indi idual p e e ences,
so he algo i hm should also be able o ind a wo kable comp omise.
Reac i i y: In he elec ical g id, planning is pe o med on he basis o o e-
cas s and es ima es, so unbalances be ween supply and demand is a equen
eali y. The e o e, he clea ing algo i hm mus be able o eac o unexpec ed
condi ions.
Flexibili y: The algo i hm mus be able o adap i sel o unexpec ed e en s
and e en ual o de s om he SO.
76
4.2 Algo i hms
Despi e all he ad an ages ha he in ol emen o so wa e agen s p omises, all
he abo e-men ioned cha ac e is ics o ene gy ma ke s, including he p esence o
combina o ial bids, make de eloping a ully e ec i e solu ion a complica ed ask.
4.2 Algo i hms
This sec ion e iews ep esen a i e wo ks ha desc ibe solu ions based on so -
wa e agen s o he managemen o dis ibu ed ene gy a eas such as ene gy cells
and mic o-g ids. Two main ca ego ies o solu ions a e conside ed: (i) hose ha
p opose mic o-ene gy ma ke s based on he SDM model; and (ii) hose ha a e
in ended o p o iding ancilla y se ices and manage eme gencies. In bo h cases,
he e iew is mainly de o ed o echniques ha come om he a i icial in elligence
ield and elec onic comme ce.
4.2.1 Supply-demand ma ching
4.2.1.1 Double-sided auc ions
In o dina y auc ions, o e s om all agen s a e collec ed by a cen al au ho i y
called he auc ionee , who is esponsible o de e mining which a e he winning
o e s and how he esou ces a e dis ibu ed among hem. When bo h p oduce s and
consume s can submi bids and o e s, auc ions a e said o be double-sided o wo-
sided auc ions [FR93]. E en hough his app oach seems o be sui able o sol ing
mos o he p oblems, in p ac ice i is ha d o implemen an algo i hm ha , a e
e alua ing all bids and o e s, de e mines which esou ces a e assigned o which
en i ies. In his ega d, he mos basic implemen a ion is o de ine a clea ing p ice
in which all o e s ha exceed i a e accep ed. This solu ion, despi e i s simplici y,
has shown o be alid and speedy o en i onmen s ha a e no expec ed o g ow
beyond p ojec ed bounda ies.
Focusing on he li e a u e o he Sma G id, Dimeas and Ha zia gy iou (2004,
[DH04]) uses English auc ions o dis ibu e gene a ion esou ces be ween con-
sume s in mic o-g ids. In his wo k, ene gy ma ke s a e held o 15-minu e pe iods.
P oduce s and consume s wo k, espec i ely, as auc ionee s and bidde s who com-
pe e o blocks o ene gy. To ensu e compe i i e p ices, a G id Ope a o agen
77

4. REVIEW OF THE AGENT-BASED ALGORITHMS FOR DENS
MANAGEMENT
announces he p ices a which he main g id is willing o sell and pu chase ene gy.
Fu he mo e, he ime limi o nego ia ions is 3 minu es. Ramachand an e al.
(2011, [RSEC11]) en iches his wo k by using a mechanism ha minimizes he uel
cos on he gene a ion side, and also implemen s s a egies o handling he playe s’
isk a i ude a he ading pe iod. Speci ically, in o de o op imize he gene a ion
cos s, he au ho s ha e de eloped a no el algo i hm ha combines echniques om
A i icial Immune Sys ems (AIS) and Pa icle Swa m Op imiza ion (PSO).
Bo h p oposals p o ide de ailed expe imen s ha simula e he ope a ion o
small mic o-g ids. In bo h cases, he pa icipa ion o so wa e agen s ha exchange
ene gy blocks by using double-sided auc ions has p o en o b ing bene i s. Howe-
e , i is impo an o no e ha none o he algo i hms suppo s he nego ia ion o
mul iple i ems a he same ime: bids and o e s a e always in ended o co e he
pe iod ahead. The e o e, o la ge and mo e complex scena ios o he Sma G id,
i is necessa y o enable he algo i hms o handle mul i-i em and combina o ial bids.
Mo eo e , i is also necessa y o poin ou ha , in gene al, double-sided auc ions
scale e y poo ly when hese condi ions a e conside ed, as is he case when he
numbe o nodes g ows, apidly esul ing in NP-Ha d p oblems.
4.2.1.2 Pa allel auc ions
When he complexi y is oo high, pa allel auc ions o e an a ac i e al e na i e.
In his scheme, each selle has he op ion o holding i s own auc ion, so many auc-
ions may be unning simul aneously. Al hough selle s may accep combina o ial
bids, hey usually occu in he con ex o single-sided auc ions ha a e signi ic-
an ly simple and as e han he double-sided op ion. On he o he hand, he main
d awbacks o pa allel auc ions a e:
Using dis ibu ed local clea ing algo i hms o ob ain he solu ion causes loss
o global insigh and, consequen ly, he capaci y o ob ain op imal solu ions.
Requi ing he so wa e agen s o hold and manage hei own auc ions may be
a demanding ea u e ha can a ec he le el o pa icipa ion.
Sending he same bid o mo e han one auc ion implies ha he bidde is ac ua-
lly o e booking his/he capaci y, which may lead o solu ions ha , in p ac ice,
canno be implemen ed. Thus, o e booking en ails an impo an isk o he
78
4.2 Algo i hms
secu i y and eliabili y o he elec ical g id. On he o he hand, when o e -
booking is o bidden, a speci ic block o ene gy can only be o e ed o a single
auc ion, wi h he esul ha many o hose blocks can emain unassigned a he
end o he p ocess, hus leading o a no iceable was e o sca ce esou ces.
Despi e hese disad an ages, pa allel auc ions may be explici ly eques ed by
he playe s o gain au onomy and con ol o e he decision p ocess.
Amin and Balla d (2000, [AB00]) p esen s a p oo o concep based on pa a-
llel auc ions. Howe e , he auc ions used in he s udy do no suppo mul i-uni
and mul i-i em bids. On he o he hand, Penya and Jennings (2005, [PJ05]) in o-
duces he algo i hm mPJ, which suppo s bids ha a e mul i-uni , mul i-i em and
combina o ial. The main cha ac e is ics o mPJ a e:
In o de ha gene a ion ollows demand (demand-d i en-supply), as elec ical
ne wo ks equi e, he ma ke is based on e e se auc ions. Tha is, auc ions
a e held by consume s, while p oduce s submi bids o selling gene a ion ca-
paci y. The e o e, con a y o he classical app oach, he en i y ha wan s o
acqui e he good holds he auc ion.
Buye s make bids h ough linea piece-wise unc ions [SS01, DJ03] like he
one depic ed in Figu e 4.1.
Auc ions a e non-i e a i e, so consume s ha e o de e mine he winne a e
he i s ound o bids.
The implemen a ion is de eloped acco ding o he Vick ey o mula [Vic61],
so p oduce s a e encou aged o alue ene gy acco ding o hei eal needs.
mPJ is a b u e o ce algo i hm ha shows good pe o mance when combina o-
ial bids a e omi ed. In he o he case, due o he la ge numbe o combina ions
ha may be in ol ed, he pe o mance o he algo i hm d ops no iceably. Howe e ,
as shown by he au ho s, in eal ene gy ma ke s only pa o he whole spec um
o possible combina o ial bids is use ul. In p ac ice, agen s will be p e e en ially
in e es ed in packages o consecu i e i ems since hey allow conduc ing consump-
ion and gene a ion ac ions ha span mul iple ime slo s wi hou aking isks. By
limi ing he se o possible combina ions o he se s o i ems ha a e p incipally
needed, he pe o mance o he algo i hm emains good and use s do no lose sig-
ni ican ac ion capaci y. Despi e mPJ being one o he mos p omising mecha-
79
4. REVIEW OF THE AGENT-BASED ALGORITHMS FOR DENS
MANAGEMENT
nisms o implemen ing ene gy ma ke s based on so wa e agen s, i s pe o mance
has only been es ed h ough heo e ical es s ha a e no ocused on dis ibu ed
ene gy en i onmen s, bu simply on p o ing i s capaci y o scale. The e o e, i is
s ill necessa y o simula e mPJ in scena ios ypical o he Sma G id.
4.2.1.3 P ice-o ien ed sea ch o equilib ium
This is he mos common app oach in classical ma ke s. Pa ies exp ess hei p e-
e ences h ough p ice unc ions, so ene gy p ice is he ac o ha de e mines he
amoun o ene gy ha each pa y will consume and p oduce. In his app oach, he
goal o he clea ing algo i hm is o ind he p ice ha op imizes he esou ces as-
signmen . The main d awback o sea ching in he space o possible p ices is ha
he op imal solu ion canno be ob ained analy ically, which makes he p ocess ime-
consuming, using la ge amoun s o compu a ional esou ces. As a esul , o ackle
i , many wo ks use he p ice esul ing om ma ching he o e and demand agg eg-
a ed unc ions, which is commonly known as equilib ium p ice, as i is supposed
o esul in he quan i y o supply being equal o he quan i y o demand. Fu he -
mo e, i he ma ke needs o be co ec ed, his app oach allows ha he au ho i y
can change esou ces dis ibu ion by simply al e ing p ice signals.
A nhei e (2000, [A n00]) desc ibes a solu ion ha uses p ice-o ien ed sea ch
o equilib ium o manage ene gy sys ems. Howe e , his p elimina y wo k does
no suppo mul i-uni and mul i-i em bids. Mo eo e , agen s canno use u ili y
unc ions. Logen hi an e al. (2008, [LSW08]) implemen s a pool ma ke ha
ope a es in he same manne as ypical wholesale ene gy ma ke s. The cen al
agen ep esen s he pool, which, a e ecei ing all buying bids and selling o e s
om loads and gene a o s, is in cha ge o de e mining he clea ing p ice. This p ice
co esponds o he highes accep ed selling (gene a ion) o e . As in he wholesale
pool ma ke s, all accep ed gene a ion bids a e paid he clea ing p ice, while loads
a e equi ed o pay a ha p ice. In gene al, pool ma ke s a e no in ended o co e
he ea u es desc ibed in Sec ion 4.1.
80
4.2 Algo i hms
4.2.1.4 Resou ce-o ien ed sea ch o equilib ium
The op imal solu ion can also be sough in he space o possible esou ce alloca-
ions. In his case, he algo i hm sea ches o he alloca ion o esou ces ha yields
he equilib ium p ice. Speci ically, i is said ha he sys em is in equilib ium i ,
a e all esou ces ha e been assigned, all agen s a e willing o pay he same p ice
o a speci ic good. The sys em is he e o e in equilib ium i i inds a Pa e o
op imal dis ibu ion o he esou ces. Compa ed wi h he p ice-o ien ed app oach,
his model has he ad an age ha agen s’ demand can be ob ained analy ically om
u ili y unc ions, which signi ican ly accele a es he sea ching p ocess.
Ygge and Akke mans (1996, [YA96]) uses an algo i hm based on his app o-
ach in o de o manage dis ibu ed loads. The algo i hm does no suppo mul i-
i em bids, so he solu ion is no sui able o implemen ing en i onmen s d i en
by supply-demand ma ching. This p oblem is o e come in Ygge and Akke mans
(2000, [YA00]), whe e he au ho s p esen he algo i hm COTREE. This desc ibes
an en i onmen whe e so wa e agen s send hei u ili y unc ions o a cen al node
ha uses New on-Raphson o ind a Pa e o op imal solu ion. COTREE assumes
an a chi ec u e ounded on hie a chical cells [KWK05], which, al hough s anda ds
a e no explici ly conside ed in he wo k, is compa ible wi h he a chi ec u e p o-
posed by he OASIS Ene gy In e ope a ion s anda d. COTREE was implemen ed
and es ed as pa o he CRISP EU p ojec [ECN06]. The es s show ha COTREE
is as and scales well, being able o ind an equilib ium solu ion in less han one
second o scena ios wi h hund eds o nodes. Also, simula ions show ha CO-
TREE is e ec i e o smoo hing demand cu es. Howe e , he algo i hm has he
ollowing handicaps:
Combina o ial auc ions a e no suppo ed. Acco dingly, nodes canno submi
bids ha span mo e han one ime slo , which is a signi ican es ic ion o
ene gy uni s.
Any change in he agen s’ plans equi es es a ing he p ocess, since a new
equilib ium solu ion is necessa y. I mus be no ed ha he es o a ion in ol es
all he nodes.
In gene al, he la e d awback is a ibu able o all solu ions based on equili-
b ium sea ch, since he solu ion ha clea s he ma ke is calcula ed om he u ili y
81
4. REVIEW OF THE AGENT-BASED ALGORITHMS FOR DENS
MANAGEMENT
4.3 Discussion
The pu pose o his sec ion is o iden i y he mos p omising mechanisms, compa e
hem, discuss hei ad an ages, and p opose new lines o esea ch. To his end,
s udies a e classi ied acco ding o he p ope ies equi ed by he ene gy ma ke s. In
addi ion, in o ma ion is gi en abou how he s udies ha e been es ed, and whe he
i is possible o no o ep oduce hei expe imen s. Table 4.1 desc ibes he cha-
ac e is ics used o classi y he wo ks, and Tables 4.2 and 4.3 show hem classi ied
acco ding o di e en se s o hose cha ac e is ics.
Table 4.2 shows ha mos wo ks o e look all he challenges ha dis ibu ed
ene gy ma ke s a e expec ed o ace. In pa icula , mos o he algo i hms do no
suppo o do no e en conside combina o ial bids. As explained, hese ypes
o bids a e essen ial in o de ha cus ome s can plan ac ions ha co e mul iple
ime slo s, which is a usual equi emen in ene gy con ex s. As a ma e o ac ,
only [CA07] (Ca lsson and Ande sson, 2007) and [PJ05] (Penya and Jennings,
2005) suppo his p ope y in i s wo ace s, including packages o complemen a y
and supplemen a y bids. Howe e , hese wo wo ks a e based on a he di e en
app oaches: he algo i hm CONSEC p esen ed in [CA07] looks o he equilib ium
p ice by using a cen alized scheme; while he algo i hm mPJ desc ibed in [PJ05]
p oposes ha each consume holds i s own auc ion, hus leading o a comple e dis-
ibu ed solu ion based on so wa e agen s. The mechanism desc ibed in [CA07]
su e s om d awbacks ypical o cen alized s a egies and hose based on equili-
b ium sea ch (see Sec ion 4.2.1.4). Fu he mo e, mPJ is mo e lexible han CON-
SEC in building combina o ial bids, as hese can be made up o disjoin i ems.
Table 4.2 also shows ha auc ions a e he me hod mos o en conside ed o
implemen ing agen -based ene gy ma ke s. The eason is wo old: (i) auc ions a e
lexible enough o mee all equi emen s o ene gy ma ke s; and (ii) auc ions is
a ypical app oach o he communi y de o ed o de eloping mul i-agen sys ems
because i is dis ibu ed and ensu es he au onomy o agen s in decision-making.
Howe e , heo y says ha ob aining an op imal solu ion by using bila e al auc-
ions becomes a NP-Ha d p oblem in la ge en i onmen s, so pa allel auc ions is,
in p inciple, he only alid me hod o accomplishing his ask. The e o e, in o de
88

4.3 Discussion
o check he sui abili y o he solu ions p oposed in [DH04] (Dimeas and Ha zi-
a gy iou, 2004) and [AB00] (Amin and Balla d, 2000), i is s ill necessa y o con-
duc simula ions wi h ealis ic scena ios conside ing all equi emen s. Fo i s pa ,
he solu ion based on pa allel auc ions p oposed in [PJ05] has s ill o be simula ed
in scena ios ypical o he Sma G id. In his pa icula case, besides checking he
e iciency o he algo i hm mPJ in a ealis ic con ex , i is also necessa y o s udy
how ypical d awbacks o pa allel auc ions may a ec he capaci y o he algo i hm
o balance he ne wo k. In pa icula , he bes -known d awbacks o pa allel auc-
ions a e: (i) poo dis ibu ion o buye s o he selle s [Hop08]; and (ii) he isk o
buye s o e booking hei capaci y o , on he con a y, he business oppo uni ies
hey may lose when hey adop o e ly conse a i e s a egies.
In p ac ice, he only wo ks ha ha e been simula ed by using an elec ical g id
simula o a e he ones ca ied ou in he CRISP EU p ojec . This is because de-
ining ealis ic scena ios and simula ing hem is a complex ask. Fu he mo e, he
in eg a ion o in elligen agen s in he Sma G id equi es a mul idisciplina y eam
wi h specialized membe s in bo h a eas. As a esul , i is no mal ha wo ks com-
ing om eams p incipally de o ed o he esea ch o so wa e agen s o e look
ace s such as simula ions suppo ed by ools belonging o he elec ical engin-
ee ing. Fo ins ance, simula o s such as G idLAB-D, which is open sou ce and
agen s-based ( hus ep esen ing an excellen oppo uni y) is no used in any o he
p e ious wo ks.
On he o he hand, Table 4.3 shows in o ma ion co esponding o o he in e es-
ing aspec s. Among hem is s anda d-o ien a ion, which is a he limi ed in wo ks
ha combine so wa e agen s wi h echniques ypical o he Sma G id. The main
eason o his is due o he de elopmen o s able s anda ds in his ield, as well as
he accep ance o hem as such, has been la e. This ac has con ibu ed o c ea e
a dis o ed idea o he Sma G id and i s goals, wi h he esul ha au ho s a e
occasionally unable o p ecise he con ex o which hei wo ks a e in ended. In he
ques o implemen ing ene gy ma ke s in he Sma G id, well-es ablished s and-
a ds a e de ini ely a aluable sou ce o in o ma ion ha p o ides au ho s he insigh
necessa y o iden i y he a eas in which hey can con ibu e on a sa e ounda ion.
89
4. REVIEW OF THE AGENT-BASED ALGORITHMS FOR DENS
MANAGEMENT
In scien i ic and echnological a eas, i is impo an ha people o he han he
au ho s can ep oduce esea ch and academic p og ess. Rep oducibili y ensu es
ha esul s can be co obo a ed, and ha o he au ho s can conduc new esea ch
based on p e ious wo ks, hus making p og ess as e and mo e eliable. The secu -
i y, eliabili y and e iciency equi ed o any solu ion ha in ol es changes in he
elec ical g id, as well as he p esence o cu ing-edge echnologies, makes ep o-
ducible esea ch especially impo an o he Sma G id, whe e any solu ion mus
be p e iously s udied and es ed in de ail. The de ini ion o amewo ks in ended o
gua an ee wo ks ep oducibili y is inc easingly common [S o09, FC09]. Howe e ,
none o he wo ks e iewed in his documen success ully mee s he condi ions o
hese amewo ks. On he con a y, hey igno ed his aspec . Howe e , he pe -
o mance and e iciency o some me hods p esen ed in he s a e o he a could be
alida ed due o he ex ensi e knowledge we ha e o some mechanisms, as well as
due o he a ailabili y o he esou ces in ol ed. F om his less s ic poin o iew,
as shown in Table 4.3, wo ks based on auc ions can be conside ed ep oducible,
since hey use algo i hms ha a e easy o implemen , and hei e icacy o he pa -
icula con ex hey ha e been designed can be checked. In con as , non- i ial
o mulas and ac o s come in o play in wo ks based on equilib ium sea ch, so addi-
ional in o ma ion is equi ed in o de o conside hei expe imen s ep oducible.
To summa ize, Penya and Jennings (2005) is he mos p omising and comple e
app oach o ully implemen ing unc ioning ene gy ma ke s based on au onomous
so wa e agen s. Ca lsson and Ande sson (2007) mee s all he equi emen s ene gy
ma ke s equi e, bu om a mo e cen alized and de e minis ic app oach. The e-
o e, when he goal is implemen ing ma ke s based on dis ibu ed decision-making,
Penya and Jennings (2005) is de ini ely mo e sui able. Howe e , as men ioned be-
o e, his me hod mus o e come he complica ions o pa allel auc ions, including
he need o so wa e agen s which can conduc hei own auc ions. In his e-
ga d, adop ing he Agency Se ices model could be e y use ul, as he esponsi-
bili y o holding pa allel auc ions would all on b oke agen s. I he e o e seems
app op ia e o assess he po en ial o his combina ion. Fu he mo e, as al eady
discussed, i is s ill necessa y o check he pe o mance o he algo i hm mPJ using
an elec ical g id simula o and ealis ic scena ios.
90
4.3 Discussion
P ope y Desc ip ion Values
Mul i-uni Indica es i i suppo s mul i-uni
bids.
yes, no, — (no conside ed)
Mul i-i em Indica es i i suppo s mul i-
i em bids.
yes, no, —
Combina o ial
complemen a y
Indica es i i suppo s combina-
o ial bids wi h complemen a y
i ems.
yes, no, —
Combina o ial
supplemen a y
Indica es i i suppo s combi-
na o ial bids wi h supplemen a y
i ems.
yes, no, —
Tes ype Indica es he ype o en i on-
men in which he p oposal ha e
been es ed.
sim: Simula ed using an elec-
ical g id simula o .
es : Tes s he e icacy and
scalabili y o he p oposal
wi hou using simula o s.
exp: Expe imen in a labo a o y
en i onmen .
Rep oducible Indica es i he p oposal is ep o-
ducible.
yes, no
S anda ds-based Indica es i i is based on he
Sma G id s anda ds.
yes, no
Table 4.1: P ope ies o desc ibing he wo ks ha aim o implemen he SDM ex-
change model a medium o la ge DENs.
91
4. REVIEW OF THE AGENT-BASED ALGORITHMS FOR DENS
MANAGEMENT
Wo k Mul i
uni
Mul i
i em
Comb.
Comp.
Comb.
Supp. Tes Type
Dimeas e al.
(2005)
yes no no no exp Symme ic
assignmen
Funabashi e
al. (2008)
yes no no no sim Symme ic
assignmen
Nunna and
Doolla (2013)
yes no no no sim Symme ic
assignmen
Dimeas e al.
(2004)
yes — — — es Double sided
auc ions
Ramachand an
e al. (2011)
yes — — — sim Double sided
auc ions
A nhei e
(2000)
yes no no no es Equilib ium
Logen hi an
(2008)
yes no no no es Equilib ium
Ygge and Ak-
ke mans (1996)
yes no no no es Equilib ium
Ygge and Ak-
ke mans (2000)
yes yes no no sim Equilib ium
Ca lsson and
Ande sson
(2007)
yes yes yes yes sim Equilib ium
Amin and Bal-
la d (2000)
yes yes — — es Pa allel auc-
ions
Penya and Jen-
nings (2005)
yes yes yes yes es Pa allel auc-
ions
Rahman e al.
(2007)
yes no — — sim. Ma chmake
Table 4.2: Desc ip ion o he wo ks acco ding o he cha ac e is ics o ene gy ma ke s.
92
4.3 Discussion
Wo k Rep oducible S anda ds-based
Dimeas e al. (2005) yes no
Funabashi e al. (2008) yes no
Nunna and Doolla (2013) yes no
Dimeas e al. (2004) yes no
Ramachand an e al. (2011) yes no
A nhei e (2000) yes no
Logen hi an (2008) yes no
Ygge and Akke mans (1996) no no
Ygge and Akke mans (2000) no no
Ca lsson and Ande sson (2007) no no
Amin and Balla d (2000) yes no
Penya and Jennings (2005) yes no
Rahman e al. (2007) yes no
Table 4.3: Desc ip ion o he wo ks acco ding o hei s anda d-o ien a ion and he
capaci y o ep oduce expe imen s based on hem.
93

Hay momen os pa a eci a poes´
ıa y momen os pa a
boxea .
“Los de ec i es sal ajes”, Robe o Bola˜
no.
CHAPTER
5
Simula ion in as uc u e
This chap e in oduces he simula ion in as uc u e used o e alua e he heo-
e ical esea ch done h oughou his hesis. I s design is ma ked by he need o
simula ing bo h he elec ical g id and mul i-agen sys ems, which leads o an in-
as uc u e composed o mul iple modules. The ollowing sec ions desc ibe he
echnologies used o implemen he in as uc u e, he pu pose and main cha ac e-
is ics o each componen , he in e ac ions be ween hem, and he li e cycle o he
simula ion p ocess.
As will be seen, he choice o echnologies is guided by p inciples which a e
pa o he aim o his p ojec . In pa icula , he in as uc u e is designed o comply
wi h s anda d-o ien ed solu ions based in ex ensi ely p o en ools, and espec s
and p omo es he condi ions o ep oducible esea ch. Al hough i migh seem ha
some o hese condi ions could limi he scope o he p esen p ojec , he eade
will ha e he oppo uni y o see ha he bes and mos capable simula ion ools in
bo h domains ( he elec ical g id and mul i-agen sys ems) ha e been implemen ed
acco ding o his hinking, wi h he esul ha , in p ac ice, he main challenge elies
hea ily on achie ing an e ec i e liaison be ween simula o s.
95
5. SIMULATION INFRASTRUCTURE
5.1 Simula ion so wa e
The esea ch p esen ed in his documen seeks o be ep oducible and e i iable
by he esea ch communi y. As desc ibed in ep oducible esea ch amewo ks
[S o09, FC09], condi ions o bo h he so wa e and da a used du ing he expe i-
men al e alua ion play essen ial oles in achie ing his goal. In pa icula , ame-
wo ks s ess he need o mee he ollowing ea u es:
Bo h, so wa e and da a se s mus be accessible. In he case o so wa e, i mus
be possible o use s o ob ain all so wa e componen s equi ed o pe o m he
simula ions. Undoub edly, his need is acili a ed when he so wa e does no
cos any hing and can be ob ained di ec ly om In e ne . As o da a se s,
he e mus be a desc ip ion o how he da a was b ough in o he o m used in
he esea ch. In his ega d, widely accep ed da a se s a e conside ed aluable
esou ces.
The licenses o bo h mus no impose condi ions limi ing he ep oduc ion o
expe imen s and dissemina ion o esul s.
The so wa e mus con ain comple e ins uc ions on how o execu e and use i ,
as well as in o ma ion abou how o ob ain and use he da a se s.
On he o he hand, he ac i i y o he elec ical g id implies he pa icipa-
ion o many subsys ems, which, in p ac ice, esul s in a mul ilaye a chi ec u e
ha en e s mul iple domains, anging om hose ocused on he elec onic be-
ha io o base componen s, o hose ocused on asks ela ed o he long- e m
managemen o gene a ion and consump ion esou ces. This condi ion makes i
pa icula ly di icul o cap u e all he complexi y o he elec ical g id wi h a
single piece o so wa e. As a esul , simula ion ools co e only some speci ic
aspec s o i ; powe low calcula ions and demand models being he mos a en-
ded unc ionali ies. Fo his eason, esea ch p ojec s o en eso o co-simula ion
[LSS+11, GMD+10, LAH11, LXJM12], a e m used o e e o he need o simu-
la e and model coupled p oblems in a dis ibu ed manne , so ha subsys ems a e
simula ed sepa a ely, in e ac ing wi h each o he by using communica ion channels.
Thus, in simula ing he elec ical g id, p ojec s dealing wi h domains o he han he
wo men ioned be o e a e commonly equi ed o use addi ional simula ion ame-
wo ks. This is he case, o ins ance, o he p ojec p esen ed in his disse a ion,
96
5.1 Simula ion so wa e
which p oposes a new a chi ec u al solu ion and he implemen a ion o agen -based
ma ke s. The e o e, o make co-simula ion possible, i is impo an ha he elec-
ical g id simula o p o ides i s unc ionali y h ough con en ional means (such as
APIs o web se ices), o , al e na i ely, acili a es he de elopmen o plug-ins ha
can accomplish his ask.
Gi en he la ge numbe o componen s and subsys ems ha pa icipa e in he
ope a ion o he elec ical g id, in o de o ensu e he expe imen s’ eliabili y, i is
also impo an ha he simula o has been es ed agains benchma k scena ios, such
as he Dis ibu ion Tes Feede s [Ch 99] de ined by IEEE Powe and Ene gy So-
cie y. Addi ionally, his ea u e imp o es he ep oducibili y o he p ojec s es ed
on he simula o .
A desc ip ion o he cha ac e is ics and quali ies o he mos capable simula o s
o he elec ical g id can be ound in [LZ14, RLS+14, S e14, FCD+13]. Fo he
p esen p ojec , he G idLAB-D simula o [CSG08] was chosen o pe o m he ex-
pe imen al e alua ion because, besides being one o he mos p omising op ions, i
mee s all he equi emen s de ined abo e. In pa icula , G idLAB-D has been de e-
loped by he US Depa men o Ene gy (DOE) as a ool o acing he o hcoming
challenges in he ene gy ield. Some in e es ing cha ac e is ics o G idLAB-D a e:
a) I ollows an agen -based app oach and is ex ensible, wi h modules which can
simula e a la ge a ie y o componen s o he elec ical g id a di e en le els
o abs ac ion, including he powe low model and o he physical cons ain s.
b) I is open sou ce, wi h a g owing, ac i e communi y o bo h de elope s and
esea ches wo king on i . In pa icula , G idLAB-D uses he Be keley So wa e
Dis ibu ion (BSD) license, which gua an ees ha use s can make use o i in
all possible manne s, and ha ep oducible esea ch p inciples can be applied
o any de i a i e wo k.
c) I s e icacy has been ex ensi ely es ed. I is wo h men ioning ha G idLAB-
D includes he de ini ion o he s anda d scena ios de ined in [Ch 99], o
which i yields he expec ed alues.
d) I is well documen ed, he in o ma ion being upda ed egula ly. In addi ion, i
has ac i e o ums.
97
5. SIMULATION INFRASTRUCTURE
name: In e nal name o he elemen . This ield is equi ed in o de ha o he
elemen s can e e o he “G idOpe a o ” objec . Fo ins ance, ASPEMs
use his oken o egis e wi h he sys em ope a o .
se icesEndpoin : URL o access he web se ices API.
The ollowing code snippe shows a basic s a emen o his ype o elemen :
objec G idOpe a o {
name "g idop";
se icesEndpoin "localhos :9090/g idop/ esou ces";
};
5.3.1.3 Elemen ASPEM
The “ASPEM” elemen de ines an ASPEM. As in he case o he sys em ope a o ,
i is a Ja a web applica ion whose unc ionali ies a e accessible h ough a web
se ices API. I mus be no ed ha “ASPEM” elemen s mus be decla ed as inne
objec s o “G idOpe a o ”. The ields ha he “ASPEM” elemen includes a e:
name: In e nal name o he elemen . This name is equi ed in o de o o he
elemen s can e e o he ASPEM. Fo ins ance, households use his oken o
de ine he ASPEM o which hey a e connec ed ia he ASBox.
pa en : In e nal name o he “G idOpe a o ” elemen .
se icesEndpoin : URL o access he web se ices API.
The ollowing code snippe shows a basic s a emen o an “ASPEM” elemen :
objec ASPEM {
name "aspem01";
pa en "g idop";
se icesEndpoin "localhos :8080/aspem/ esou ces";
};
5.3.1.4 Elemen ASBox
In o de ha a household can pa icipa e in he managemen sys em, i mus con-
ain an elemen o ype “ASBox”. This is esponsible o communica ing wi h he
104

5.3 Modules and applica ions
co esponding ASPEM, and hus being able o apply he o de s sen by he b oke
agen s, as well as sending in o ma ion on he s a e o he local esou ces. This
elemen is also used o se use p e e ences, including he oles hey will play in
he ma ke s, and cha ac e is ics ela ed o hei beha io s. The ields o “ASBox”
objec s a e:
name: In e nal name o he elemen . ASBoxes a e end nodes (no o he elemen
e e s o hem). Howe e , speci ying he in e nal name is s ill impo an o
debugging and logging ac i i y.
pa en : In e nal name o he “house” elemen o which he ASBox is linked.
aspem: In e nal name o he “ASPEM” elemen o which he ASBox is con-
nec ed.
le elsSchedule: In e nal name o he schedule elemen ha associa es alues
o ime pe iods. The alue co esponding o a pe iod iden i ies: (i) he mode
o ope a ion o he household (ha d,easy o no mal); and (ii) he le el o
consump ion when he selec ed mode is ha d o easy, which can be no mal,
mode a e,high o c i ical. In e nally, hese wo alues a e combined in o a
numbe ; howe e , o he sake o eadabili y, hence o h his combina ion is
shown in plain ex . As o schedules de ini ion, G idLAB-D p o ides a na-
i e ype ha uses he same syn ax and seman ic as he well-known piece o
so wa e C on ab [Rez93].
s a ingP ice: Maximum p ice 1a which he b oke agen is willing o buy
ene gy blocks in pa allel auc ion ma ke s.
le elP iceX: P ice a which he b oke agen is willing o sell ene gy blocks
co esponding o he mode a e (le elP ice1), high (le elP ice2) and
c i ical (le elP ice3) le el o consump ion in ene gy ma ke s.
p io i y: In ege alue use ul o esol ing con lic s and/o p io i izing some
cus ome s o e o he s. The alue o his pa ame e is mean o be con ac ed
wi h he ASPEM.
The ollowing code snippe shows a basic s a emen o an “ASBox” elemen :
1The ma ke is based on e e se auc ions, so he de ini ion o concep s such as he s a ing p ice
is in e ed.
105
5. SIMULATION INFRASTRUCTURE
objec ASBox {
name i1B645;
aspem aspem01;
pa en house1B_ m_B_1_645;
le elsSchedule sch_02;
s a ingP ice 140;
le elP ice1 105;
le elP ice2 140;
le elP ice3 143;
p io i y 3;
};
As an example, he code snippe shown below con ains a G idLAB-D schedule
wi h h ee pe iods o he Augus 1s . I can be seen ha om 13:30h o 15:30h
he node wo ks as a ha d-load, willing o p o ec he mode a e le el as minimum;
whe eas om 15:30h o 17:00h, i is willing o adop he high le el e en i i is no
equi ed. The ea e , he node adop s he no mal mode. I is assumed ha he node
wo ks in no mal mode o all hose pe iods ha a e no explici ly de ined.
schedule sch_02 {
30 13 1 8 *ha d(mode a e);
30 15 1 8 *easy(high);
00 17 1 8 *no mal;
};
5.3.2 Sys em Ope a o applica ion
The sys em ope a o is ep esen ed by a Ja a web applica ion unning in a Je y
se e . The applica ion p o ides he ollowing unc ionali ies: (i) a use in e ace
o easy con igu a ion o scena ios; (ii) ins an ia ion o he so wa e agen ha ca-
ies ou he sys em ope a o asks, including li e cycle managemen o DR e en s;
and (iii) web se ices API ha G idLAB-D uses o communica e wi h he sys em
ope a o . The implemen a ion and ins an ia ion o he so wa e agen is add essed
using he module Ene gyAgen s (see Sec ion 5.3.4).
106
5.3 Modules and applica ions
The p ima y objec i e o he use in e ace is o p o ide an easy means whe eby
use s can con igu e he scena ios a ailable. Speci ically, scena ios a e c ea ed
h ough he o m New Scena io (Figu e 5.3), which eques s he ollowing in o -
ma ion:
Code: Unique code ela ed o he scena io being c ea ed. When G idLAB-
D communica es wi h he sys em ope a o applica ion, his code is used o
iden i y he scena io being simula ed (see Sec ion 5.3.1.1).
S o e name: Name o he da abase o demand es ima es o be used in simula ions.
P og am: File wi h he de ini ion o he DR p og am o be applied du ing he
simula ion (see Sec ion 5.4.1).
Desc ip ion: Use commen s abou he scena io.
Figu e 5.3: Fo m o egis e a new scena io.
Mo eo e , he applica ion p o ides iews o lis ing he cu en scena ios and
p o iding in o ma ion on he simula ions pe o med so a . These iews include
basic ac ions ha allow use s o c ea e, modi y and dele e i ems.
107
5. SIMULATION INFRASTRUCTURE
Table 5.1 desc ibes he REST ul web se ices API p o ided by his applica ion.
As can be seen, se ices a e ela ed o he managemen o he simula ion p ocess.
Essen ially, G idLAB-D uses his se o se ices o in o m he sys em ope a o o
simula ion e en s, and egis e new ASPEMs when a new simula ion is s a ed.
/se ices/simula ions
Name URL Ope . Desc ip ion Pa ame e s
s a / POST S a a new simula ion, and e u n
he iden i ie ela ed o he simula-
ion.
scena iocode: Code o he
scena io being simula ed.
usp: Flag indica ing whe he using
s a ing p ices (1) o no (0).
nd: Flag indica ing whe he using
uni o m dis ibu ion o buye s (1)
o no (0) (see Chap e 7).
c : Value o he cons an C (see
Chap e 7).
inish {id}/ inish PUT End he simula ion ela ed o he
iden i ie id.
id: Iden i ie o he simula ion.
pause {id}/paused PUT Pause he simula ion ela ed o he
iden i ie id.
id: Iden i ie o he simula ion.
newAspem {id}/aspem POST Ins an ia e a new ASPEM, and e-
u n he iden i ie ela ed o he AS-
PEM.
id: Iden i ie o he simula ion.
con en : XML desc ip ion o he
ASPEM (see Lis ing A.1, Appendix
A).
Table 5.1: Desc ip ion o he laye o REST ul web se ices p o ided by he applica-
ion ha ep esen s he sys em ope a o .
5.3.3 ASPEM applica ion
ASPEMs a e implemen ed as Ja a web applica ions ha p o ide he ollowing
unc ionali ies: (i) a use in e ace; (ii) web se ices API ha G idLAB-D uses o
communica e wi h he ASPEM; (iii) ins an ia ion o he so wa e agen esponsible
o handling he DR e en s; (i ) c ea ion o he i ual en i onmen in which nego-
ia ions a e pe o med; and ( ) ins an ia ion and deploymen o he b oke agen s
ha ac on behal o use s. The c ea ion and ac i i y o so wa e agen s, inclu-
108
5.3 Modules and applica ions
ding he en i onmen hey un, is add essed h ough he module Ene gyAgen s (see
Sec ion 5.3.4).
Fo ASPEMs, he use in e ace does no p o ide g ea unc ionali y, since
b oke agen s as well as o he in o ma ion ha migh be o in e es , a e dynamically
loaded when he simula ion s a s. The e o e, he use in e ace only con ains an
in o ma ion sc een ha allows use s o check i he applica ion is p ope ly wo king.
On he o he hand, Table 5.2 desc ibes he REST ul web se ices API o he
ASPEM applica ion. This is mean o in o m he ASPEM when simula ions a e
ini ia ed and comple ed. In addi ion, once a simula ion is unning, i o e s a se -
ice h ough which G idLAB-D egis e s he ASBoxes ha will be connec ed o
a pa icula ASPEM. This se ice ecei es use p e e ences as a pa ame e . Fo
ins ance, in he pa icula case o auc ions ma ke , o each ASBox, G idLAB-D
epo s he ASPEM when he use pa icipa es as easy-load and ha d-load, he p ice
a which he/she is willing o buy o sell ene gy, and he s a ing p ice when he/she
ac s as auc ionee . Mo eo e , he REST ul API p o ides a se ice ha epo s he
s a us o he ASPEM. Thanks o his, he sys em ope a o can check i ASPEMs
a e ope a ing as expec ed.
5.3.4 Module Ene gyAgen s
The Ene gyAgen s module implemen s all he unc ionali ies ela ed o so wa e
agen s ha bo h he sys em ope a o and ASPEMs equi e. This module he e o e
wo ks as an ex e nal API ha ees oo applica ions om dealing wi h specialized
concep s and p ocedu es belonging o he a i icial in elligence a ea. Ene gyAgen s
is, in u n, buil on he Jade amewo k, which acili a es he implemen a ion o
mul i-agen sys ems ully complian wi h he FIPA speci ica ions [FIP96].
The module c ea es a pla o m composed o mul iple con aine s [FIP03a] which
wo k as unning en i onmen s o he so wa e agen s. Following he Agency Se -
ices model’s guidelines, he sys em ope a o and each ASPEM owns a con aine .
Fu he mo e, o each managemen me hod used in simula ions, he Ene gyAgen s
module implemen s all beha io s o : (i) he agen which ac s on behal o he sys-
em ope a o esponsible o managing he e en ’s li e cycle; (ii) he agen ha
109

5. SIMULATION INFRASTRUCTURE
/se ices/simula ion
Name URL Ope . Desc ip ion Pa ame e s
s a /{id}/s a POST Repo ha a new simula ion has
been ini ia ed.
id: Iden i ie o he simula ion.
inish /{id}/ inish PUT Repo ha he simula ion ela ed o
he iden i ie id is comple e.
id: Iden i ie o he simula ion.
newASBox /{id}/asbox POST Ins an ia e a new ASBox inside he
ASPEM and ela e i o he simula-
ion wi h iden i ie id.
id: Iden i ie o he simula ion.
clien code: Code ha iden i ies
he ASBox in G idLAB-D.
con en : XML desc ip ion o he
use p e e ences (see Lis ing A.2,
Appendix A).
/se ices/aspem
s a us /id/s a us GET Re u n whe he he ASPEM is
a ailable o no .
Table 5.2: Desc ip ion o he laye o REST ul web se ices p o ided by he applica-
ion ha ep esen s ASPEMs.
ep esen s a pa icula ASPEM, which, a e ecei ing he o de o handle a pa -
icula e en om he sys em ope a o , decides how o implemen i ; and (iii) he
b oke agen s ha pa icipa e in he managemen sys em when i is implemen ed as
a dis ibu ed mechanism.
In o de o achie e a well-de ined solu ion, he module de ines on ologies ha
implemen all he concep s, ac ions and p edica es in ol ed in he in e ac ions and
dialogues be ween agen s. Speci ically, an on ology is de ined o each mana-
gemen me hod used in he expe imen al e alua ion. In addi ion, all in e ac ions
be ween b oke agen s a e conduc ed h ough s anda d FIPA in e ac ion p o ocols.
Speci ically, he FIPA Reques [FIP02c] is used when one agen eques s ano he
o pe o m an ac ion, he FIPA Que y [FIP02b] o asking o speci ic in o ma ion,
and he FIPA In o m [FIP01] communica e ac o sending in o ma ion abou s a es
and ac s.
In he sou ce code, bo h on ologies and agen s’ beha io s a e g ouped in Ja a
packages so ha each o hese is associa ed wi h a speci ic me hod and expe imen .
110
5.3 Modules and applica ions
The co espondence be ween packages and managemen me hods is no ed in he
code i sel .
The Ene gyAgen s module also c ea es i s own laye o en i ies and unc ionali -
ies, which, in addi ion o hiding he in e ace o Jade, esul s in a highly simpli ied
in e ac ion laye . P oo o his is he code o he sys em ope a o and ASPEMs
applica ions, whose asks conce ning agen s consis o a ew calls.
5.3.5 Addi ional modules
The simula ion in as uc u e p o ides wo addi ional modules ha aim o acili a e
specialized asks. These a e: (i)SimpleDR, which acili a es he c ea ion and ma-
nagemen o en i ies and unc ions ypical o he OpenADR s anda d; and (ii)Won-
es , which p o ides an API o accessing web se ices o ype REST ul using he
C++ p og amming language.
To wo k wi h OpenADR i is necessa y o ex ac he en i ies model om he
XML schemes p o ided by he s anda d. The gene a ed model is a he complex,
so, in p ac ice, using i leads o code di icul o main ain and ead. The SimpleDR
p ojec aims o o e come his p oblem by p o iding a much mo e simpli ied model
o en i ies. This new model success ully mee s all equi emen s o he expe imen al
e alua ion, eeing he de elope om ha ing o manage many en i ies, which a e
au oma ically illed. On he o he hand, he Won es lib a y implemen s a simple
REST ul clien o he C++ p og amming language. I does no p e end o be a ull
implemen a ion o he REST ul speci ica ion, bu a simple in e ace o essen ial
unc ions, hiding much o he complexi y ha in ol es wo king di ec ly wi h he
Cu l lib a y [cu 15], which is he classical app oach.
Thanks o hese complemen a y so wa e modules, he business logic o bo h
he Ene gyAgen s and he AgencySe ices modules is ac ually ocused on he issues
hey eally ha e o sol e.
111
5. SIMULATION INFRASTRUCTURE
5.4 Da a iles
5.4.1 DR p og ams ile
The DR p og ams a e de ined h ough XML iles. The de ini ion o DR p og am
used in his p ojec consis s o ze o o mo e e en s, which in u n may consis o one
o mo e in e als. The e en elemen also speci ies, in o m o a ibu e, he ype
o signal o be sen , and he s a ing ime. As o in e al elemen s a e in ended o
speci y he alue o he signal o a speci ic ime pe iod.
The ollowing code snippe shows he de ini ion o a DR p og am which con-
sis s o wo e en s. The i s e en decla es a signal o ype del a ha s a s a
14:00h. This signal is implemen ed by using wo in e als. The i s one las s 3600
seconds and speci ies ha 1000 kW mus be disca ded; while he second in e al
las s 1800 seconds and speci ies a alue o 2000 kW. On he o he hand, he second
e en decla es a signal o ype simple ha s a s a 15:30h. This consis s o h ee
e en s, each one las ing 1800 seconds. Respec i ely, hey o de he cus ome s o
adop he le els mode a e (1), high (2) and c i ical (3).
<?xml e sion="1.0" encoding="UTF-8" ?>
<p og am xmlns="h p://www.siani.es/agencyse ices/e en s/1.0"
name="h14-d3500">
<e en ype="del a" s a ="2000-08-01T14:00:00Z" p io i y="0"
no i Du a ion="300">
<in e al du a ion="3600" alue="1000" />
<in e al du a ion="1800" alue="2000" />
</e en >
<e en ype="le el" s a ="2000-08-01T15:30:00Z" p io i y="0"
no i Du a ion="300">
<in e al du a ion="1800" alue="1" />
<in e al du a ion="1800" alue="2" />
<in e al du a ion="1800" alue="3" />
</e en >
</p og am>
The elemen s con aining a ile de ining a DR p og am a e:
p og am: Main elemen o he XML documen . I may con ain ze o o mo e
elemen s o ype “e en ”. The a ibu es o “p og am” a e:
112
5.4 Da a iles
qxmlns:Namespace associa ed wi h he XML elemen s. I mus be he
same as ha de ined in he p e ious example.
qname: In e nal name o he p og am. This name is shown by he Ja a
applica ion o he sys em ope a o in o de ha use s can iden i y he p o-
g ams.
e en : An OpenADR e en . An e en consis s o one o mo e in e als. In e -
als a e supposed o span consecu i e ime blocks, so he du a ion o he e en
is equal o he sum o all he in e als du a ion. The a ibu es o “e en ”
a e:
q ype: Type o he OpenADR e en . The possible alues a e le el and del a
(see Sec ion 2.3.2, page 34).
qs a : Time when he OpenADR e en s a s.
qp io i y: In ege alue ha de ines he e en ’s impo ance. In he simula-
ion in as uc u e p esen ed in his documen , when p io i y is highe han
ze o, ASPEMs a e equi ed o send he signal di ec ly o households, hus
indica ing ha ma ke s canno be used.
qno i Du a ion: Time slo wi hin which e en s mus be no i ied o house-
holds.
in e al: Elemen ha associa es a pa icula alue o a ime pe iod o he
co e ing e en . The a ibu es o “in e al” a e:
qdu a ion: In e al du a ion in seconds.
q alue: Value associa ed o he block o ime spanned by he in e al. I he
ype o he co e ing e en is del a, he alue exp esses kW; whe eas i he
ype is le el, he possible alues a e: 0(no mal), 1(mode a e), 2(high)
and 3(c i ical).
This syn ax means a signi ican simpli ica ion compa ed o ha p o ided by
he OpenADR s anda d. Fo ins ance, he la e o e s he possibili y o se ing he
ype o signal in he in e al elemen . Al hough his ea u e is no suppo ed,
he same esul can be achie ed wi h he new syn ax by de ining e en s o each
in e al. The main objec i e o his new syn ax is he sea ch o simplici y and
cla i y wi hou missing impo an unc ionali ies o he expe imen al e alua ion o
he managemen me hods.
113

– Y la na u aleza ¿ ambi´
en es una on e ´
ıa? — p onunci´
o
A kadi mi ando pensa i o a lo lejos, a los campos
abiga ados, que el sol ya en decli e iluminaba he mosa y
sua emen e[. . . ]
— Tambi´
en lo es, en el sen ido que ´
u le das. La
na u aleza no es un emplo, sino un alle , y el homb e es
un abajado del alle .
“Pad es e hijos”, I ´
an Tu gu´
ene . CHAPTER
6
Expe imen al e alua ion o
he ASPEM ole
The ASPEM ole in oduced in Chap e 3, as discussed, is designed o wo k as
a i ual en i onmen in which so wa e agen s can conduc nego ia ions on behal
o cus ome s. This concep ep esen s a signi ican imp o emen o e he classical
a chi ec u e, in which end-nodes a e supposed o be able o pe o m all ype o
unc ions. Speci ically, in he con ex o he Sma G id, besides managing local
esou ces, hey a e also expec ed o access ex e nal da a se ices, p ocess da a se s
and conduc dialogues and nego ia ions. To some ex en , his is he eason why eal
solu ions do no inco po a e ad anced unc ionali ies, bu o he s which a e mo e
p ac ical and less ambi ious. In con as , he ASPEM ole, as a cha ac e is ic inhe -
i ed om he Cloud Compu ing model, p omises o ee end-nodes om demanding
asks. When i comes o OpenADR, his ea u e implies ha end-nodes should only
ha e o suppo he simples p o ile o he p o ocol (p o ile 2.0a) while s ill enjoy-
ing he unc ionali ies and bene i s o he mos ad anced p o iles (2.0b and 2.0c).
One o he goals o he p esen Chap e , which shows he esul s o simula ing he
pa icipa ion o ASPEM nodes in OpenADR p og ams, is o demons a e his con-
e sion skill: o conduc he managemen o he g id by sending signals o he
simples p o ile, ega dless o he p o ile o he signal sen by he sys em ope a o .
121
6. EXPERIMENTAL EVALUATION OF THE ASPEM ROLE
Wi h he aim o demons a ing he bene i s ha using so wa e agen s can b ing
o he cus ome s, he simula ions a e implemen ed wi h pa allel auc ion ma ke s
[PJ05]. As discussed in Chap e 4, his app oach mee s all he condi ions o ene gy
ma ke s (see Sec ion 4.1, page 74); and, no leas o all, i is comple ely dis ibu ed,
so i is close o he u u e ision o he elec ical g id. In pa icula , pa allel auc-
ions gi e he use s au onomy and con ol o e he decision p ocess. Howe e , his
app oach is known o shi he bu den o holding auc ions om he sys em ope a o
o he cus ome s. This Chap e also ul ills he mission o showing how his chal-
lenge can be me by using he concep o b oke agen , which is a so wa e agen
con ac ed as a Cloud Compu ing se ice. In his ega d, i is wo h men ioning
ha his esea ch wo k is he i s o i s kind o simula e pa allel auc ion ma ke s in
he con ex o he elec ical g id, hus gi ing an insigh in o he ac ual e ec i eness
o he mechanism.
The simula ions a e pe o med in he con ex o OpenADR p og ams, his ech-
nology being one o he mos immedia e and ealis ic miles ones o he Sma G id.
To achie e ma ke s in DR p og ams, which by de aul is a con ex in which end-
nodes a e expec ed o me ely apply incoming signals, he model based on easy-
loads and ha d-loads desc ibed in Chap e 2 is used (see Sec ion 3.5, page 69).
Fu he mo e, choosing DR p og ams as simula ion con ex demons a es he abil-
i y o ASPEM nodes o display hei capabili ies in es ic i e en i onmen s.
6.1 Expe imen al e alua ion
Two ypes o expe imen s a e ca ied ou below: he i s one consis s o a cen-
alized app oach ha p io i izes cus ome s’ p e e ences acco ding o hei con-
ac wi h he ASPEM, hus ocusing on he se ices ace o he ASPEM nodes;
while he second pu s o wa d a decen alized solu ion ha shows he capabili y o
ASPEM nodes o manage he e en s h ough ene gy ma ke s. In bo h cases, he
p ocess consis s o he ollowing gene al s eps:
i.The sys em ope a o sends cu ailmen e en s. These can be de ined as signals
o ype simple o del a. In he case o simple signals, he sys em ope a o ex-
pec s ha nodes adop one o he ollowing p ede ined OpenADR le els: no -
122
6.1 Expe imen al e alua ion
mal,mode a e,high o c i ical. As o del a signals, hey exp ess an amoun
o load ha mus be disca ded by each ASPEM.
ii.ASPEM p ocesses cu ailmen e en s. In he solu ion based on he use s’ p io-
i y, he ASPEMs selec he nodes ha pa icipa e in he e en , as well as he
le el o consump ion hey mus apply o hei de ices. On he o he hand, when
ma ke s a e used, ha d-loads ini ia e nego ia ions wi h easy-loads in o de o
a oid applying he le el o consump ion speci ied in he signal.
iii.Cus ome s ecei e signals o ype simple om he ASPEM. The local agen
applies he ac ions co esponding o he le el indica ed by he signal sen by
he co esponding ASPEM. These ac ions consis o swi ching o de ices.
In he expe imen s, OpenADR le els a e implemen ed as ollows:
No mal: No es ic ion is applied, so he household can consume as usual.
Mode a e: HVAC uni s mus be swi ched o .
High: HVAC uni s and wa e -hea e s mus be swi ched o .
C i ical: All uni s, including ligh s, mus be swi ched o .
This de ini ion o le els is oo agg essi e. Ac ually, o eal scena ios he adop-
ion o solu ions ha a y he com o le el is commonly p oposed. To achie e
his, anges o ope a ing poin s a e se o each o he consump ion le els, so use s
do no lose he en i e se ice p o ided by he uni s, bu only pa o he com o
hey p o ide. Howe e , we p e e he abo e de ini ion o simula ions because i
p o ides demand cu es whe e i is easie o di e en ia e he applica ion o each
DR signal, besides helping o keep he ocus on he exchange mechanism.
In bo h ypes o expe imen s, a ypical summe day (Augus 1s , 2000) is simu-
la ed in he scena io IEEE 13-node [Ch 99], which is con igu ed wi h wo ASPEMs
ha ha e p ac ically he same numbe o clien s and he same le el o consump-
ion pe OpenADR le el (Table 6.1) and pe ype o load (Table 6.2). In o de o
con igu e he nodes’ p e e ences, h ee ypes o schedules a e de ined (Lis ing 6.1).
They con igu e he nodes o ac as easy-, ha d- o no mal-loads du ing he whole
pe iod o simula ion. In pa icula , he ha d-loads a e con igu ed o p o ec all hei
demand, so hey s i e o p ese e he no mal s a e o ope a ion; and he easy-loads
a e willing o disca d all he demand. Table 6.3 summa izes he numbe o nodes
associa ed o each ype o schedule.
123
6. EXPERIMENTAL EVALUATION OF THE ASPEM ROLE
schedule easy {
01418*easy(c i ical);
01518*easy(c i ical);
01618*no mal;
};
schedule ha d {
01418*ha d(no mal);
01518*ha d(no mal);
01618*no mal;
};
schedule no mal {
0018*no mal;
};
Lis ing 6.1: Schedules de ini ion o he auc ions ma ke .
no mal
(kW)
mode a e
(kW)
high
(kW)
c i ical
(kW)
aspem 1 61.137 27.103 15.552 0
aspem 2 66.461 30.871 15.194 0
Table 6.1: Consump ion o he ASPEMs pe OpenADR le el.
Easy load
(kW)
Ha d load
(kW)
aspem 1 4.670 32.652
aspem 2 5.819 34.891
Table 6.2: Consump ion o he ASPEMs pe ype o load.
124
6.1 Expe imen al e alua ion
No mal Easy Ha d
aspem 1 109 31 174
aspem 2 107 36 172
Table 6.3: Numbe o nodes associa ed o each ype o schedule.
In absence o any DR e en , he beha io o he demand cu e is ha depic ed
in Figu e 6.1. I shows ha he e is a peak demand in he a e noon, which is espe-
cially high be ween 14:00h and 15:30h. The ollowing expe imen s a e ocused in
his pa icula zone o he cu e, o which OpenADR e en s o ype le el and del a
a e o de ed.
Figu e 6.1: Demand cu e co esponding o 1s Augus , 2000 in he IEEE 13-node
when no signal is applied.
125

6. EXPERIMENTAL EVALUATION OF THE ASPEM ROLE
As will be app ecia ed, he join eco e y o all he consump ion de ices a e
a DR e en p oduces no iceable peaks in he demand cu e. This is a well-known
e ec o which he OpenADR s anda d p o ides an ope a ing p ocedu e. Howe e ,
being beyond he scope o his documen , no mechanism has been implemen ed in
he expe imen s.
I is also wo h men ioning ha , in o de o gi e he sys em ope a o a channel
o bypass he ac ion o ASPEMs, all OpenADR messages wi h p io i y highe han
ze o a e con eyed in ac o he child nodes. This channel can be use ul o handling
si ua ions ha equi e pa icula beha io s, such as eme gencies.
6.1.1 Dispa ch based on p io i ies
In his expe imen , he ASPEM uses an in e nal cen alized algo i hm o de e mine
which nodes ake on he cu ailmen job when an OpenADR signal o ype del a is
ecei ed. The decision is based on he con en o he con ac ha links he use o
he ASPEM. Fo his speci ic expe imen , we assume ha use s can speci y in hei
con ac :
i. The ime pe iods a which hey wan o pa icipa e as ha d-, easy- and no mal-
loads.
ii. A p io i y le el ha he ASPEM uses o de e mine which nodes a e used o
mee he cu ailmen signal. The highe he alue o he pa ame e , he mo e
he p e e ences de ined by he cus ome a e conside ed.
Acco ding o he syn ax o he ad-hoc plugin de eloped o G idLAB-D (see
Sec ion 5.3.1, page 101), he abo e in o ma ion is exp essed by using elemen s o
ype “ASBox” as ollows:
objec ASBox {
name i1B645;
aspem aspem01;
pa en house1B_ m_B_1_645;
le elsSchedule sch_02;
p io i y 3;
};
126
6.1 Expe imen al e alua ion
When an OpenADR signal is sen , he ASPEM ga he s he p e ious in o ma ion
by asking he b oke agen s 1. Nex , he algo i hm con e s he incoming signal
(which in his case is o ype del a) in o mul iple signals o ype simple ha a e
con eyed o he AS-Boxes. I mus be no ed ha OpenADR e en s a e composed
o in e als, and he e o e he exchange p ocess is ca ied ou o each o hese
in e als.
The i s s ep o he algo i hm is o de e mine which esou ces a e necessa y o
co e he amoun o load speci ied in he del a signal. This in o ma ion is ob ained
by asking each b oke agen i s demand es ima es o each consump ion le el. Nex ,
depending on his esul , h ee se ings a e possible:
i.The amoun can be en i ely co e ed by using easy-loads (Algo i hm 6, Ap-
pendix B). This is he less d ama ic case o he sys em, since all he demand
o disca d is collec ed om nodes ha a e ac ually willing o shed i . In his
case, he algo i hm so s he easy-loads in descending o de acco ding o he
p io i y alue. Nex , he algo i hm, ollowing he es ablished o de , akes he
minimum numbe o loads necessa y o co e he amoun speci ied by he del a
signal.
ii.The amoun can be co e ed wi hou using ha d-loads (Algo i hm 7, Appendix
B). In his case, i is possible o co e he del a amoun by doing ha no-ha d
loads adop one o he p ede ined le els o he OpenADR s anda d (mode a e,
high o c i ical). The i s s ep in his case is o calcula e he le el ha he
no-ha d loads ha e o adop . This lis o nodes is so ed in ascending o de
acco ding o he p io i y alue. In addi ion, as in he p e ious case, a lis o
easy-loads is buil . In o de o co e he del a amoun , i s ly, he lis o easy-
loads is used; nex , he no-ha d loads a e aken ( ollowing he o de o he lis )
un il he del a amoun is co e ed.
iii.The amoun has o be co e ed using ha d loads (Algo i hm 8, Appendix B).
Fi s ly, all he no-ha d loads a e equi ed o adop he c i ical le el. Nex , he
lis o ha d-loads is so ed in ascending o de acco ding o he p io i y alue.
Nodes om his lis a e o de ed o adop he c i ical le el un il co e ing he
del a amoun .
1Since his is a cen alized solu ion, he in o ma ion could also be ob ained di ec ly om he
da abase sys em.
127
6. EXPERIMENTAL EVALUATION OF THE ASPEM ROLE
Figu e 6.2 illus a es he case in which he ope a o o de s a educ ion o 30.000
kW be ween 14:00h and 15:30h. In his scena io, he ope a o aims o educe
he equi ed amoun and a oid aking he mode a e le el. In gene al, when he
di e ence o demand be ween he p ede ined OpenADR le els is signi ican , del a
signals enable o apply educ ions wi h esul s ha a e mo e accu a e and equi e
less in ol emen o clien s. Table 6.4 shows ha , o handling he e en , only easy-
loads a e used, so he e is no need o he nodes wo king as ha d-loads o con ibu e
in he load educ ion. Table 6.5 shows he numbe o signals o each ype used o
gene a e he expec ed esul .
Figu e 6.2: Demand cu e co esponding o 1s Augus , 2000 in IEEE 13-node when
a del a signal o 30.000 kW is applied.
Mo eo e , Figu e 6.3 illus a es he example in which a del a signal o 75.000
kW is applied. I aims o a oid he need o he adop ion o he high le el. Due o
128
6.1 Expe imen al e alua ion
Del a Co e ed
(kW)
No mal
(kW)
Easy
(kW)
Ha d
(kW)
aspem 1 14.465 14.569 9.899 4.670 0
aspem 2 15.534 15.680 9.861 5.819 0
Table 6.4: Consump ion o he ASPEMs pe le el when a del a signal o 30.000 kW
is applied.
No mal Mode a e High C i ical
aspem 1 187 91 36 0
aspem 2 204 80 31 0
Table 6.5: Numbe o signals pe ype when a del a signal o 30.000 kW is applied.
he huge size o he educ ion, as shown in Table 6.6, he ha d-loads a e in ol ed
in he ac ion. Howe e , in ela ion o he o al capaci y o his ype o loads, hei
pa icipa ion is ac ually limi ed: only he 11% o hem a e used in his case, and
always a e all he capaci y co esponding o he no mal- and easy-loads has been
used. Table 6.7 shows he numbe o signals pe ype ha a e applied in his case.
All nodes o which a signal o le el no mal is sen a e ac ually ha d-loads ha ha e
con ac ed he maximum ype o p o ec ion o he co esponding ASPEM.
Del a Co e ed
(kW)
No mal
(kW)
Easy
(kW)
Ha d
(kW)
aspem 1 36.163 36.295 27.996 4.670 3.628
aspem 2 38.836 38.977 29.487 5.819 3.671
Table 6.6: Amoun o load pe ype when a del a signal o 75.000 kW is applied.
In bo h cases, he cha s show ha he ASPEM ole is able o handle del a sig-
nals by using use p ope ies, which in his case, o he sake o simplici y, a e
129
C. RESUMEN EN ESPA ˜
NOL
equie e que el hoga es inja su consumo; mode a e, las unidades de ai e acondi-
cionado deben desconec a se; high las unidades de ai e acondicionado y los e mos
deben desconec a se; c i ical, odos los disposi i os de consumo, incluidas las lu-
ces, deben desconec a se. En ealidad, es a de inici´
on de las se˜
nales es demasiado
ag esi a, de o ma que, en escena ios eales, es com´
un p opone implemen aciones
que a ´
ıan el ni el de con o . No obs an e, en es e abajo se op a po una de i-
nici´
on como la expues a po que pe mi e cen a la a enci´
on en los mecanismos de
ges i´
on, y po que asimismo p opo ciona cu as de demanda en las que es m´
as ´
acil
dis ingui el e ec o de las se˜
nales aplicadas.
Los agen es b ´
oke se comunican con el agen e local a a ´
es del p o ocolo
XMPP, siendo es e uno de los mecanismos p opues os en el es ´
anda OpenADR.
Po an o, como esul ado, la in aes uc u a de simulaci´
on incluye un se ido
XMPP. O o componen e a des aca de la in aes uc u a es el eposi o io de es ima-
ciones de demanda. Es e es necesa io pa a que los nodos dispongan de indicado es
que les in o men de la can idad ap oximada de ene g´
ıa que consumi ´
an du an e un
pe ´
ıodo espec´
ı ico de iempo y pa a cada uno de los ni eles de consumo posibles.
La Figu a C.8 mues a los componen es p incipales de la in aes uc u a de
simulaci´
on, incluyendo cada uno de los componen es so wa e que se usan pa a
implemen a los
C.3.7 Simulaci´
on de me cados de subas as pa alelas usando no-
dos ASPEM
Como se puede conclui del an´
alisis del es ado del a e, el mecanismo que mejo se
adap a a la na u aleza dis ibuida del Sma G id, y que asimismo cumple odos los
equisi os de los me cados de ene g´
ıa, son las subas as pa alelas in e sas. Bajo es e
esquema, un usua io que desea p o ege su demanda es un usua io que o ece blo-
ques de demanda en el me cado (subas ado o endedo ); mien as que un usua io
dispues o a o ece pa e de su demanda es un usua io dispues o a puja pa a cub i
la demanda de e ce os (comp ado ). De es a o ma, usando los concep os de ini-
dos en la Secci´
on C.3.4, las ca gas de ipo ha d son ´
ıpicas de los subas ado es, y
las de ipo easy de los pos o es. En es e sen ido, se debe no a que el ol desem-
pe˜
nado po el agen e puede cambia en e me cado y me cado, cuya du aci´
on en
232

C.3 Apo aciones o iginales
ASPEM
Con enedo FIPA
Agen es B óke
Simulado del
G id Eléc ico
AS-Box
Agen es Locales
Ja a Web Applica ion
Jade
G idLAB-D
OpenADR sob e XMPP
Módulo G idLAB-D
(AgencySe ices)
Reposi o io
es imaciones Se ido
XMPP Open i e
Pos g eSQL
Figu a C.8: Componen es de la in aes uc u a de simulaci´
on.
las simulaciones se es ablece en 30 minu os. Po consiguien e, los e en os con una
du aci´
on supe io a 30 minu os se ges ionan a a ´
es de secuencias de me cados.
Despu´
es de ecibi un e en o OpenADR, los ASPEM ins ancian los agen es
b ´
oke que ep esen an a los clien es en la ges i´
on del e en o. A con inuaci´
on, cada
b ´
oke egis a en un di ec o io FIPA los oles que desempe˜
na ´
a en cada uno de los
me cados que componen el e en o, los cuales pueden se : endedo (subas ado ),
comp ado o ninguno (si decide no pa icipa ). Pos e io men e, los consumido es
( endedo es) consul an el di ec o io pa a ob ene la lis a de p oduc o es (comp a-
do es) disponibles. Los consumido es que no encuen an su icien es p oduc o es
pa a cub i un ni el comple o de la se˜
nal OpenADR (mode a e,high oc i ical)
deben cancela la subas a y aplica la se˜
nal OpenADR de en ada.
Despu´
es de ecibi las in i aciones, los p oduc o es deciden en qu´
e subas as
pa icipa . Es impo an e acla a que los p oduc o es no pueden pa icipa en odas
las subas as simul ´
aneamen e po que es o segu amen e implica ´
ıa puja po encima
de sus posibilidades eales. Es deci , supond ´
ıa lle a a cabo una es a egia basa-
da en el concep o de o e booking, la cual supone un iesgo pa a el comp ado y,
233
C. RESUMEN EN ESPA ˜
NOL
en gene al, pa a el sis ema. Po ´
ul imo, los consumido es deciden qu´
e o e as son
acep adas. Como esul ado, los p oduc o es a los que se hayan acep ado o e as
debe ´
an adop a ni eles de consumo m´
as es ic i os, mien as que los consumi-
do es que hayan log ado ce a acue dos pod ´
an aumen a o man ene su ni el de
consumo en e a la se˜
nal OpenADR de en ada.
Las simulaciones se ealizan en un d´
ıa ´
ıpico de e ano (1 de agos o de 2000)
usando el escena io es ´
anda IEEE-13. En es e se incluyen dos ASPEM que p ´
ac i-
camen e ges ionan el mismo n´
ume o y ipo de nodos (Tabla C.2), los cuales se
con igu an pa a ac ua como ca gas de ipo no mal,easy oha d du an e el pe ´
ıodo
comple o que aba ca cada me cado. En conc e o, las ca gas ha d se con igu an pa a
p o ege oda su demanda (in en an conse a su consumo habi ual); mien as que
las ca gas easy se de inen pa a es a dispues as a desca a , si es necesa io, oda su
demanda.
no mal
(kW)
mode a e
(kW)
high
(kW)
c i ical
(kW)
Easy load
(kW)
Ha d load
(kW)
aspem 1 61.137 27.103 15.552 0 4.670 32.652
aspem 2 66.461 30.871 15.194 0 5.819 34.891
Tabla C.2: Consumo de los ASPEM po cada ni el de consumo p opio de las se˜
nales
de ipo simple de OpenADR.
La Tabla C.3 desc ibe el pe il del me cado. Como indica su con enido, cuando
el alo de la se˜
nal es mode a e, la elaci´
on en e la capacidad de puja y la can idad
subas ada (en adelan e Rca) es 2,29. Po consiguien e, cuando los subas ado es no
es ablecen p ecio de en ada, se puede a i ma que la o e a duplica la demanda. La
abla ambi´
en mues a que, en es e caso, la elaci´
on en e el n´
ume o de p oduc o es
y consumido es (en adelan e Rnp) es 3,10. Cuando el alo de la se˜
nal es ha d, la
si uaci´
on es menos ideal, ya que Rca es igual a 0,91. Es deci , la can idad subas ada
es supe io a la capacidad de comp a; o, lo que es lo mismo, la o e a es mayo
que la demanda. Se debe no a que odos es os alo es son meno es cuando exis en
234
C.3 Apo aciones o iginales
p ecios de en ada, pues o que la o e a de un comp ado puede no se ´
alida pa a
odas las subas as.
Se˜
nal
Demanda
(kW)
O e a
(kW) Rca Rnp
mode a e 15271 35114 2,29 3,10
ha d 20416 18569 0,91 2,27
Tabla C.3: Pe il del me cado de subas as.
El me cado de subas as pasa po las siguien es e apas:
i. Anuncio: los ASPEM in o man a los agen es b ´
oke de la ins anciaci´
on de
un nue o ciclo de me cados pa a ges iona e en os DR. Espec´
ı icamen e, se
in o ma ace ca de la du aci´
on del e en o, la du aci´
on de los me cados y los
plazos de las e apas subsiguien es.
ii. Regis o: los agen es b ´
oke egis an en un di ec o io FIPA el ol que desem-
pe˜
na ´
an en cada me cado, que puede se endedo (subas ado ), comp ado o
ninguno si deciden no pa icipa . Al a a se de subas as in e sas, el p ime ol
es p opio de los consumido es y el segundo de los p oduc o es. Adem´
as, los
agen es egis an in o maci´
on espec´
ı ica a cada ol: los subas ado es de inen el
p ecio m´
aximo al que es ´
an dispues os a comp a ; y los comp ado es de inen
la can idad m´
axima de ca ga que es ´
an dispues os a suminis a .
iii. O e a: los subas ado es consul an el di ec o io FIPA pa a localiza p oduc-
o es e in i a los a que pujen en sus subas as. Al inal de es a e apa, odos
los subas ado es que no hayan encon ado su icien e o e a, as´
ı como odos
los p oduc o es que no hayan encon ado subas ado es, in o man al ASPEM y
cancelan su pa icipaci´
on en el me cado. Es os nodos es ´
an obligados a aplica
la se˜
nal OpenADR que en i´
o o iginalmen e el ope ado del sis ema.
i . Subas a: en e odas las o e as ecibidas, los p oduc o es deciden en qu´
e con-
jun o de subas as pa icipa . Las o e as se en ´
ıan en o ma de unciones linea-
les a ozos como la ep esen ada en la Figu a C.9. Cada ni el de la unci´
on
se co esponde con un ni el de consumo que el p oduc o o ece (mode a e,
235
C. RESUMEN EN ESPA ˜
NOL
high,c i ical). El n´
ume o de secciones de la unci´
on depende del alo de la
se˜
nal OpenADR de ipo simple que en ´
ıa el ope ado , el cual ma ca el ni el de
inicio.
. Resoluci´
on: los consumido es deciden qu´
e o e as acep a . El algo i mo im-
plemen ado en es e abajo selecciona las o e as con los p ecios m´
as bajos.
Pa a ello, en p ime luga , el algo i mo o dena odas las o e as seg´
un el p e-
cio, en endiendo po o e a una secci´
on de la unci´
on lineal a ozos en iada
po el p oduc o . A con inuaci´
on, acep a o e as de la lis a de o ma i e a i a
has a cub i oda la demanda. Po ´
ul imo, el consumido in o ma a cada agen e
sob e el esul ado de la subas a.
i. Cie e: los ASPEM egis an odos los acue dos que han sido ce ados y, con-
o me a ellos, en ´
ıa a cada agen e b ´
oke la se˜
nal OpenADR que es e debe
aplica . Es a se˜
nal es ansmi ida a los agen es locales pa a que la apliquen
sob e los ecu sos locales. En gene al, los p oduc o es que hayan log ado en-
de bloques de ene g´
ıa debe ´
an adop a ni eles m´
as es ic i os de consumo,
y los consumido es que hayan ce ado acue dos pod ´
an man ene o incluso
inc emen a su demanda.
Figu a C.9: Funci´
on lineal a ozos que en ´
ıan los p oduc o es a los consumido es
pa a ep esen a sus o e as.
La Figu a C.10 ilus a el esul ado de una simulaci´
on en la que el ope ado o -
dena una se˜
nal de ipo mode a e que comienza a las 2:00 pm y e mina a las 3:30
pm. Como esul ado de los in e cambios en e los nodos, el me cado gene a una
cu a de demanda muy simila a la co espondien e al ni el mode a e, siendo es e
236
C.3 Apo aciones o iginales
el e ec o espe ado. Po su pa e, la Figu a C.11 mues a el esul ado de la simu-
laci´
on pa a una se˜
nal de ipo ha d. En ambos casos, el esul ado que se ob iene
usando mecanismos de me cado es segu o po que, cuando un subas ado no log a
su icien es o e as pa a cub i un ni el en e o de demanda, es e es obligado a can-
cela la subas a y a aplica la se˜
nal OpenADR de en ada. Con es e compo amien o
se log a que la demanda, como m´
aximo, iguale el ni el de consumo o denado po
el ope ado .
Figu a C.10: Cu as de demanda cuando se aplica una se˜
nal OpenADR de ipo mo-
de a e.
La Tabla C.4 mues a los da os co espondien es a las simulaciones. La colum-
na “Can idad in e camb.” hace e e encia al po cen aje de la can idad subas ada
que, g acias a los in e cambios del me cado, ha log ado se cubie a. Como se pue-
de obse a , el po cen aje es signi ica i amen e mayo cuando no exis en p ecios
de en ada. Es o se debe a que, en es e caso, las o e as de odos los p oduc o es
son ´
alidas pa a odas las subas as. No obs an e, el hecho m´
as des acable es que,
237

C. RESUMEN EN ESPA ˜
NOL
Figu a C.11: Cu as de demanda cuando se aplica una se˜
nal OpenADR de ipo ha d.
238
C.3 Apo aciones o iginales
pese a que un n´
ume o conside able de subas as se deben cancela po no ecibi su-
icien es o e as, una can idad impo an e de o e as no log an pa icipa en ning´
un
in e cambio. La az´
on pa a es e hecho apa en emen e con adic o io se halla en la
mala dis ibuci´
on de las o e as en e las subas as. En conc e o, si muchos p o-
duc o es eligen pa icipa en el mismo conjun o de subas as, muchas o as quedan
excluidas de las negociaciones, no ecibiendo su icien es o e as pa a cub i su de-
manda, y, po an o, debiendo se canceladas. En las simulaciones p esen adas en
es e apa ado los p oduc o es seleccionan las subas as de o ma alea o ia. Bajo es a
con igu aci´
on, la mayo pa e de las subas as eciben pujas, pe o los ac o es Rnp
yRca no son lo su icien emen e al os como pa a p opicia que odas las subas as
eciban el n´
ume o m´
ınimo de o e as de p oducci´
on que equie en. La ine icien-
cia que causa la mala dis ibuci´
on de los pa icipan es en los me cados de subas as
pa alelas se es udia en el siguien e apa ado de es e ap´
endice.
Con p ecio de en ada Sin p ecio de en ada
Can idad
in e camb.
(kW)
Subas as
cubie as
( %)
O e a
endida
( %)
Can idad
in ecamb.
(kW)
Subas as
cubie as
( %)
O e a
endida
( %)
mode a e 5798 37.97 18.06 12930 84.67 36.82
ha d 3334 16.33 23.80 3843 18.82 20.70
Tabla C.4: Da os co espondien es a las simulaciones de los me cados de subas as
pa alelas cuando se aplican las se˜
nales mode a e yha d.
El ope ado ambi´
en puede en ia se˜
nales de ipo del a, las cuales en ez de
o dena la adopci´
on de un ni el p ede inido de consumo, indican la can idad exac a
de demanda que se debe desca a . Pa a ges iona es a clase de se˜
nales en la esis se
usa el pa ´
ame o p io idad, cuyo alo pueden aco da los usua ios con el ASPEM
como pa e del con a o. En es e caso, el p ime paso del p ocedimien o es c ea
una lis a de las ca gas compues a de es secciones. Es as, espec i amen e, se co-
esponden con los ipos de ca ga easy,no mal yha d. Las p ime as dos secciones,
omando la p io idad como alo de e e encia, se o denan en o ma descenden e,
239
C. RESUMEN EN ESPA ˜
NOL
mien as que la secci´
on co espondien e a las ca gas de ipo ha d se o dena de o -
ma ascenden e. A con inuaci´
on, se i e a sob e la lis a has a cub i la can idad de
demanda de inida en la se˜
nal del a. En la Figu a C.12 se mues a un caso en el que
el ope ado o dena una educci´
on de 30.000 kW en e las 2:00 pm y las 3:30 pm.
El obje i o de es a se˜
nal es e i a que los nodos deban adqui i el ni el mode a e,
el cual, en conjun o, esul a ´
ıa m´
as es ic i o pa a los usua ios.
Figu a C.12: Cu as de demanda cuando se aplica una se˜
nal OpenADR de ipo del a
que o dena una educci´
on de 30.000 kW.
Los esul ados demues an que usando el modelo de Se icios de Agencia se
pueden ges iona se˜
nales OpenADR a a ´
es de me cados. Es o, en con aposici´
on
con el esquema cl´
asico, posibili a que los usua ios puedan pa icipa ac i amen e en
el p oceso de ges i´
on y, con ello, de ende sus in e eses. Los expe imen os ambi´
en
si en pa a cons a a que el modelo de Se icios de Agencia acili a que, sin pe de
capacidad de ges i´
on, los nodos clien e s´
olo engan que p ocesa se˜
nales de ipo
simple, as´
ı simpli icando no ablemen e su in aes uc u a.
240
C.3 Apo aciones o iginales
C.3.8 Dis ibuci´
on de los comp ado es en me cados de subas as
pa alelas
En en o nos g andes, como pueden se el Sma G id y las edes compu aciona-
les, un comp ado puede ecibi cien os o miles de in i aciones pa a pa icipa en
subas as, ya que los endedo es es ´
an in e esados en inc emen a la pa icipaci´
on,
y con ella la compe i i idad. Cuando exis en m´
as comp ado es que endedo es, la
con igu aci´
on ideal es que la pa icipaci´
on de los p ime os se dis ibuya de ma-
ne a uni o me en e las subas as de los segundos. Sin emba go, es a condici´
on no
se puede espe a en en o nos donde los comp ado es son agen es independien es,
au ´
onomos e in e esados en sus p opias me as. Adem´
as, a ello se ha de suma que
pueden exis i ac o es obje i os que conduzcan a los comp ado es a p e e i al-
gunas subas as sob e o as. En pa icula , en es e ´
ul imo caso cabe espe a que los
comp ado es adop en es a egias a la ho a de selecciona las subas as en las que
desean pa icipa y que, adem´
as, es as es a egias sean compa idas, ocasionando
as´
ı que los comp ado es se acaben concen ando en un peque˜
no g upo de subas as.
Pa a sol en a es e p oblema, en es a esis se desa olla el m´
e odo HUDP (Hash-
based Uni o m Dis ibu ion o Playe s), el cual es ´
a inspi ado en el uncionamien o
de las ablas hash. El compo amien o b´
asico de HUDP se esume en dos pasos:
i. Los comp ado es se egis an en HUDP, de o ma que a cada uno de ellos se
le asigna un iden i icado .
ii. Cada comp ado accede al HUDP pa a ob ene la lis a de subas as en las que
puede pa icipa .
La p esencia de un mecanismo como HUDP puede al e a las condiciones de
la compe ici´
on y, po an o, es necesa io de ini eglas que ga an icen no mas esen-
ciales. En la esis se p oponen las siguien es cua o P e oga i as:
1. Si en la con igu aci´
on o iginal un comp ado puede ecibi su icien es in i a-
ciones pa a aloja oda su capacidad, el mecanismo debe p ese a es a condi-
ci´
on.
2. Si en la con igu aci´
on inicial un endedo puede ecibi pujas su icien es pa a
ende oda su o e a, en onces el mecanismo debe p ese a es a condici´
on.
3. El mecanismo no puede ac ua en pe juicio de un pa icipan e de o ma deli-
be ada.
241
C. RESUMEN EN ESPA ˜
NOL
sin HUDP con HUDP
Alea o io Es a egia Alea o io Es a egia
In e cambiado (kW ) 10612 492 11672 11521
O e a cubie a 84 % 4 % 93 % 91 %
Subas as canceladas 22 1 11 12
Subas as ac´
ıa 22 245 0 0
Tabla C.6: HUDP: Escena io #1: Resul ados pa a los casos en los que no se usan
p ecios de en ada.
iende a segui una dis ibuci´
on uni o me. Sin emba go, des aca que, pese a la g an
capacidad de o e a, exis e un n´
ume o signi ica i o de subas as ac´
ıas (subas as que
no eciben ninguna o e a) y subas as canceladas (subas as que no eciben o e as
su icien es como pa a aloja oda la demanda subas ada). Los esul ados mejo an
cuando se usa HUDP. En conc e o, g acias a la mejo dis ibuci´
on de los comp a-
do es, no quedan subas as ac´
ıas y el n´
ume o de subas as canceladas se educe a la
mi ad. Po o a pa e, cuando los comp ado es ienen incen i os pa a concen a se,
HUDP p ´
ac icamen e log a anula su e ec o; po el con a io, cuando HUDP no se
usa y los comp ado es ienden a ag upa se, el n´
ume o de in e cambios, y po an o
la e iciencia del sis ema, se educe d ´
as icamen e.
Como se mues a en la Tabla C.7, cuando las subas as es ablecen p ecio de
en ada, el mecanismo de dis ibuci´
on ambi´
en log a mejo a la e iciencia del sis-
ema, anulando en g an medida la concen aci´
on de los comp ado es cuando es os
adop an es a egias comunes.
En el Escena io 6 la capacidad de comp a es in e io a la can idad subas ada,
y asimismo el n´
ume o de comp ado es es in e io al n´
ume o de endedo es. Como
se explic´
o, en es os casos la ´
unica mejo a que se puede log a es e i a la concen-
aci´
on de los comp ado es. Los da os de las ablas C.8 y C.9 demues an que se
log a es e obje i o y que, adem´
as, cuando los comp ado es no ienen incen i os
pa a ag upa se, el mecanismo no causa ning´
un pe juicio.
248

C.3 Apo aciones o iginales
sin HUDP con HUDP
Alea o io Es a egia Alea o io Es a egia
In e cambiado (kW ) 8493 950 9949 8727
O e a cubie a 67 % 7 % 79 % 69 %
Subas as canceladas 31 8 33 49
Subas as ac´
ıas 41 222 1 4
Tabla C.7: HUDP: Escena io #1: Resul ados pa a los casos en los que se usan p ecios
de en ada y pujas en o ma de unciones lineas a ozos.
con HUDP sin HUDP
Alea o io Es a egia Alea o io Es a egia
In e cambiado (kW ) 11663 487 11535 11565
O e a cubie a 45 % 2 % 45 % 45 %
Subas as canceladas 249 3 285 164
Subas as ac´
ıas 179 739 177 271
Tabla C.8: HUDP: Escena io #6: Resul ados pa a los casos en los que no se usan
p ecios de en ada.
249
C. RESUMEN EN ESPA ˜
NOL
con HUDP sin HUDP
Random S a egy Random S a egy
In e cambiado (kW) 10051 752 10824 10883
O e a cubie a 39 % 3 % 42 % 42 %
Subas as canceladas 168 10 285 94
Subas as ac´
ıas 314 715 177 431
Tabla C.9: HUDP: Escena io #6: Resul ados pa a los casos en los que se usan p ecios
de en ada y pujas en o ma de unciones lineas a ozos.
La Figu a C.18 ilus a el endimien o de HUDP en cada uno de los escena-
ios. La endencia gene al es que el ni el de mejo a log ado po HUDP se eduzca
cuando ambi´
en lo hace Rca, ya que a menos pa icipan es, menos posibilidades de
log a una mejo dis ibuci´
on de ellos. El aspec o m´
as des acable es que el e ec o
de las es a egias es anulado en g an pa e. Asimismo, o o e ec o isible de HUDP
es que esul a inocuo cuando los comp ado es no adop an es a egias. El g ´
a ico
ambi´
en mues a que, cuando no se adop an es a egias y Rca no iene asociado un
alo al o, no exis e cla o ganado . Es o se debe a que en es e caso HUDP no iene
espacio pa a abaja .
Pa a conclui , se puede a i ma que HUDP: (i) anula casi po comple o la con-
cen aci´
on de comp ado es cuando es os se gu´
ıan po es a egias comunes; (ii) e-
duce el n´
ume o de subas as canceladas; y (iii) esul a inocuo cuando no iene po-
sibilidad de ac ua . Adem´
as, como se ha expues o, el uncionamien o de HUDP
se basa en ope aciones sencillas e independien es, de o ma que sopo a accesos
concu en es.
C.4 Conclusiones
Muchas de las ca ac e ´
ıs icas que habi ualmen e se a ibuyen al Sma G id deman-
dan la implan aci´
on de un sis ema de ges i´
on dis ibuido. Es os son p oyec ados
de o ma que los pun os de p oducci´
on y consumo, ep esen ados po unidades de
con ol in eligen e, son capaces de plani ica y negocia sus acciones di ec amen e
250
C.4 Conclusiones
Figu a C.18: His og ama de la can idad de o e a cubie a cuando se usa el mecanismo
HUDP y los pa icipan es adop an es a egias.
251
C. RESUMEN EN ESPA ˜
NOL
con el es o de en idades. En la p ´
ac ica, la implemen aci´
on de es e modelo, conoci-
do como SDM, adquie e la o ma de un me cado de ene g´
ıa que se ca ac e iza po
se ins anciado bajo demanda y de co a du aci´
on. Pa a hace posible la pa icipa-
ci´
on au ´
onoma y au oma izada de los usua ios en es a clase de en o nos se p opone
el uso de agen es in eligen es, los cuales, a p io i, e´
unen odas las ca ac e ´
ıs icas
eque idas. Sin emba go, como se desc ibe en es e documen o, los agen es in eli-
gen es no han log ado el ´
exi o espe ado en en o nos simila es, haciendo necesa io
la elabo aci´
on de nue os modelos de in e acci´
on y despliegue. Adem´
as, como am-
bi´
en concluye es e documen o, esponsabiliza a los disposi i os de con ol locales
de a eas de negociaci´
on, coo dinaci´
on y acceso a da os in oduce e os es uc u-
ales y ecnol´
ogicos que con a ienen muchas de las cualidades que se espe an
del Sma G id. Con el obje i o de supe a es a ba e a y de ga an iza un sis ema
el´
ec ico eac i o, lexible y iable, es a esis, inspi ´
andose en la expe iencia ob e-
nida en ´
a eas simila es, desa olla el modelo de Se icios de Agencia. Es e libe a
a los nodos locales de oda la complejidad que implica pa icipa en sociedades
i uales y, a a ´
es de una soluci´
on inspi ada en el pa adigma Cloud Compu ing,
log a conse a odas las p opiedades que habi ualmen e se a ibuyen a los agen es
in eligen es. En es e sen ido, cabe des aca que el hecho de ecu i a la analog´
ıa de
campos ecnol´
ogicos simila es, iden i icando an o las soluciones de ´
exi o como los
p oblemas que a on an, ha cons i uido un ecu so de g an u ilidad pa a compone
una soluci´
on nue a y p ´
ac ica.
El modelo de Se icios de Agencia combina la o ien aci´
on a se icios con los
agen es in eligen es, cons i uyendo ambos una soluci´
on que supe a muchos de los
e os de in e acci´
on que p esen a el Sma G id. Po un lado, la o ien aci´
on a se i-
cios acili a el desa ollo de una ed el´
ec ica en la que los clien es puedan con a a
se icios con o me a sus necesidades; mien as que las cualidades de los agen es
so wa e, que el nue o modelo conse a en su o alidad, p opo cionan au onom´
ıa a
los usua ios y, con ello, la capacidad de implemen a sis emas de ges i´
on dis ibui-
dos, eac i os e in eligen es.
El es ado del a e de los algo i mos pa a la ges i´
on de en o nos p opios del
Sma G id e ela que ca ac e ´
ıs icas impo an es de los me cados de ene g´
ıa son
casi siemp e omi idas. En e ellas des aca la necesidad de acep a o e as combi-
nadas de ipo complemen a io y suplemen a io. S´
olo los algo i mos CONSEC y
252
C.4 Conclusiones
mPJ conside an es a uncionalidad; y s´
olo el ´
ul imo de los dos, basado en subas-
as pa alelas in e sas, p opo ciona un en oque dis ibuido que explo a odas las
capacidades de los agen es au ´
onomos. Es a esis es el p ime abajo en simula
el algo i mo mPJ en el en o no de las edes de ene g´
ıa, demos ando su alidez.
Asimismo es el p ime o en des aca el ue e impac o que iene la dis ibuci´
on de
la pa icipaci´
on de los comp ado es en la e iciencia de las subas as pa alelas. Po
o a pa e, la e isi´
on del es ado del a e mues a que la mayo pa e de los a-
bajos no es ´
an basados en es ´
anda es p opios del Sma G id, ni ampoco, en su
mayo ´
ıa, son ep oducibles, lo cual a ec a a la e i icabilidad de las p opues as y,
en consecuencia, a las ga an ´
ıas que o ecen los mismos.
La in aes uc u a de simulaci´
on basada en G idLAB-D y Jade ha demos ado
se e ec i a, siendo capaz de simula en de alle ambos campos de aplicaci´
on: el de
las edes de ene g´
ıa y el de los agen es in eligen es. Adem´
as, el uso de in e aces
bien de inidas pa a la in e comunicaci´
on y sinc onizaci´
on de las dos he amien as
de simulaci´
on ha p obado gene a un dise˜
no limpio y escalable, o aleciendo as´
ı la
apues a po soluciones de simulaci´
on conjun a. Del mismo modo, se ha de des aca
que la sinc onizaci´
on de ambos simulado es, la cual es econocida como la a ea
m´
as compleja en es a clase de soluciones, se io simpli icada po el hecho de que
los dos p oyec os ue an de c´
odigo abie o.
Como demues an los expe imen os, los cuales es ´
an especialmen e dise˜
nados
pa a se ealis as y ep oducibles, el modelo de Se icios de Agencia es capaz de
ins ancia me cados de ene g´
ıa en p og amas DR. En es os en o nos, que en p in-
cipio es ´
an pensados pa a ope a en base a compo amien os p ep og amados, la
p esencia de nodos ASPEM, que incluyen en o nos i uales de negociaci´
on pa a
los agen es so wa e, demues a: p opo ciona au onom´
ıa a los usua ios, simpli ica
la in aes uc u a necesa ia en las ins alaciones del clien e, y acili a la implemen-
aci´
on de me cados guiados po los in e eses de los usua ios. Asimismo, los esul-
ados de los expe imen os demues an que, usando me cados basados en subas as
pa alelas y agen es so wa e au ´
onomos, se puede log a la cu as de demanda co-
espondien es a las se˜
nales DR de en ada. Adem´
as, el modelo de Se icios de
Agencia log a comple a odos sus obje i os espe ando los es ´
anda es del Sma
G id desa ollados po OASIS y NIST.
253

C. RESUMEN EN ESPA ˜
NOL
La concen aci´
on de comp ado es es una condici´
on que puede a ec a se e a-
men e a la e ec i idad de los mecanismos basados en subas as pa alelas. Los expe-
imen os demues an que, cuando su ge es a condici´
on, el sis ema puede o na se
inope a i o, in alidando comple amen e el uso de subas as pa alelas, y con ello uno
de los mecanismos m´
as comple os pa a implemen a me cados en sis emas dis i-
buidos como las edes compu acionales o el Sma G id. Es a esis es el p ime
abajo en es udia en p o undidad es e e ec o y en p opone una soluci´
on al p o-
blema. En conc e o, el mecanismo HUDP, inspi ado po el uncionamien o de las
ablas hash, log a dis ibui los comp ado es de mane a uni o me en e las subas as
sin al e a eglas b´
asicas de los me cados. Adem´
as, es una soluci´
on especialmen e
dise˜
nada pa a en o nos concu en es, dis ibuidos y eac i os, como el Sma G id.
Pa a conclui , se ha de des aca que odo el abajo desa ollado a lo la go de
es a esis cumple con los c´
anones de la in es igaci´
on ep oducible. El p op´
osi o
de ello es que las con ibuciones puedan se con as adas y, en caso de in e ´
es,
ex endidas po o os g upos de in es igaci´
on.
254
Bibliog aphy
[AB00] M. Amin and D. Balla d. De ining new ma ke s o in elligen
agen s. IT P o essional, 2(4):29–35, 2000. 31, 79, 89
[ACC+99] Gul A. Agha, Ka hleen M. Ca ley, Jose Cuena, Edmund H. Du ee,
Cla ence Ellis, Les Gasse , Michael P. Geo ge , Michael N. Huhns
andTo u Ishida, Nadeem Jamali, Sascha Ossowski, H. Van Dyke
Pa unak, Anand S. Rao, Tuomas W. Sandholm, Sandip Sen, Mu-
ninda P. Singh, La y M. S ephens, Ge a d Tel, Jacques Waine ,
Ge ha d Weiss, Michael J. Woold idge, and Mako o Yokoo. Mul-
iagen Sys ems: A Mode n App oach o Dis ibu ed A i icial In el-
ligence. MIT P ess, 1999. 51
[All10] Robe John Allan. Su ey o agen based modelling and simula ion
ools. Technical epo , Science and Technology Facili ies Council
(STFC), 2010. 98
[A n00] Tim A nhei e . Modeling and simula ion o an agen -based decen -
alized wo-commodi y powe ma ke . Mul i-Agen Sys ems, In e -
na ional Con e ence on, 0:361–362, 2000. 31, 80
[ARP+02] C. And ieu, B. Raison, D. Penko , M. Fon ela, S. Bacha, and
N. Hadjsaid. Faul de ec ion, analysis and diagnos ics in high-dg
dis ibu ion sys ems. Technical epo , CRISP, 2002. 25
[ATS04] S ephanos And ou sellis-Theo okis and Diomidis Spinellis. A su -
ey o pee - o-pee con en dis ibu ion echnologies. ACM Com-
pu ing Su eys (CSUR), 36(4):335–371, 2004. 41
255
BIBLIOGRAPHY
[BAV05] Rajkuma Buyya, Da id Ab amson, and S ikuma Venugopal. The
g id economy. P oceedings o he IEEE, 93(3):698–714, 2005. 44
[BBCP05] Fabio Belli emine, Fede ico Be gen i, Gio anni Cai e, and
Agos ino Poggi. Jade — a ja a agen de elopmen amewo k.
In Ra aelH. Bo dini, Mehdi Das ani, J¨
u gen Dix, and Amal Fal-
lah Segh ouchni, edi o s, Mul i-Agen P og amming, olume 15 o
Mul iagen Sys ems, A i icial Socie ies, and Simula ed O ganiza-
ions, pages 125–147. Sp inge US, 2005. 98
[BBD+06] Ra ael H Bo dini, La s B aubach, Mehdi Das ani, Amal El Fallah-
Segh ouchni, Jo ge J Gomez-Sanz, Joao Lei e, G ego y MP
O’Ha e, Alexande Pokah , and Alessand o Ricci. A su ey o p o-
g amming languages and pla o ms o mul i-agen sys ems. In-
o ma ica (Slo enia), 30(1):33–44, 2006. 98
[BCG+98] F. B azie , F. Co nelissen, R. Gus a sson, C. M. Jonke ,
O. Lindebe g, B. Polak, and J. T eu . Agen s nego ia ing o load
balancing o elec ici y use. In The 18 h In e na ional Con e ence
on Dis ibu ed Compu ing Sys ems, page 622. IEEE Compu e So-
cie y, 1998. 31, 87
[Bea04] Da id Boo h and e al. Web Se ice A chi ec u e, Feb 2004. 62
[B a97] Je ey M. B adshaw. So wa e Agen s. AAAI P ess, 1997. 51
[CA07] P. Ca lsson and A. Ande sson. A lexible model o ee-s uc u ed
mul i-commodi y ma ke s. Elec onic Comme ce Resea ch, 7:69–
88, 2007. 82, 88
[Ch 99] Rich Ch is ie. Powe sys ems es case a chi e. h p://www.
ee.washing on.edu/ esea ch/ps ca, 1999. 97, 123
[CL07] Manish Chhab a and Hongen Lu. Towa ds Agen Based Web Se -
ice. In 6 h IEEE/ACIS In e na ional Con e ence on Compu e and
In o ma ion Science, 2007. 62
256
BIBLIOGRAPHY
[CM06] Cama inha-Ma os. Vi ual O ganiza ions, chap e ICT In as uc-
u es o VO, pages 83–103. Sp inge US, 2006. 46
[CMA03] Luis M Cama inha-Ma os and Hamideh A sa manesh. Elemen s
o a base e in as uc u e. Compu e s in indus y, 51(2):139–163,
2003. 46
[CPPS06] G. Celli, F. Pilo, G. Pisano, and G. Soma. Op imal pa icipa ion
o a mic og id o he ene gy ma ke wi h an in elligen ems. In In-
e na ional Powe Enginee ing Con e ence, pages 663–668, 2006.
28
[CRI02] CRISP. Case s udy simula ions and esul s. Technical epo ,
Dis ibu ed In elligence in C i ical In as uc u es o Sus ainable
Powe , 2002. 26, 83
[CSG08] DP Chassin, K Schneide , and C Ge kensmeye . G idLAB-D: An
open-sou ce powe sys ems modeling and simula ion en i onmen .
In T ansmission and Dis ibu ion Con e ence and Exposi ion, 2008.
T&D. IEEE/PES. IEEE, 2008. 97
[cu 15] cu l. cu l and libcu l. h p://cu l.haxx.se, 2015. 111
[DH04] A. L. Dimeas and N. Ha zia gy iou. A mul iagen sys em o mi-
c og ids. In IEEE Powe Enginee ing Socie y Gene al Mee ing,
numbe 55-58, 2004. 31, 77, 89
[DH05] A. Dimeas and N. Ha zia gy iou. Ope a ion o a mul iagen sys-
em o mic og id con ol. IEEE T ansac ions on Powe Sys ems,
20:1447–1455, 2005. 27, 31, 84, 85
[DJ03] Vie Dung Dang and Nicholas R. Jennings. Op imal clea ing al-
go i hms o mul i-uni single-i em and mul i-uni combina o ial
auc ions wi h demand/supply unc ion bidding. In P oceedings o
he 5 h In e na ional Con e ence on Elec onic Comme ce, ICEC
’03, pages 25–30, New Yo k, NY, USA, 2003. ACM. 74, 79, 160
257
BIBLIOGRAPHY
[LKG05] R. Lum, D.B. Ko ak, and W.A. G u e . Mul i-agen coo dina ion o
dis ibu ed ene gy sys ems. In Sys ems, Man and Cybe ne ics, 2005
IEEE In e na ional Con e ence on, olume 3, pages 2584 – 2589,
10-12 2005. 31, 86
[LRHT10] I. Lopez-Rod iguez and M. He nandez-Teje a. Agen -based se -
ices o building ma ke s in dis ibu ed ene gy en i onmen s. In
Renewable Ene gy & Powe Quali y Jou nal (ICREPQ), 2010. 187
[LRHT11a] I. Lopez-Rod iguez and M. He nandez-Teje a. Challenges o us-
ing sma local de ices o he managemen o he sma g id. In
Sma G id Communica ions (Sma G idComm), 2011 IEEE In e -
na ional Con e ence on, pages 73–77, Oc 2011. 187
[LRHT11b] Ignacio Lopez-Rod iguez and Ma io He nandez-Teje a. So wa e
agen s as cloud compu ing se ices. In Y es Demazeau, Michal
Pechoucek, JuanM. Co chado, and Ja ie Bajo P´
e ez, edi o s, Ad-
ances on P ac ical Applica ions o Agen s and Mul iagen Sys ems,
olume 88 o Ad ances in In elligen and So Compu ing, pages
271–276. Sp inge Be lin Heidelbe g, 2011. 187
[LRHT15] I. Lopez-Rod iguez and M. He nandez-Teje a. In as uc u e based
on supe nodes and so wa e agen s o he implemen a ion o
ene gy ma ke s in demand- esponse p og ams. Applied Ene gy,
158:1–11, 2015. x iii, 188, 189
[LRHTHC15] I. Lopez-Rod iguez, M. He nandez-Teje a, and J. He nandez-
Cab e a. Regula ion o he buye s’ dis ibu ion in managemen sys-
ems based on simul aneous auc ions and in elligen agen s. Expe
Sys ems wi h Applica ions, 42(21):8014 – 8026, 2015. x iii, 188,
190
[LRHTa ] I. Lopez-Rod iguez and M. He nandez-Teje a. Me hods o he ma-
nagemen o dis ibu ed ene gy en i onmen s using so wa e agen s
and ma ke mechanisms: A su ey. Elec ic Powe Sys ems Re-
sea ch, o appea . 188
264

BIBLIOGRAPHY
[LSM09] J. Lago se, M.G. Simoes, and A. Mi aoui. A mul iagen uzzy-
logic-based ene gy managemen o hyb id sys ems. Indus y Appli-
ca ions, IEEE T ansac ions on, 45(6):2123–2129, No 2009. 28
[LSS+11] Hua Lin, S. Sambamoo hy, S. Shukla, J. Tho p, and L. Mili. Powe
sys em and communica ion ne wo k co-simula ion o sma g id
applica ions. In Inno a i e Sma G id Technologies (ISGT), 2011
IEEE PES, pages 1–6, Jan 2011. 96
[LSW08] T. Logen hi an, D. S ini asan, and D. Wong. Mul i-agen coo dina-
ion o de in mic og id. In Sus ainable Ene gy Technologies, 2008.
ICSET 2008. IEEE In e na ional Con e ence on, pages 77–82, No
2008. 80
[LXJM12] Ma in L´
e esque, Da Qian Xu, G´
eza Jo´
os, and Ma in Maie . Com-
munica ions and powe dis ibu ion ne wo k co-simula ion o mul-
idisciplina y sma g id expe imen a ions. In P oceedings o he
45 h Annual Simula ion Symposium, ANSS ’12, pages 2:1–2:7, San
Diego, CA, USA, 2012. Socie y o Compu e Simula ion In e na-
ional. 96
[LZ14] Weilin Li and Xiaobin Zhang. Simula ion o he sma g id com-
munica ions: Challenges, echniques, and u u e ends. Compu e s
& Elec ical Enginee ing, 40(1):270 – 288, 2014. 40 h-yea com-
memo a i e issue. 97
[Ma 98] Mau icio Ma ´
ın. Asynch onous ( ime-wa p) e sus synch on-
ous (e en -ho izon) simula ion ime ad ance in bsp. In Da id
P i cha d and Je Ree e, edi o s, Eu o-Pa ’98 Pa allel P ocessing,
olume 1470 o Lec u e No es in Compu e Science, pages 897–
905. Sp inge Be lin Heidelbe g, 1998. 117
[MFPF01] Ma Ta Ma inez, Philippe Foule ie , Ka Ha Pa k, and Joel Fa el.
Vi ual en e p ise–o ganisa ion, e olu ion and con ol. In e na-
ional jou nal o p oduc ion economics, 74(1):225–238, 2001. 45
265
BIBLIOGRAPHY
[MG11] Pe e Mell and Timo hy G ance. The NIST de ini ion o cloud com-
pu ing. Technical epo , Na ional Ins i u e o S anda ds and Tech-
nology (NIST), 2011. 53
[ML07] P. McBu ney and M. Luck. The agen s a e all busy doing s u !
In elligen Sys ems, IEEE, 22(4):6–7, July 2007. 51
[NEGR15] Nada A. Nabeeh, Hai ham A. El-Gha eeb, and A. M. Riad. In eg-
a ing so wa e agen s and web se ices in se ice o ien ed a chi ec-
u e based cloud se ices disco e y amewo k. Jou nal o Cases
on In o ma ion Technology (JCIT), 10:67–79, 2015. 187
[NER11] NERC. Balancing and equency con ol. Technical epo , No h
Ame ican Elec ic Reliabili y Co po a ion (NERC), 2011. 7
[NIS12] NIST F amewo k and Roadmap o Sma G id In e ope abili y
S anda ds, Release 2.0, 2012. 32
[NT07] Mohammad Reza Nami and Djamshid Ta anga ian. Vi ual O gan-
iza ion: A New App oach in IT. In IEEE In e na ional Con e ence
on Compu e and In o ma ion Technology (CIT), olume 1, pages
93–98, 2007. 45
[OAD] OpenADR Alliance. h p://www.openad .o g. 11, 35
[OAD12] Openad 2.0a p o ile speci ica ion, 2012. 35
[OASa] Ene gy in e ope a ion e sion 1.0. 32
[OASb] Ene gy ma ke in o ma ion exchange (emix) e sion 1.0. 34
[OASc] Ws-calenda e sion 1.0. 34
[OJ05] J. Oya zabal and J. Jimeno. Agen based mic o g id managemen
sys em. In Fu u e Powe Sys ems (FPS), pages 16–18, 2005. 30
[O a00] O acle. Ja a Ne wo k Launching P o ocol & API Speci ica ion
(JSR-56), Sep embe 2000. 59
266
BIBLIOGRAPHY
[PAG+09] Pie e, Ma y Ann, Gi ish Gha ika , Sila Kilicco e, Ed Koch, Dan
Hennage, Pe e Palensky, and Cha les McPa land. Open Au o-
ma ed Demand Response Communica ions Speci ica ion (Ve sion
1.0). Technical epo , Law ence Be keley Na ional Labo a o y,
2009. 64
[Pal12] Mau icio Pale a. Cloud Compu ing o Teaching and Lea n-
ing: S a egies o Design and Implemen a ion, chap e In elligen
Clouds: By Means o Using Mul i-Agen Sys ems En i onmen s,
pages 254–279. In o ma ion Science Re e ence, 2012. 187
[PAP] Requi emen s o sma me e upg adeabili y. 68
[PFR09] M. Pipa anasompo n, H. Fe oze, and S. Rahman. Mul i-agen sys-
ems in a dis ibu ed sma g id: Design and implemen a ion. In
Powe Sys ems Con e ence & Exposi ion (PSCE), 2009. 31, 87
[PJ05] Y. K. Penya and N.R. Jennings. Combina o ial ma ke s o e icien
elec ici y managemen . In IEEE/WIC/ACM In Con on In elligen
Agen Technology, pages 626–632, 2005. 79, 88, 89, 122, 131, 178
[PL06] Yoseba Penya Landabu u. Op imal Alloca ion and Scheduling o
Demand in De egula ed Ene gy Ma ke s. PhD hesis, Vienna Uni-
e si y o Technology, 2006. 131
[PLSW06] L. Phillips, M. Link, R. Smi h, and L. Weiland. Agen -based con-
ol o dis ibu ed in as uc u e esou ces. Technical epo , Sandia
Na ional Labo a o ies, 2006. 24, 31
[PS06] Michael Pe e s and Se gei Se e ino . In e ne auc ions wi h many
ade s. Jou nal o Economic Theo y, 130:220–245, 2006. 178
[Rez93] La y Reznick. Using c on and c on ab. Sys Admin, 2(4):29–32,
1993. 105
267
BIBLIOGRAPHY
[RLS+14] S. Rohjans, S. Lehnho , S. Schu e, F. And en, and T. S asse .
Requi emen s o sma g id simula ion ools. In Indus ial Elec-
onics (ISIE), 2014 IEEE 23 d In e na ional Symposium on, pages
1730–1736, June 2014. 97
[RPH98] Michael H. Ro hkop , Aleksanda Peke, and Ronald M. Ha s ad.
Compu a ionally manageable combina ional auc ions. Manage-
men science, 44(8):1131–1147, 1998. 76
[RPT07] S. Rahman, M. Pipa anasompo n, and Y. Teklu. In elligen dis-
ibu ed au onomous powe sys ems (idaps). In IEEE PES Annual
Gene al Mee ing, pages 1–8, 2007. 31, 86
[RSEC11] B. Ramachand an, S.K. S i as a a, C.S. Ed ing on, and D.A.
Ca es. An in elligen auc ion scheme o sma g id ma ke using
a hyb id immune algo i hm. Indus ial Elec onics, IEEE T ansac-
ions on, 58(10):4603–4612, Oc 2011. 78
[RZ94] J. S. Rosenschein and G. Zlo kin. Rules o Encoun e : Designing
Con en ions o Au oma ed Nego ia ion Among Compu e s. The
MIT P ess, 1994. 87
[San02] Tuomas Sandholm. Algo i hm o op imal winne de e mina ion
in combina o ial auc ions. A i icial In elligence, 135(1–2):1–54,
2002. 52, 76, 178
[SBLH06] Nigel Shadbol , Tim Be ne s-Lee, and Wendy Hall. The Seman ic
Web Re isi ed. IEEE In elligen Sys ems, 21(3):96–101, July 2006.
62
[Sch01] Rudige Schollmeie . A de ini ion o pee - o-pee ne wo king
o he classi ica ion o pee - o-pee a chi ec u es and applica i-
ons. Pee - o-Pee Compu ing, IEEE In e na ional Con e ence on,
0:101–102, 2001. 41
268
BIBLIOGRAPHY
[SJ06] Manuel Sanchez-Jimenez. Sma elec ici y ne wo ks based on
la ge in eg a ion o enewable sou ces and dis ibu ed gene a ion.
Mas e ’s hesis, Kassel Uni e si y, 2006. 6
[SS01] Tuomas Sandholm and Subhash Su i. Ma ke clea abili y. In In-
e na ional Join Con e ence on A i icial In elligence, olume 17,
pages 1145–1151. LAWRENCE ERLBAUM ASSOCIATES LTD,
2001. 74, 79, 160, 178
[S a12] Tom S an on. A e sma mic og ids in you u u e? explo ing chal-
lenges and oppo uni ies o s a e public u ili y egula o s. Tech-
nical epo , Na ional Regula o y Resea ch Ins i u e, 2012. 187
[S e14] F. Bush S ephen. Sma G id: Communica ion-Enabled In elli-
gence o he Elec ic Powe G id, chap e Appendix: Sma G id
Simula ion Tools, pages 489–492. John Wiley & Sons, L d, 2014.
97
[S o09] Vic o ia S odden. Enabling ep oducible esea ch: Licensing o
scien i ic inno a ion. In e na ional Jo nal o Communica ions Law
& Policy, 13:1, 2009. 90, 96
[S 05] Ma a S yszowska. Las -minu e and mul iple bidding in simul-
aneous and o e lapping second p ice in e ne auc ions. Cen ER
Wo king Pape Se ies, 2005. 178
[Tal12] D. Talia. Clouds mee agen s: Towa d in elligen cloud se ices.
In e ne Compu ing, IEEE, 16(2):78–81, Ma ch 2012. 187
[Tal14] Domenico Talia. Towa ds in e ne in elligen se ices based on
cloud compu ing and mul i-agen s. In Sal a o e Gaglio and Gi-
useppe Lo Re, edi o s, Ad ances on o he In e ne o Things,
olume 260 o Ad ances in In elligen Sys ems and Compu ing,
pages 271–283. Sp inge In e na ional Publishing, 2014. 187
269

BIBLIOGRAPHY
[TTL05] Douglas Thain, Todd Tannenbaum, and Mi on Li ny. Dis ibu ed
compu ing in p ac ice: he condo expe ience. Concu ency and
Compu a ion: P ac ice and Expe ience, 17(2-4):323–356, 2005. 43
[Vic61] William Vick ey. Coun e specula ion, auc ions, and compe i i e
sealed ende s. The Jou nal o inance, 16(1):8–37, 1961. 79
[W3C] Soap e sion 1.2 pa 0: P ime (second edi ion). 36
[WJ95] Michael Woold idge and Nicholas Jennings. In elligen agen s:
Theo y and p ac ice. Knowledge Enginee ing Re iew, 10:115–152,
1995. 50
[Woo09] Michael Woold idge. An In oduc ion o Mul iAgen Sys ems - Se-
cond Edi ion. John Wiley & Sons, 2009. 51
[XMP] XMPP S anda ds Founda ion. Ex ensible Messaging and P esence
P o ocol (XMPP): Ins an Messaging and P esence. 35, 59, 99
[YA96] F. Ygge and J. M. Akke mans. Powe load managemen as a com-
pu a ional ma ke . In Second In e na ional Con e ence on Mul i-
Agen Sys ems, ICMAS 1996, Kyo o, Japan, pages 393–400. AAAI
P ess, 1996. 81
[YA00] F. Ygge and H. Akke mans. Resou ce-o ien ed mul icommodi y
ma ke algo i hms. Au onomous Agen s and Mul i-Agen Sys ems,
3:53–71, 2000. 81
[YWM+05] Zhihong Ye, R Walling, N Mille , P Du, and K Nelson. Facili y
Mic og ids. Na ional Renewable Ene gy Labo a o y, 2005. 87
[ZCB10] Qi Zhang, Lu Cheng, and Raou Bou aba. Cloud compu ing: s a e-
o - he-a and esea ch challenges. Jou nal o In e ne Se ices and
Applica ions, 1:7–18, 2010. 53
270
BIBLIOGRAPHY
[ZK13] T.A. Mohamed Ziyad and D .K.R. Shanka Kuma . Agen based
in elligen sys ems: Gene al iews om lea ning and esea ch pe -
spec i e. A i icial In elligen Sys ems and Machine Lea ning, 5(4),
2013. 187
271
Decla a ion
I he ewi h decla e ha I ha e p oduced his wo k wi hou he p ohibi ed as-
sis ance o hi d pa ies and wi hou making use o aids o he han hose spe-
ci ied; no ions aken o e di ec ly o indi ec ly om o he sou ces ha e been
iden i ied as such. This wo k has no p e iously been p esen ed in iden ical
o simila o m o any examina ion boa d.
The disse a ion wo k was conduc ed unde he supe ision o F ancisco
Ma io He n´
andez Teje a a he Uni e si y o Las Palmas de G an Cana ia.
This page is in en ionally le blank