A me hodological esea ch on so wa e enginee ing applied o he design o Sma
G ids using a Complex Sys em app oach
Au ho : Jos´
e´
E o a-G´
omez
Ad iso : F ancisco Ma io He n´
andez-Teje a
Ad iso : Jos´
e Juan He n´
andez-Cab e a
Tex p in ed in Las Palmas de G an Cana ia
Fi s edi ion, Oc obe 2014
Dedico es a esis a mis se es m´
as que idos.
A mi mad e, mi pad e y mi he mano Chenko
po apoya me siemp e. A Paula po da me ´
animos,
escucha me y much´
ısimo m´
as. A mi abuela Mamen y
mi ´
ıa Ma ga i a que siemp e se han p eocupado po mi.
CONTENTS
9 Simula ion esul s analysis 101
9.1 Business In elligence me hodologies o analysing da a . . . . . . 103
9.2 On-Line Analy ical P ocessing o Sma G ids . . . . . . . . . . 106
10 Demand Side Managemen policy design 115
10.1 Swa ming In elligence echniques o designing policies . . . . . 116
10.2 Pa icle Swa m Op imisa ion . . . . . . . . . . . . . . . . . . . . 116
10.3 Mul i Objec i e Op imisa ion . . . . . . . . . . . . . . . . . . . . 117
10.4 Mul i Objec i e Pa icle Swa m Op imisa ion . . . . . . . . . . . 119
IV Expe imen a ion 121
11 Expe imen a ion conside a ions 123
11.1 Modelling in Ta a . . . . . . . . . . . . . . . . . . . . . . . . . 124
11.2 Expe imen s compa ing di e en simula ion imings . . . . . . . 132
12 Agen -Based Modelling o Elec ical Load a Household Le el 137
12.1Desc ip ion ............................. 138
12.2CaseS udy ............................. 141
12.3Resul s................................ 142
12.4Discussion.............................. 146
13 Modelling li es yle aspec s in luencing he esiden ial load-cu e 149
13.1Desc ip ion ............................. 150
13.2Cases udy.............................. 151
13.3Resul s................................ 154
13.4Discussion.............................. 158
14 A Mul i-Objec i e Pa icle Swa m Op imisa ion me hod o Di ec Load
Con ol in Sma G id 161
14.1Desc ip ion ............................. 162
14.2CaseS udy ............................. 169
14.3Resul s................................ 173
14.4Discussion.............................. 178
III

CONTENTS
15 A la ge-scale elec ical g id simula ion o massi e in eg a ion o dis-
ibu ed pho o ol aic ene gy sou ces 181
15.1Desc ip ion ............................. 182
15.2Cases udy.............................. 184
15.3 Simula ion esul s . . . . . . . . . . . . . . . . . . . . . . . . . . 185
15.4Discussion.............................. 186
16 C i icali y in complex socio echnical sys ems, an empi ical app oach 187
16.1Desc ip ion ............................. 188
16.2Cases udy.............................. 191
16.3Resul s................................ 193
16.4Discussion.............................. 203
17 Vehicle o G id 205
17.1Desc ip ion ............................. 205
17.2Cases udy.............................. 210
17.3Resul s................................ 215
17.4Discussion.............................. 218
18 F equency Managemen wi h Swa m In elligence: a case s udy in Sma
G ids 219
18.1Desc ip ion ............................. 219
18.2Cases udy.............................. 220
18.3Resul s................................ 222
18.4Discussion.............................. 226
19 Agen -based modelling o designing an EV cha ging dis ibu ion sys-
ems: a case s udy in Sal ado o Bahia 227
19.1Desc ip ion ............................. 227
19.2Cases udy.............................. 229
19.3Resul s................................ 230
19.4Discussion.............................. 233
IV
CONTENTS
V Conclusions 235
20 Resul s 237
20.1 Abou Sma G id modelling . . . . . . . . . . . . . . . . . . . . 239
20.2 Abou da a analysis . . . . . . . . . . . . . . . . . . . . . . . . . 240
20.3 Abou s a egy design . . . . . . . . . . . . . . . . . . . . . . . . 241
20.4 Empi ical hypo heses alida ion . . . . . . . . . . . . . . . . . . 241
20.5T ans e ence............................. 243
20.6P ojec s ............................... 244
20.7Publica ions............................. 244
21 Discussion 249
21.1Con ibu ions ............................ 249
21.2O he ema ks............................ 256
21.3Fu u ewo k............................. 259
Bibliog aphy 265
V
Lis o Figu es
1.1 DSM policy design me hod . . . . . . . . . . . . . . . . . . . . . 9
1.2 Model o e alua ing DSM policies . . . . . . . . . . . . . . . . . 10
1.3 Addi ion o laye s o implemen ing he policy . . . . . . . . . . 11
2.1 Powe g ids’ challenges . . . . . . . . . . . . . . . . . . . . . . . 18
3.1 Powe consump ion a iabili y . . . . . . . . . . . . . . . . . . . 29
3.2 Demand modi ica ion wi h DSM . . . . . . . . . . . . . . . . . . 34
4.1 Complex sys em illus a ion . . . . . . . . . . . . . . . . . . . . 39
5.1 MDE abs ac ion le els . . . . . . . . . . . . . . . . . . . . . . . 53
5.2 MDEexample............................ 54
6.1 DSS concep e olu ion . . . . . . . . . . . . . . . . . . . . . . . 58
6.2 Decision making p ocess unde a DSS en i onmen . . . . . . . . 59
6.3 BIda a amewo k ......................... 62
8.1 Abs ac ion le els o modelling powe g ids . . . . . . . . . . . . 76
8.2 Ta a amewo k a chi ec u e . . . . . . . . . . . . . . . . . . . . 78
8.3 A Me amodel example o a powe g id . . . . . . . . . . . . . . 81
8.4 Dependencies examples be ween objec s . . . . . . . . . . . . . . 88
8.5 Modelexample ........................... 89
8.6 Synch onous s Asynch onous simula ion . . . . . . . . . . . . . 90
8.7 Da a eques ............................. 91
8.8 Reques ge s blocked . . . . . . . . . . . . . . . . . . . . . . . . 92
VII
LIST OF FIGURES
8.9 The da a is deli e ed . . . . . . . . . . . . . . . . . . . . . . . . 92
8.10Cyclicdependency ......................... 93
8.11Messagesending .......................... 94
8.12 Message is ecei ed and applied . . . . . . . . . . . . . . . . . . 94
8.13 Simula ion Time p opaga ion . . . . . . . . . . . . . . . . . . . . 95
8.14 Simula ion Time ecep ion . . . . . . . . . . . . . . . . . . . . . 95
8.15Scallingup ............................. 96
8.16 Ta a asynch onous simula ion a chi ec u e . . . . . . . . . . . . 98
8.17 In e ac ion example wi h agen s . . . . . . . . . . . . . . . . . . 99
9.1 S uc u e o expo simula ion esul s . . . . . . . . . . . . . . . . 102
9.2 Cube o esiden ial consump ion . . . . . . . . . . . . . . . . . . 104
9.3 An OLAP Cube s uc u e . . . . . . . . . . . . . . . . . . . . . . 104
9.4 Desc ip ion o a ac . . . . . . . . . . . . . . . . . . . . . . . . . 106
9.5 Scena io composi ion . . . . . . . . . . . . . . . . . . . . . . . . 107
9.6 Householdcube........................... 107
9.7 Household dimension . . . . . . . . . . . . . . . . . . . . . . . . 108
9.8 TVcube............................... 108
9.9 TVdimension............................ 109
9.10Radia o cube............................ 109
9.11 Radia o dimension . . . . . . . . . . . . . . . . . . . . . . . . . 110
9.12 Radia o da a mine . . . . . . . . . . . . . . . . . . . . . . . . . 111
9.13 In o ma ion isualisa ion example . . . . . . . . . . . . . . . . . 112
11.1 Case s udy’s me amodel . . . . . . . . . . . . . . . . . . . . . . 125
11.2 A e aged cu e a e 100 simula ion uns. . . . . . . . . . . . . . 129
11.3 Cu es o 100 simula ion uns. . . . . . . . . . . . . . . . . . . . 130
11.4 Cu e o wo simula ion uns wi h signi ican di e ences. . . . . . 130
11.5 Cu es o 100 simula ions uns wi h ixed numbe o agen s. . . . 131
11.6 Cu es o wo simula ions uns wi h ixed numbe o agen s. . . . 132
11.7Couplingde ails1.......................... 133
11.8Couplingde ails2.......................... 133
11.9 Times in which each en i y kind inished . . . . . . . . . . . . . . 135
VIII

LIST OF FIGURES
12.1 Load cu e example . . . . . . . . . . . . . . . . . . . . . . . . . 139
12.2 Agen a chi ec u e . . . . . . . . . . . . . . . . . . . . . . . . . . 140
12.3 Socio-demog aphic g oups used in he case s udy . . . . . . . . . 141
12.4 Simula ed load cu e . . . . . . . . . . . . . . . . . . . . . . . . 143
12.5 Model cons uc ion pa e n . . . . . . . . . . . . . . . . . . . . . 144
12.6 Compa ison wi h eal da a . . . . . . . . . . . . . . . . . . . . . 145
13.1 Absence om home on weekdays . . . . . . . . . . . . . . . . . 155
13.2 Simula ed load cu e wi h eal dis ibu ions . . . . . . . . . . . . 156
13.3 Simula ed load cu e o conse a i e well-o . . . . . . . . . . . 157
13.4 Simula ed load cu e o en e ainmen seeke s . . . . . . . . . . 158
14.1DLCs uc u e............................ 163
14.2MOPSOpa icles.......................... 165
14.3Sea chspace............................. 169
14.4 Con olle s and consume s . . . . . . . . . . . . . . . . . . . . . 170
14.5 Res ic ions sen by he g id ope a o . . . . . . . . . . . . . . . . 171
14.6 Indi idual appliances consump ion . . . . . . . . . . . . . . . . . 174
14.7 Indi idual appliances consump ion wi h DLC . . . . . . . . . . . 175
14.8 Neighbou hoods consump ion . . . . . . . . . . . . . . . . . . . 176
14.9 Neighbou hoods consump ion wi h DLC . . . . . . . . . . . . . . 176
14.10To al consump ion . . . . . . . . . . . . . . . . . . . . . . . . . . 177
14.11To al consump ion wi h DLC . . . . . . . . . . . . . . . . . . . . 177
14.12Non-DLC sDLC.......................... 177
15.1 GIS laye s ela ed o he p o ided da a . . . . . . . . . . . . . . . 182
15.2 DLC s a egy lowcha . . . . . . . . . . . . . . . . . . . . . . . 183
15.3 Powe o e -demand unbalances . . . . . . . . . . . . . . . . . . 184
15.4 Non-DLC s DLC o a building . . . . . . . . . . . . . . . . . . 185
15.5 Non-DLC s DLC o all cus ome s . . . . . . . . . . . . . . . . 186
16.1S abili y egimes .......................... 192
16.2 Pa ame e s and hei classi ica ion . . . . . . . . . . . . . . . . . 194
16.3 S abili y conside ing he e ige a o sha e . . . . . . . . . . . . . 196
16.4 Numbe o oscilla ion s sha es . . . . . . . . . . . . . . . . . . . 197
IX
LIST OF FIGURES
16.5 S abili y conside ing he doo opening a e . . . . . . . . . . . . . 198
16.6 S abili y conside ing he equency h eshold . . . . . . . . . . . 199
16.7 Edge o chaos conside ing he equency h eshold . . . . . . . . 200
16.8 S abili y conside ing he ex e nal empe a u e (scena io 4-a) . . . 201
16.9 Edge o chaos conside ing he ex e nal empe a u e (scena io 4-a) 202
16.10S abili y conside ing he ex e nal empe a u e (scena io 4-b) . . . 203
16.11Edge o chaos conside ing he ex e nal empe a u e (scena io 4-b) 203
17.1 Cha ging s a ions consump ion o bo h s a egies . . . . . . . . . 217
18.1 F equency eac ion acing a gene a ion uni ailu e . . . . . . . . 221
18.2 Indica o s ha a e used o e alua e policies . . . . . . . . . . . . 222
18.3 F equency eac ion o a ailu e wi h sma e ige a o s . . . . . . 223
18.4 F equency eac ion o he policy #2 . . . . . . . . . . . . . . . . 223
18.5 F equency eac ion o he policy #3 . . . . . . . . . . . . . . . . 224
18.6 Bes policy s base case . . . . . . . . . . . . . . . . . . . . . . . 225
19.1 EVs consump ion when hey s a cha ging as soon as hey a e
plugged ............................... 231
19.2 Compa ison o he o iginal load cu e o 2014 (g ey) wi h he o al
consump ion including EVs cha ging o 2030 (black) . . . . . . . 232
19.3 EVs consump ion acco ding o he RTP based policy which makes
cheape cha ging om 1am o 8am . . . . . . . . . . . . . . . . . 232
19.4 Compa ison be ween o iginal load cu e o 2014 (g ey) wi h he
o al consump ion including EVs cha ging (black) using he RTP-
basedpolicy o 2030........................ 233
20.1 Policies design p oblems . . . . . . . . . . . . . . . . . . . . . . 239
21.1Kuhn’scycle ............................ 250
X
Lis o Tables
11.1 Con igu a ion o he agen s ha uses he shopping cen es. . . . . 127
11.2 Timing o e e y de ice a each simula ion . . . . . . . . . . . . . 134
11.3 Pe o mance compa ison be ween synch onous and asynch onous
cases. ................................ 135
13.1 O e li es yle g oups in S u ga . . . . . . . . . . . . . . . . . . 152
14.1 Elemen s exis ing in he model scene. . . . . . . . . . . . . . . . 172
14.2 MOPSO pa ame isa ion used. . . . . . . . . . . . . . . . . . . . 173
14.3 MOPSO execu ion imes. . . . . . . . . . . . . . . . . . . . . . . 178
17.1 PSO pa ame isa ion used. . . . . . . . . . . . . . . . . . . . . . 216
18.1 Indica o s o bo h cases. . . . . . . . . . . . . . . . . . . . . . . 225
19.1Vehicles............................... 229
XI
1.1 The u u e o powe g ids
su icien elec ici y [Mas13]. This solu ion is e ec i e bu no e icien since hese
backup uni s mus be ope a i e e en when enewable ene gy is being p oduced.
On he one hand, in as uc u es (buildings, powe lines, ans o me s, gene a o s)
should eplica e he enewable ins alled powe . This in as uc u e mus be ins alled
and main ained. On he o he hand, backup gene a o s keep bu ning ossil- uels
while enewable sou ces a e p oducing ene gy. The e o e, enewable ene gy is no
ze o-emission, since i is indi ec ly p oducing GHG.
An al e na i e solu ion in ol es he use o ba e ies o o he ene gy s o age
echnology. This is he case o El Hie o island whe e an ene gy s o age in a-
s uc u e has been buil wi h a bidi ec ional hyd oelec ic powe plan [BC05]. The
main p oblem wi h his solu ion is he in es men cos which is e y high. How-
e e , he cos o ene gy s o age echnology could dec ease in he coming yea s and
his solu ion could become mo e compe i i e [And09].
Howe e , SGs could p o ide al e na i e solu ions ha could be mo e e icien .
Tha means, he same esul s could be achie ed wi h less expense. Fo example,
when he sun is hidden by clouds, PV p oduc ion may be a ec ed, making p oduc-
ion lowe han consump ion. In his case, some loads can be selec i ely discon-
nec ed in o de o ebalance consump ion and p oduc ion. In ano he case, when
clouds disappea and p oduc ion is eco e ed, loads can be econnec ed. These
ac ions a e espec i ely known as load shedding and load shi ing [S 08] and hey
a e cen al opics in his hesis. Since hese policies ac di ec ly on he consump ion
side, hey a e classi ied as Di ec Load Con ol wi hin Demand Side Managemen
(DSM) [GMR+03].
F om a gene al poin o iew, DSM includes all he policies ha adap he de-
mand o g id equi emen s [GMR+03, NNdGW09]. Some o DSM policies a e
based on making he people awa e o e icien ene gy use. O he policies a e e-
la ed o he modi ica ion o p ices, encou aging consump ion in o -peak hou s and
discou aging consump ion in peak hou s.
Howe e , nowadays he e is in o ma ion echnology (IT) ha could au oma -
ically modi y he consump ion o speci ic loads in o de o ebalance he g id
[NNdGW09]. This echnology is an oppo uni y ha could bene i u u e g ids.
The main p e equisi e o his solu ion is de eloping an IT ne wo k o moni o -
ing and ac ing o e he g id componen s. In his way, powe g ids could become
5

1. CONTEXTUALISATION
sma e since hey would be able o pe cei e he en i onmen and ac acco dingly.
This is a dis up i e concep ion ha in oduces a new powe g id pa adigm, SGs,
whe e he managemen o bo h sides, p oduc ion and consump ion, can be ca ied
ou .
Nowadays, many ins i u ions ela ed o powe g ids a e esea ching and design-
ing he ansi ion owa ds SGs. This esea ch and design p ocess is no an easy
ask since he e a e many ac o s ha ha e o be de ined: in as uc u e, ma ke ,
de ice, communica ion, s a egies and conce ns, among o he s. All o hese ac o s
need o be well de ined and s udied since hey will be implemen ed in eal powe
g ids which a e c i ical sys ems ha wo k a ound he clock. Conside ing all hese
conce ns men ioned abo e, he ocus o his hesis is made on he esea ch o new
DSM policies.
IT-based DSM policies need o be designed, e alua ed and alida ed p io o
hei implemen a ion in a eal powe g id in as uc u e. This is necessa y because
hey may in ol e isks o he s abili y and pe o mance o powe g ids. The sys em
could spin ou o con ol, p o oking mal unc ioning and in he wo s case, powe
ailu es. This could happen in he case o a p oduc ion-consump ion unbalance
when DSM policies a e modi ying he consump ion.
Since i is necessa y o s udy he impac o DSM policies on SGs, his esea ch
con ibu es o he ield o compu e simula ions ha add ess his kind o s udies.
The e a e speci ic challenges ha mus be aced when simula ing DSM policies.
1.2 Simula ing u u e scena ios
To illus a e he need o simula ions o DSM policies, some examples o DSM in
SGs a e p esen ed. In all hese examples, i is assumed ha he g id unde s udy
has a communica ion ne wo k which allows he emo e con ol o he appliances
on he demand side.
Le us hink o a policy whe e he c i e ia o make decisions is based on he s a e
o he powe g id. When he e is an unbalance be ween gene a ion and demand,
his policy will eac . When he e is mo e gene a ion han demand, he policy will
y o inc ease consump ion in appliances (e.g. swi ching wa e hea e s on). In he
6
1.2 Simula ing u u e scena ios
opposi e case, when he e is mo e demand han gene a ion, he policy will ac on
de ices o dec ease he demand (e.g. swi ching wa e hea e s o ).
In he case o a powe g id unbalance in which demand is highe han gen-
e a ion, he applica ion o he policy educes he demand o all e ige a o s o
he g id. Since e e y household has a e ige a o , his ac ion has a huge impac
on he powe demand causing he opposi e unbalance: mo e gene a ion han de-
mand. In his si ua ion, he applica ion o he policy swi ches all he e ige a o s
on again, ying o balance he g id. Ob iously, going back o he o iginal posi ion,
i.e. when all e ige a o s a e on, unbalances he g id since demand is once again
highe han gene a ion. In his case, he applica ion o he policy may lead o an
uns able si ua ion whe e e ige a o s a e con inuously being swi ched on and o
and he consequences could include a powe ailu e o damage o he e ige a o s
[VKE+13]. An expe imen o his si ua ion has been done and mo e in o ma ion
can be ound in chap e 16.
Ano he example o policy is educing demand when he e is an unbalance by
con olling he in ensi y o ligh ing. Le ’s assume ligh ing in ensi y can be emo ely
modi ied by he policy applie . Then, whene e he e is a g id unbalance, his policy
educes he in ensi y o all ligh ing by 30%. A i s glance, his seems like a good
idea and much ene gy demand can be educed in he ace o a powe g id unbalance
p oblem. Ne e heless, wha happens i his occu s a nigh ? Would all cus ome s
be in a ou o ha ing ligh s a 70% o hei no mal in ensi y? Ligh s may be
conside ed as an essen ial powe usage, especially a nigh . This policy does no
ake in o accoun he quali y o he se ice ha is o e ed o cus ome s.
These examples poin ou he need o s udy DSM policies p io o hei im-
plemen a ion. E en hough he policy designs a e ob iously w ong, he e may be
many o he issues (e.g. echnological, social, e c.) ha a e no as easy o in e as
hese. These issues a e no easy o in e due o he ac ha powe g ids a e complex
sys ems ha con ain many di e en ac o s whose own sel -in e es ed decisions a -
ec he g id [PAB12]. The e o e, hese issues mus be iden i ied and aken in o
accoun be o e implemen ing policies in eal g ids.
Simula ion is a way o es hese policies. Th ough simula ion, policies can
be es ed on a la ge-scale making i possible o see e ec s a di e en le els o
agg ega ion. Simula ions o es ing DSM policies may be de eloped by using
7
1. CONTEXTUALISATION
disagg ega ed models since hese policies, such as Di ec Load Con ol, a e de-
o ed o ac ing a he lowes le el o consump ion (de ices on he demand side).
Examples o hese simula ions ha e been ci ed in many documen a y i ems as in
[CGLP94, PKZK08, EKM+11, RVRJ11].
When many decisions a e being made in a local and dis ibu ed manne , he
agg ega ed e ec s o hese decisions esul in an eme gen beha iou ha canno
be easily p edic ed. In gene al, his is a common cha ac e is ic o complex sys ems
and he bes way o s udy hem is h ough simula ions. These simula ions may be
use ul o analysing he policies in SGs and e ining hei design [PFR09].
Complexi y in SGs a ises when many componen s a e ac ing and small a i-
a ions in hei beha iou may cause he sys em o e ol e in an unp edic able man-
ne . The challenges ha a e add essed in his documen a e ela ed o deal wi h
his complexi y when de eloping simula ions o es hese policies. Powe g ids a e
huge enginee ing cons uc ions which consis o many di e en elemen s linked o
each o he . In his sense, di e en app oaches ha e been esea ched in his docu-
men o deal wi h hese le els o complexi y.
1.3 P oblem de ini ion
In he p e ious sec ions, he di icul ies o simula ing DSM policies ha e been
desc ibed. Ne e heless, hey a e also challenges ha mus be aced. On he one
hand, he need o disagg ega ing he demand in o de o s udy DSM policies has
been jus i ied. In addi ion, he need o unning he s udies h ough simula ion
expe imen s has been jus i ied as well.
The design o complex sys ems is usually app oached h ough a ial and e o
me hod. In [Sim91], his is exp essed: “ he mo e di icul and no el he p oblem,
he g ea e is likely o be he amoun o ial and e o equi ed o ind a solu ion”.
This is he case in he design o DSM policies which a e bo h di icul and no el.
This ial and e o me hod is no comple ely andom o blind; i is ac ually highly
selec i e, since he knowledge acqui ed in p e ious expe imen s is used o design
new ones.
Following his idea, he way in which DSM policies a e designed can be based
on an i e a i e p ocess (Figu e: 1.1). This p ocess s a s by de ining he objec i es
8
1.3 P oblem de ini ion
Figu e 1.1: A ial and e o me hod o designing DSM policies.
ha he applica ion o he policy mus achie e. Acco ding o hese, he beha iou
o he indi iduals ha make decisions is de ined. This beha iou mus be o ien ed
o achie e he objec i e o he policy.
Thus, he powe g id needs o be modelled and simula ed in o de o e alua e
he policy. A e he simula ions a e execu ed, he da a hey p o ide is analysed in
o de o compa e i o he objec i es p e iously de ined. Depending on he esul s
o his compa ison, he i e a i e p ocess may be comple ed o no , s a ing a new
i e a ion.
The execu ion o he s ages o his ial and e o me hod may be e o -consuming.
The main di icul ies ha hese s ages ha e a e discussed in he ollowing pa a-
g aphs h ough he desc ip ion o a case s udy ca ied ou in his esea ch.
Some expe imen s de eloped wi hin he con ex o he Millene p ojec [EDFb,
CSMJ13] we e o ien ed o es ing DSM policies on a F ench island called La
R´
eunion [AER]. These expe imen s equi ed he modelling and simula ion o he
island’s en i e g id in ol ing he demand side.
Conc e ely, he expe imen was o ien ed o simula ing he e ec s o applying
DSM policies o e de ached homes. To his end, e e y de ached home has o be
modelled including, a leas , he appliances ha can be emo ely con olled. This
in ol ed a o al numbe o elemen s, including hei espec i e beha iou al models,
o app oxima ely 2.5 million.
Facing his complexi y is no a i ial p oblem. The managemen o such com-
plex models is an a duous ask. Modelling a scena io such as he one p esen ed
equi es he in eg a ion o mul iple da a coming om mul iple sou ces (Figu e:
1.2).
9
1. CONTEXTUALISATION
Figu e 1.2: Base model enabled o he e alua ion o Demand Side policies.
Fi s o all, da a o ep esen ing he powe lines o he g id has o be in eg a ed.
La e on, a da abase o buildings h oughou he island mus be included in he
model. A his poin , appliances ha a e equi ed o simula ing he demand need
o be modelled. Fu he mo e, he beha iou o he people li ing in homes mus be
designed and implemen ed in o de o simula e he way people use he appliances
a he esiden ial le el.
This modelling ask desc ibed abo e is jus one pa o he whole p ocess since
i only p o ides he base scena io on which DSM policies can be es ed. The in-
clusion o a policy like his may in ol e he implemen a ion o a communica ion
ne wo k which enables he ansmission o messages among di e en ac o s in he
g id o command he demand side (Figu e: 1.3). In his ne wo k a se o de ices
ha a e esponsible o making decisions, ansmi ing messages, applying com-
mands, moni o ing he powe g id, e c. has o be placed. This is, he addi ion
o he elemen s ha a e necessa y depends on how he policy is designed and i s
equi emen s.
In his kind o simula ion in which he e a e millions o elemen s in ol ed he
quan i y o esul s p o ided by he simula o is huge. E e y single elemen o
10

1.3 P oblem de ini ion
Figu e 1.3: Addi ion o a communica ion and a decision making laye o he base
model.
he simula ion is p o iding i s own a iable pa ame e s a each ime s ep. Ha ing
2.5 million elemen s ha a e ou pu ing h ee di e en pa ame e s each o one
simula ion day wi h a ime s ep o minu es (so 1440 minu es / day), he o al numbe
o esul s eaches 1010. Wi h so many esul s, modelle s usually ocus only on
ce ain ea u es o he simula ion since i is almos impossible o check all he
ou pu . The e o e, many impo an conclusions ha may be d awn om he esul s
may be missed since he whole da a ou pu is no mally no being e iewed.
Ano he issue ha has been obse ed is ela ed o he concep ion o DSM
policies. The oo p oblem, and he eason why hese policies a e c ea ed, is ha
he e a e limi ed esou ces (ene gy) and many di e en ac o s ha wan o ha e
access o hose esou ces (cus ome s). Fu he mo e, he e is an en i onmen which
has limi ed esou ces and he e ogeneous ac o s ha ha e hei own pa icula in-
e es s. The e o e, he challenge is how o e icien ly dis ibu e hese esou ces
acco ding o ce ain objec i es and cons ain s.
In conclusion, h ee majo p oblems ha e been desc ibed in his sec ion:
• SGs modelling and simula ion: he complexi y o dealing wi h a huge num-
be o elemen s.
• Simula ion esul s analysis: he complexi y o dealing wi h a huge numbe
o esul s om simula ions.
• DSM policies design: he complexi y o dealing wi h so many ac o s ha
wan access o he limi ed ene gy esou ces.
11
1. CONTEXTUALISATION
1.4 Resea ch ques ions
Complexi y is a common issue in he p oblems desc ibed in he p e ious sec ion.
The ques ions p esen ed in his sec ion add ess he p oblems ha ha e been p esen-
ed.
In igu e 1.1, he way in which DSM policies a e designed is p esen ed. Fi s ,
a e de ining he policy’s objec i es, he way in which decision make s will ac
mus be de ined. Second, o es he policy, i is necessa y o de elop “in silico”
expe imen s. Ne e heless, ca ying ou hese expe imen s is an a duous ask i he
powe g id ep esen ed is huge. Thi d, hese simula ions can ou pu huge quan i ies
o da a which mus be analysed.
Al hough he selec i e ial and e o me hod is assumed o be he base o
designing DSM policies, he e is a gene al ques ion ha a ises: how o imp o e
he execu ion o his me hod? Tha is o say: how he e o o design a policy can
be lessened. In o he wo ds, his is a ques ion o e iciency and p oduc i i y. This
can be add essed by making less i e a ions o educing he e o o ca y ou each
i e a ion.
This esea ch is ocused on he educ ion o he e o o ca y ou he i e a ions
as his issue can be deal wi h me hodological and echnological app oaches. The
p oblem o educing he numbe o i e a ions has o do wi h de ining heu is ics ha
sugges he pa hs ha should be ied i s [Sim91]. This has no been add essed in
his esea ch, al hough i could be u he explo ed in u u e esea ch. Thus, his
esea ch is o ien ed o explo ing me hods and echnology o imp o ing e iciency
and p oduc i i y in he design o DSM policies.
To be mo e speci ic, his gene al ques ion is elabo a ed h ough ques ions ha
add ess he p oblem o e iciency and p oduc i i y a each s age: de ini ion o he
indi idual’s beha iou ; modelling and simula ion o complex sys ems; and analysis
o simula ion esul s.
Rega ding he de ini ion o he indi idual’s beha iou : he app oach esea ched
is based on na u al compu ing, ega dless o o he ideas ha could be explo ed.
“Na u al compu ing” me hods ake inspi a ion om na u e o he de elopmen
o p oblem-sol ing echniques [RBK11]. These me hods ha e been p e iously
12
1.4 Resea ch ques ions
applied in o he domains by ou esea ch g oup 1. Fo example, Swa m In elli-
gence me hods ha e been used in op imisa ion p oblems. Fo his eason, he ques-
ion “can Swa m In elligence help o deal wi h he complexi y o designing DSM
policies?” has been used o s udy hese speci ic me hods in he ield o DSM.
In addi ion, his ques ion is also ela ed o he eme gen beha iou o he powe
g id as he idea consis s in using Swa m In elligence o ob ain he powe g id ob-
jec i es o consump ion. In o he wo ds, Swa m In elligence echniques could be
used o ob ain a desi ed eme gen beha iou by ac ing on he indi iduals. The e-
o e, ano he ques ion could be: “can Swa m In elligence help ob ain a desi ed
eme gen beha iou in a SGs?” The e may be many di e en echniques o ad ess
his issue. Howe e , apa om he ac ha we ha e used hese echniques in he
pas , we ha e obse ed ha o he esea che s ha e used hem in he powe g ids
ield (Sec ion: 4.4).
Conce ning he modelling and simula ion o complex sys ems, se e al ques-
ions can be de ined. The mos gene al one would be, how can he complexi y be
add essed? Howe e , his ques ion is oo gene al and i mus be mo e speci ic. A
common s a egy o sol ing a p oblem is using me hods. So, a di e en ques ion
could be: “which me hod would be app op ia e o de ining complex models?”
This ques ion assumes as an axiom ha i is be e o apply a me hod a he han
no hing. Ne e heless, his ques ion ocuses on sea ching o a me hod.
Since his esea ch has been de eloped by a g oup 1wi h esea ch lines in Model
D i en Enginee ing and Business In elligence me hodologies, i is logical o ex-
plo e how hese me hodologies could be applied o he p oblem o complexi y in
SGs. The e o e, he abo e ques ion can be p ecisely exp essed: “could Model
D i en Enginee ing help o deal wi h he complexi y o modelling and simula ing
SGs?” and “could Business In elligence me hodology help o deal wi h he com-
plexi y o he da a analysis om simula ions?”
The na u e o hese ques ions de ines an explo a o y esea ch, whe e hese
me hodologies can be adap ed o his p oblem. I is no possible o explo e how
hese me hodologies a e applied o all he p oblems ela ed o complexi y in SGs.
1CES (Calidad, E iciencia y Sos enibilidad - Quali y, E iciency and Sus ainabili y) di ision o
SIANI (Sis emas In eligen es y Aplicaciones Num´
e icas en la Ingenie ´
ıa - In elligen Sys ems and
Nume ical Applica ions in he Enginee ing) ins i u e, Uni e si y o Las Palmas de G an Cana ia.
13
1. CONTEXTUALISATION
Howe e , i is possible o explo e hem in ce ain cases. The goal ha has been
de ined o his esea ch is o p o ide e idence o o agains hese me hodologies
ins ead o e i ying hem. Fu he mo e, he p oblem canno be exp essed as he
disco e y o he bes me hodology o dealing wi h complexi y. Such ques ions a e
no wo hwhile because i is no possible o de ine sui able expe imen a ion.
1.5 S uc u e o he documen
The es o his documen is o ganised as ollows: a e his pa ha was in ended
o con ex ualise he esea ch, he “s a e o he a ” o opics ela ed o his hesis
a e e iewed: SGs, DSM, modelling and simula ion, model d i en enginee ing
and da a analysis. This pa aims o expand upon he in o ma ion p o ided in he
in oduc ion. I also shows how o he esea che s a e ca ying ou he DSM expe i-
men a ion. Fu he mo e, i desc ibes modelling app oaches, applica ions o model
d i en enginee ing and echnology o da a analysis.
The hi d pa p esen s he hypo heses. These hypo heses a e ela ed o explo -
ing he ques ions p e iously de ined. The e o e, his pa is de o ed o desc ibing
he hypo heses and hei de elopmen in e ms o how he esea ch p ocess has
been add essed and he desc ip ion o he ools ha ha e been de eloped as a con-
sequence o he esea ch.
The ou h pa is de o ed o alida ing he esea ch p ocess desc ibed in he
hi d pa by applying i o case s udies. In his pa o he documen , wo syn he ic
case s udies a e p esen ed o show ea u es ela ed o he modelling and simula ion
o SGs. The ollowing case s udies desc ibed he e a e based on eal da a. They a e
so ed om case s udies in which only he demand is simula ed o hose in which
he e a e policies ha ac o e he demand.
The las pa o he documen , he conclusion, shows he esul s and discus-
sions o his esea ch. This is o say, in he “ esul s” chap e , he ou comes o his
esea ch a e desc ibed in e ms o wha has been done, amewo ks, p ojec s, pa-
pe s, e c. The “discussion” chap e p o ides some in e es ing e lec ions and u u e
conside a ions ela ed o his esea ch.
14
2.3 Scope
Technology Pla o m and he o he om he USA Depa men o Ene gy. They a e
ci ed below:
In [ p eno ], he Eu opean Technology Pla o m o Sma G ids s a ed ha
SGs a e “elec ici y ne wo ks ha can in elligen ly in eg a e he beha iou and ac-
ions o all use s connec ed o i – gene a o s, consume s and hose ha do bo h – in
o de o e icien ly deli e sus ainable, economic and secu e elec ici y supplies”.
In [Dep], he Depa men o Ene gy o US poin s ou ha SGs mus include
he ollowing ea u es: “sel -healing om powe dis u bance e en s; enabling ac -
i e pa icipa ion by consume s in demand esponse; ope a ing esilien ly agains
physical and cybe a ack; p o iding powe quali y o 21s cen u y needs; accom-
moda ing all gene a ion and s o age op ions; enabling new p oduc s, se ices, and
ma ke s; op imizing asse s and ope a ing e icien ly”.
As i can be seen, he Depa men o Ene gy is mo e p ecise when de ining
he aims assigned o a SG, highligh ing he impo ance o add essing sa e y issues
[F e08].
2.3 Scope
SGs we e ini ially concei ed o add ess he imp o emen o he DSM, ene gy e i-
ciency and sa e y h ough he cons uc ion o g ids ha a e obus agains sabo age
and na u al disas e s [RPT07]. Howe e , new equi emen s ha e expanded he
ini ial scope o he SGs o some hing b oade which includes he c ea ion o ame-
wo ks o achie e he in e ope abili y o all he ac o s wi hin he SG [FMXY12].
An in e es ing way o analyse wha SGs a e supposed o add ess is h ough he
p oposal o amewo ks. These amewo ks a e usually de eloped by ins i u ions
and one o he main poin s is he de ini ion o he scope o SGs. The nex pa a-
g aphs will summa ise he scope ha some ele an amewo ks ha e de ined o
SGs.
a) NIST F amewo k p oposal. Acco ding o a epo om NIST (Na ional In-
s i u e o S anda ds and Technology o US) [JG12] de eloped in 2010, SGs scope
mus be ocused on:
• Imp o ing powe eliabili y and quali y
21

2. SMART GRIDS
• Op imizing acili y u iliza ion and a e ing cons uc ion o back-up plan s
• Enhancing capaci y and e iciency o exis ing elec ic powe ne wo ks
• Imp o ing esilience o dis up ion
• Enabling p edic i e main enance and sel -healing esponses o dis u bances
• Facili a ing expanded deploymen o RES
• Accommoda ing dis ibu ed powe sou ces
• Au oma ing main enance and ope a ion
• Reducing GHG emissions by enabling EVs and new powe sou ces
• Oil usage by educing he need o ine icien gene a ion in demand peaks
• P esen ing oppo uni ies o imp o e g id secu i y
• Enabling ansi ion o plug-in EVs and new ene gy s o age op ions
• Inc easing consume choice
• Enabling new p oduc s, se ices, and ma ke s.
b) ETP amewo k p oposal. The Eu opean Technology Pla o m (ETP) con-
side s ha SGs ha e been concei ed o mee he challenges and oppo uni ies o
he 21s cen u y [C+06]. The use o e olu iona y new echnologies, p oduc s and
se ices is conside ed as he main pilla o SGs. In pa icula , he SGs scope mus
be o ien ed o educe peaks and was e, encou age manu ac u e s o de elop mo e
ene gy-e icien appliances and sense and p e en blackou s by isola ing dis u b-
ances on he g id.
c) EEGI amewo k p oposal. The Eu opean Elec ici y G id Ini ia i e (EEGI)
conside s ha SGs scope mus include inc easing hos ing capaci y o enewable
and dis ibu ed gene a ion, in eg a ion o na ional ne wo ks in o ma ke -based ne -
wo ks, ac i e pa icipa ion o use s in ma ke s and ene gy e iciency, and opening
business oppo uni ies and ma ke s [EE10] including he s anda disa ion and in e -
ope abili y.
d) IEA DSM Task XVII amewo k p oposal. The In e na ional Ene gy Agency
DSM ask XVII conside s ha SGs mus be o ien ed o ace in eg a ion o DSM,
Dis ibu ed Gene a ion, RES, ene gy s o ages [In 09]. This in eg a ion is con-
side ed as impo an since Dis ibu ed Gene a ion, dis ibu ed ene gy s o ages and
Demand Response can be seen as dis ibu ed ene gy esou ces [HHM11] ha may
help o in eg a e in e mi en RES.
22
2.4 Challenges
In eg a ing all iews and de ini ions om he li e a u e ha has been e iewed,
SGs can be in ex enso de ined by lis ing hei main obse ed ea u es:
• SGs a e de o ed o imp o e ene gy e iciency, eliabili y and secu i y.
• SGs will inco po a e capaci ies o moni o ing and con olling de ices p o id-
ing lexibili y.
• SGs will edesign he way in which ene gy ma ke s a e wo king nowadays,
including he pa icipa ion o many new ac o s in he decision making p o-
cess o e he g id.
2.4 Challenges
In he li e a u e e iew, di e en s a egies ha SGs may be add essing in he u u e
ha e been shown. This las sec ion o he e iew is in ended o summa ise SG
s a egies ollowing he li e a u e e iew made in [XABO+14].
a) Sma me e s and demand lexibili y. Sma me e s appea o be a genuinely
c oss cu ing componen o SGs, al hough hey a e no uni e sally pe cei ed as
necessa y o a SG [Eu 10]. This echnology is expec ed o help educe demand
based on be e in o ma ion and by shi ing load consump ion o o -peak imes
[LC11].
Nowadays, he e is a massi e in oduc ion o he mal loads in he powe g id
such as hea ing sys ems, ai condi ioning sys ems (HVAC) and e ige a o s. In a
nea u u e he massi e in oduc ion o Elec ical Vehicles (EVs) is p edic ed. Bo h
he mal loads and EVs may become a majo d i e o SGs [XABO+14]. This is
due o he ac ha bo h may help sa is y demand peaks by ei he injec ing s o ed
ene gy o s opping consump ion [CMG11]. Howe e , hei in oduc ion in he g id
in ol es a conside able inc emen in he o e all ene gy consump ion. In his si u-
a ion, he challenge is di ided in o wo sub-challenges: imp o ing he capaci y o
powe g ids o suppo hei in oduc ion and adding mechanisms o con ol hese
kinds o de ices o suppo demand peaks. The second one is ela ed o a concep
ha is u he de eloped in chap e 3 called DSM.
b) Secu i y o supply. As men ioned be o e, his is an impo an opic since
he in oduc ion o RES may in ol e secu i y o supply weaknesses due o hei
23
2. SMART GRIDS
in e mi en na u e. As said in [Spe10], we a e s a ing o become awa e o he
h ea o supply dis up ion which may ame he de elopmen o SG a ound ene gy
secu i y.
c) Cybe secu i y, p i acy, and con ol. The in oduc ion o an in as uc u e o
con ol and moni o makes i possible o cap u e da a ha may be sensi i e. Fo his
eason, da a secu i y becomes an impo an conce n bo h om a da a go e nance
and a cybe -secu i y poin o iew [Ho 11]. The i s one conce ns he p i ileges o
access his da a, whe eas he second one conce ns how ha da a mus be handled
o p e en i om being accessed by in ude s [TM11]. The in oduc ion o Sma
Me e s makes his issue e en mo e u gen and mus be app oached by SG s a egies.
d) Sys em de agmen a ion. The non-coo dina ed ope a ion o companies ha
ope a e in he ene gy sec o in ol es se e al di e en s anda d echnologies and
p o ocols. This leads o di e en business models ha equi e o be me ged in o de
o o e come such di e ences [Jac11]. A he same ime, his makes a ansi ion o
a decen alised sys em mo e di icul , which is co e o SGs [XABO+14]. This
agmen a ion p oblem mus be add essed by SGs.
e) Mic ogene a ion and decen aliza ion. Mic o-gene a ion is gene ally e y
impo an o low ca bon elec ici y sys ems, e.g. o alle ia e sys em conges ion
[BMW10]. Mic ogene a ion o e s bene i s o educing he demand, especially in
a wide decen alised con ex . Indi iduals would be esponsible o hei ene gy
p oduc ion and consump ion and, he e o e, would be awa e o how hey ha e o
use ene gy in o de o op imise bo h a iables [DW07].
Howe e , apa om he high cos s hese echnology ha e, he e a e wo main
ac o s which a e p e en ing he de elopmen o hese echnologies [XABO+14].
On he one hand, Dis ibu ion Ne wo k Ope a o s discou age hese in es men s on
inno a ion. On he o he hand, many ene gy ma ke s a e based on cen al gene -
a ion. The inclusion o hese echnologies in he g id would open his ma ke o
small ene gy selle s which, on he agg ega e, may become impo an playe s o he
ma ke .
An in e es ing idea [PRS08] o add ess his kind o egula o y issues is he use
o Vi ual Powe Plan s. Vi ual Powe Plan s can eme ge gi en he comme cial
and egula o y suppo . These plan s will depend on he coope a ion o hose who
a e in con ol [Wol12].
24
2.4 Challenges
) In e ope abili y. P e iously p esen ed amewo ks a e no only in ended o
de ine SGs scope, bu many mo e hings. One o he mos impo an is he de ini-
ion o in e ope abili y. Fo ins ance, NIST has de ined a SG In e ope abili y Panel
(SGIP). This panel is o ien ed o p o ide a o um ha suppo s s akeholde pa icip-
a ion and ep esen a ion wi h he aim o de eloping and e ol ing in e ope abili y
s anda ds [JG12]. SGIP has h ee p ima y unc ions:
• To o e see ac i i ies in ended o expedi e he de elopmen o in e ope abili y
and cybe -secu i y speci ica ions.
• To p o ide echnical guidance o acili a e he de elopmen o s anda ds o
a secu e and in e ope able SG, and
• To speci y es ing and ce i ica ion equi emen s necessa y o assess he in-
e ope abili y o SG- ela ed equipmen .
25

CHAPTER
3
Demand side managemen
In he li e a u e, he e a e many sou ces ha e iew DSM. F om all o hem,
[S 08] is he one ha has been ollowed, due o i s cla i y, o conduc he s uc-
u e o his chap e . A e e iewing DSM, di e en app oaches o simula e DSM
policies a e p esen ed as his is a co e concep in his hesis.
Wi hin he SG concep , moni o ing and con olling he demand is one o he
key aspec s o imp o e powe sys em e iciency. Then, he app oach o new powe
g ids may consis in, besides ac ing on he p oduc ion, ac ing on he demand oo.
Powe g ids may ac upon bo h he p oduc ion and demand side, he e o e hey
bo h become lexible g id elemen s. The objec i es o DSM a e, among o he s, he
minimisa ion o peak demand and he imp o emen a he sys em ope a ion and
planning le el [GMR+03].
3.1 Oppo uni ies o Demand Side Managemen
The e a e di e en sec o s whe e oppo uni ies o he DSM can be ound: gene a-
ion, ansmission, dis ibu ion and demand. These sec o s a e e iewed below.
27
3. DEMAND SIDE MANAGEMENT
3.1.1 Gene a ion
Powe g ids a e designed o suppo he maximum peak o demand. This peak o
demand a ies o e ime on a daily and seasonal basis. Apa om his, powe g ids
a e buil wi h a 20% ma gin o gene a ion capaci y o deal wi h he unce ain y
o gene a ion capaci y and unp edic ed demand inc eases [S 08]. The a e age
u ilisa ion o he gene a ion capaci y in a yea is below 55%.
The e is ano he issue a ising wi h he in oduc ion o in e mi en enewable
sou ces as wind o pho o ol aic. These sou ces o ene gy a e no p edic able and
his causes he in oduc ion o unce ain y in he powe g id gene a ion [BI04].
Nowadays, his is deal by using backup gene a ion uni s ha a e a ailable o sub-
s i u e in e mi en enewable gene a ion in case i is needed. This equi es a high
in es men [Mas13].
This is an oppo uni y o DSM o apply load shi ing (mo e some usages o
ene gy) om peak o o -peak pe iods [GMR+03]. Load shi ing can help, on he
one hand, o educe he equi ed gene a ion capaci y since peaks a e smalle and,
on he o he hand, imp o e he e iciency by inc easing he a e age u ilisa ion o
gene a ion capaci y. Conce ning he in oduc ion o in e mi en enewable gene -
a ion, DSM can help o abso b ene gy in cases in which he e is a high p oduc ion
o in e mi en enewable sou ces and a low consump ion.
3.1.2 T ansmission and dis ibu ion
F om he poin o iew o he ansmission and dis ibu ion, powe g ids a e de-
signed o be obus in case a ci cui is los . When his happens, emaining ci cui s
ake o e he load o he aul y one. Howe e , hese ci cui s canno become o e -
loaded. Tha is, unde peak-load condi ions, hese ci cui s a e usually loaded below
50% [S 08].
An oppo uni y eme ges o he DSM in his ield when powe g ids a e de-
signed o ha e a disagg ega ed gene a ion. Nowadays, mos o he gene a ion ca-
paci y is cen alised. Howe e , he dis ibu ion o gene a ion a oids he need o
ha ing low-used ci cui s jus in case a ci cui ails [S 08]. DSM may help in his
ask o make an ac i e con ol o he dis ibu ed p oduc ion acco ding o he eal-
ime needs o he g id.
28
3.1 Oppo uni ies o Demand Side Managemen
Figu e 3.1: Le : he consump ion o a win e weekday. Righ : he consump ion o a
summe weekday. (Sou ce: Red El´
ec ica de Espa˜
na).
3.1.3 Demand
Demand side is la gely uncon ollable in cu en powe g ids and a ies wi h ime
o he day and season [S 08]. Fo ins ance, in G ea B i ain, he minimum con-
sump ion occu s in summe nigh s and is abou a 30% o he win e peak. The
a iabili y o he consump ion o a powe g id depends on egional en i onmen al
condi ions.
In Spain, as in G ea B i ain and many o he powe g ids, he e is a high a i-
abili y in he amoun o ene gy consumed du ing he day and be ween seasons.
In igu e 3.1, on he le side, he consump ion o a win e weekday in he Spanish
mainland is p esen ed and, on he igh side, a cu e o a summe weekday (Sou ce:
Red El´
ec ica de Espa˜
na [REE]). On he one hand, a iabili y conce ning he day-
ime can be obse ed since he e a e isible alleys and peaks. On he o he hand,
a iabili y be ween seasons can be obse ed since no only he amoun o ene gy
changes bu also he shape o he cu e.
DSM can ha e an oppo uni y o balance his unbalanced condi ion ha exis s
be ween peaks and o -peaks [PD11]. This can be made by shi ing loads om
peak pe iods o consume when he g id is in an o -peak pe iod. A way o add ess
a load shi ing is h ough he use o elec ic ene gy s o ages (ba e ies) so ha hey
can injec ene gy in peak-pe iods and consume ene gy in o -peak pe iods.
Since he massi e in oduc ion o ba e ies is nowadays e y expensi e, o he
ene gy s o ages ha a e ound in he demand side can be used, such as he mal
loads [PD11]. These loads can be shi ed om peak-pe iods o o -peak pe iods.
29
3. DEMAND SIDE MANAGEMENT
The ene gy consump ion o hese loads is no mally no educed bu pos poned.
3.2 Re iew o Demand Side Managemen echniques
DSM is di ided in se e al kinds o in e en ions a he cus ome le el including
Di ec Load Con ol (DLC) and Demand Response (DR) [NNdGW09]. DLC is an
in e en ion a he de ice le el in o de o shi cus ome s’ consump ion ega ding
g id s a e objec i es. DR consis s in modi ying he cus ome s’ ene gy usage om
hei no mal consump ion pa e ns in esponse o changes (e.g. p ice o elec ici y
o e ime) [AES07]. The nex subsec ions in oduce di e en DSM echniques.
3.2.1 Di ec load con ol
The nex sec ions p esen se e al echniques in which a di ec -load con ol ap-
p oach is being used. In hese echniques, appliances a e di ec ly add essed by
emo e de ices in o de o modi y hei consump ion acco ding o ce ain c i e ia.
a) Nigh - ime hea ing wi h load swi ching. The idea o his echnique is o se
hea e s o consume ene gy a nigh in o de o balance he consump ion o he g id.
This echnique equi es he use o addi ional de ices which allows o swi ch hea ing
de ices emo ely by using adio ele-swi ching. This echnique can i in o wha is
known as Di ec -Load Con ol.
b) Comme cial and indus ial p og ammes. The e a e some p og ammes ha
a e aimed a comme cial and indus ial pu poses. Pa icula ly popula a e load-
in e up ible p og ammes which a e o ien ed o p o ide ese e se ices enhancing
he sys em eliabili y. O he p og ammes ha a e no mally a ailable o comme -
cial cus ome consis in con olling loads by using building con ol sys ems o
HVAC. These de ices can be connec ed o powe g id agg ega o s o command o
ei he educe o inc ease consump ion acco ding o he s a e o he g id.
3.2.2 Demand esponse
The nex sec ions p esen se e al echniques in which a demand esponse app oach
is being used. These echniques a e cha ac e ized o using ex e nal s imulus, such
30
4.1 Powe g id simula o s
TEFTS [Uni00]: “p og am has been designed o do ansien s abili y and en-
e gy unc ion analyses o educed dynamic models o ac/dc powe sys ems, wi h
addi ional capabili ies o ol age s abili y (bi u ca ion) s udies based on con inu-
a ion me hods”.
MATPOWER [ZG97]: “is a package o MATLAB M- iles o sol ing powe
low and op imal powe low p oblems. I is in ended as a simula ion ool o
esea che s and educa o s ha is easy o use and modi y”.
Vol age S abili y Toolbox (VST) [Nwa02]: “de eloped a he Cen e o Elec ic
Powe Enginee ing, D exel Uni e si y combines p o en compu a ional and analy -
ical capabili ies o bi u ca ion heo y and symbolic implemen a ion and g aphical
ep esen a ion capabili ies o MATLAB and i s Toolboxes. I can be used o ana-
lyze ol age s abili y p oblem and p o ide in ui i e in o ma ion o powe sys em
planning, ope a ion, and con ol”.
Powe Sys em Analysis Toolbox (PSAT) [Mil05]: “is a Ma lab oolbox o elec-
ic powe sys em analysis and simula ion. The main ea u es o PSAT a e: Powe
Flow; Con inua ion Powe Flow; Op imal Powe Flow; Small Signal S abili y Ana-
lysis; Time Domain Simula ion; Comple e G aphical Use In e ace; Use De ined
Models; FACTS Models; Wind Tu bine Models; Con e sion o Da a”.
In e PSS (In e ne echnology based Powe Sys em Simula o ) [Zho]: “is a ee
and open so wa e de elopmen p ojec . Simula ion is key o enhancing powe
sys em design, analysis, diagnosis, and ope a ion”.
AMES Ma ke Package [Tesa]: “is ou so wa e implemen a ion, in Ja a, o
he AMES Wholesale Powe Ma ke Tes Bed. Ou objec i e is he acili a ion o
esea ch, eaching, and aining, no comme cial-g ade applica ion”.
DCOPFJ [Tesb]: “is a ee open-sou ce Ja a sol e o bid/o e -based DC op-
imal powe low (DC-OPF) p oblems sui able o esea ch, eaching, and aining
applica ions”.
OpenDSS [ADHM]: “is a simula o speci ically designed o ep esen elec-
ic powe dis ibu ion ci cui s. OpenDSS is designed o suppo mos ypes o
powe dis ibu ion planning analysis associa ed wi h he in e connec ion o dis ib-
u ed gene a ion (DG) o u ili y sys ems”.
TSAT [Pow]: “is a leading-edge ull ime-domain simula ion ool designed o
comp ehensi e assessmen o dynamic beha io o complex powe sys ems”.
37

4. MODELLING AND SIMULATION FOR SMART GRIDS
G idLAB-D [Pac]: “is a new powe dis ibu ion sys em simula ion and analysis
ool ha p o ides aluable in o ma ion o use s who design and ope a e dis ibu ion
sys ems, and o u ili ies ha wish o ake ad an age o he la es ene gy echnolo-
gies”.
Some o hese simula o s a e able o simula e he demand side a he le el o
disagg ega ion ha is necessa y. Howe e , hey p esen some limi a ions ha a e
discussed in he hi d pa o his documen .
4.2 Powe g ids as complex sys ems
Powe g ids a e composed o many elemen s a di e en le els connec ed o each
o he , a a physical le el, h ough a ne wo k in as uc u e [K e13]. The pa adigm
shi in he ene gy sec o which is cha ac e ised by new ma ke ules, oge he wi h
he in oduc ion o enewable ene gies and dis ibu ed sys ems ha e inc eased i s
deg ee o complexi y. The implemen a ion o SG echnologies may in ol e he
in oduc ion o a ne wo k laye which enhances he capabili y o communica ion
among he di e en ac o s o he g id.
SG concep s, such as DSM, dis ibu ed gene a ion and ene gy e iciency usage,
encou age new s udies ha equi e new app oaches. P e ious s udies on powe
g ids we e mainly ocused on how o be mo e e icien om he poin o iew o
scheduling powe gene a ion uni s acco ding o a demand which was conside ed
as an agg ega ed load. New s udies ha a e ocused on demand side, dis ibu ion
and local e ec s o applying demand side s a egies equi e he use o app oaches
whe e he powe g id mus be ep esen ed as a complex sys em [PKZK08].
A complex sys em is comp ised o a (usually la ge) numbe o (usually s ongly)
in e ac ing en i ies, p ocesses o agen s, he unde s anding o which equi es he de-
elopmen , o he use, o new scien i ic ools, nonlinea models, ou -o equilib ium
desc ip ions and compu e simula ions [Wo 10] (Figu e: 4.1).
An in e es ing conce n is ha he complex sys em e e s o he way in which
eali y is ep esen ed. Tha is, e e y eal sys em is complex on i s own. Fo ex-
ample, a sys em composed by a ca going h ough a pa h is a complex sys em.
Howe e , i s beha iou can be modelled as a simple equa ion which conside s
38
4.2 Powe g ids as complex sys ems
Figu e 4.1: Complex sys em illus a ion. Ne wo k o di e en ypes o en i ies and
links (colou s) as example o a complex sys em.
some ew ac o s. In his case, he eal complex sys em is simpli ied in o a equa-
ion. Ne e heless, his sys em can be ep esen ed as a complex sys em in which all
main componen s a e sepa a ely modelled and in e connec ed wi h each o he . Fo
example, he ca can be modelled by i s componen s (wheels, engine, s uc u e, e c)
and each o hem will ha e an in luence in he way he ca beha es d i ing along
he pa h. Models and simula o s a e no mally de eloped o a end he equi emen s
o he expe imen hypo heses.
Simila ly, his may be applied o powe g ids. Ini ially, hey we e simply mod-
elled since he manageable sec ion o he powe g id was he p oduc ion side. To
his end, many models and simula ions o hese powe g ids we e ep esen ed by
a sys em o equa ions. Nowadays, he in oduc ion o new echnologies which
enhance he capabili y o ha ing a disagg ega ed con ol on he g id whe e many
di e en ac o s can in luence i boos he s udy o powe g ids o a highe le el o
complexi y [KdH09]. This way, we can unde s and be e he way in which SG
concep s can impac on powe g ids conce ning a wide a ie y o anges: p oduc-
ion, demand, ma ke s, dis ibu ed gene a ion, on-line mic og ids, e c.
The nex pa ag aphs a e in ended o p o ide some insigh s on he mos im-
po an p ope ies o complex sys ems in o de o ully unde s and he in e es o
inco po a ing his app oach o he simula ion o powe g ids. These p ope ies a e
he e ogenei y, ne wo ks and eme gence [BY97].
One o he main p ope ies o complex sys ems is he he e ogenei y o he ele-
men s ha con o m he sys em. This p ope y is one o he ac o s ha de e mine
39
4. MODELLING AND SIMULATION FOR SMART GRIDS
how complex a sys em is.
In his sense, e e ing back o he example o he ca d i ing along a pa h,
we can app ecia e he pa icipa ion o di e en and he e ogeneous en i ies: ou
wheels, an engine, a ca s uc u e (which can be decomposed) and a pa h. Le ’s
assume ha he expe imen is aimed a measu ing how long i akes he ca o go
om one poin o he pa h o ano he . In his sense, hese elemen s which ha e
conc e e beha iou s will in luence on he o al amoun o ime ha i will ake he
ca o do his ajec o y.
Powe g ids a e also composed o he e ogeneous en i ies such as gene a o s,
consume s, dis ibu ion echnologies, e c. Mo eo e , each ca ego y o hem con-
ains mo e he e ogeneous elemen s inside. Each elemen wi hin he powe g id has
an in luence on he sys em beha iou (which is an eme gen beha iou . Eme gence
in complex sys ems is u he de eloped in his sec ion).
Ne wo ks a e ano he impo an p ope y wi hin complex sys ems. This p op-
e y also conce ns he he e ogenei y p ope y as connec ions wi hin complex sys-
ems may be he e ogeneous oo. This he e ogenei y a he ne wo k le el is mainly
due o wo ac o s: ype o connec i i y and kind o connec ions. Connec i i y
ypes in ne wo ks may be any o he lis below [K e13]:
• Fully connec ed ne wo ks: e e y elemen o he sys em is connec ed o all
o he nodes o he sys em.
• Dis ance based ne wo ks: elemen s a e connec ed among hem acco ding o
dis ance c i e ia. Fo ins ance, e e y elemen is connec ed o i s wo closes
elemen s.
• Random ne wo ks: in his kind o ne wo ks, elemen s a e connec ed o each
o he ollowing a andom c i e ia which is based on a ce ain p obabili y.
• Scale ee ne wo ks [B+09]: all sys em ne wo ks ha canno be modelled
ollowing any o he app oaches abo e. Fo example, social ne wo ks do no
wo k ollowing he kind o ne wo ks exposed be o e. In his sense, scale
ee ne wo ks ha e been iden i ied in o de o ca ego ise all hose sys em
ne wo ks which do no i in he o he ca ego ies. In social ne wo ks, iend-
ship among people does no ollow he pa e ns desc ibed abo e. As hey
ollow di e en shapes, hey a e conside ed o be scale ee ne wo ks. Unde
40
4.2 Powe g ids as complex sys ems
hese ne wo ks, elemen s a e connec ed eely o each o he acco ding o he
sys em ha is being modelled.
In he same way ha he e is he e ogenei y among componen s wi hin a com-
plex sys em, connec ions can be he e ogeneous om he poin o iew o he ela-
ion hey exp ess.
Re e ing back o he example o he social ne wo k, links among people wi hin
he social ne wo k may be o di e en na u e: iendship, ela ionship, amily, e c.
In his sense, links a e he e ogeneous as hey exp ess di e en kinds o ela ion
among en i ies and, he e o e, hey de ine he way in which componen s will in e -
ac wi h each o he . Especially in he case o SGs, we can ind many di e en kinds
o links among componen s as, o ins ance, powe lines, communica ion links o
con ainmen ela ions.
The coope a ion among en i ies h ough hese links leads us o he las p ope y:
eme gence. Eme gence is an impo an key issue wi hin complex sys ems. Com-
plex sys ems o en beha e in unexpec ed ways ha canno be in e ed di ec ly om
he beha iou o hei componen s; his is known as eme gen beha iou [NN12].
Eme gen phenomena a e a consequence o complexi y. The ope a ing mode o
a sys em ha is being analysed ollowing a adi ional app oach can be p edic ed
om he sys em model as his ope a ing mode has had o be p og ammed [SPT06].
In he case o complex sys ems, eme gen beha iou s a e no p edic able h ough
ools de o ed only o he da a analysis, which is he eason why simula ion becomes
an essen ial ool o he analysis.
In he case o a complex sys em app oach, e e y single componen beha iou
o he sys em is indi idually modelled acco ding o he beha iou ha each uni
mus ha e. Howe e , he esul s a mac o-scale a e he consequence o all o hese
indi idual beha iou s unning and in e ac ing h ough hei links o each o he .
In o de o explain he main di e ence, a case based on he demand side o a
powe g id is used. Demand side on powe g ids can be modelled in many di e -
en ways ha a e discussed in he nex pa ag aphs. A i s app oach can consis in
aking in o accoun p o iles o consump ion on agg ega ed le els o he g id o be
simula ed (e.g. p o iles modelled h ough an equa ion). E en hough his equa ion
41
4. MODELLING AND SIMULATION FOR SMART GRIDS
may p o ide di e en esul s wi h espec o he p o iles o consump ion, he be-
ha iou o he sys em can be di ec ly in e ed. Ano he app oach can be based on
collec ing da a om appliances and consume s in o de o ully ep esen he de-
mand o a egion. On he one hand, da a om appliances can be used o, acco ding
o his da a, model each indi idual appliance ha is in he g id. On he o he hand,
da a om cus ome s can be used o model he way in which use s in e ac wi h he
appliance wi hin a household.
In his sense, e e y single componen o his complex sys em has been modelled
om hei own poin o iew. Ne e heless, he coope a ion o all o hem wi hin
he sys em p o ides di e en esul s in he o e all consump ion o he g id. This
consump ion canno be in e ed di ec ly om he way in which appliances and
cus ome s can be modelled.
Then, he demand in powe g ids is he esul o he agg ega ion o all he
consump ions o he powe g id. Should dis ibu ed gene a ion be inco po a ed o
he powe g id, his gene a ion can be conside ed as nega i e consump ion. Taking
in o accoun ha demand side policies a e o ien ed o play a ole a he highes
le el o de ail (appliances), i is necessa y o ep esen no only he model o each
di e en kind o de ice, bu also he social beha iou o he people li ing in he
household [PKZK08]. Since people li ing in he household a e sel -in e es ed,
hey a e conside ed as agen s whose ac s cause an e ec a he mac o-scale. This
way o concei ing he modelling o a powe g id as a complex sys em in which
he e a e in elligen agen s is known as agen -based modelling (ABM).
4.3 Agen -based modelling and simula ion
ABM is he compu a ional s udy o social agen s as e ol ing sys ems o au onom-
ous in e ac ing agen s [Jan05]. ABM is a ool ha allows o he s udy o social
sys ems om he poin o iew o an adap i e complex sys em. The e o e, he
esea che is in e es ed in he way in which mac o phenomena a e eme ging as a
consequence o he he e ogeneous indi idual beha iou s ha a e aking place a
he mic o le el [Hol92]. This app oach makes i possible o sys ema ically es di -
e en hypo heses ha a e ela ed o he a ibu es o he agen s, hei beha iou al
ules, hei ypes o in e ac ions and he way in which hey a ec he sys em.
42

4.3 Agen -based modelling and simula ion
An in e es ing discussion which conce ns he use ulness o his app oach o
unning expe imen s is gi en in [Jan05]. The e a e esea che s ha wonde why an
app oach based on ABM may be needed o unning expe imen s. Is i no possible
o app oach he expe imen s based on equa ions? The au ho ’s answe is ha his
decision may depend on he ypes o issue ha a e add essed. Many p oblems
may be aced by using equa ion-based models. Howe e , p oblems ha conce n
coo dina ion o s a egy in e ac ion whe e mul iple agen s a e pa icipa ing need o
be add essed in a di e en way using ABM.
One o he main issues in ABM is he possibili y o ep esen he complex s uc-
u es o social in e ac ions. In some sys ems (e.g. a powe g id), he mac oscale
p ope ies (e.g. o e all demand o ene gy) a e sensi i e o he s uc u e o in e -
ac ions among agen s and social ne wo ks (e.g. cus ome s making decisions ha
a ec ene gy consump ion). Howe e , using an equa ion-based app oach, hese
agen s a e supposed o be implici in hese equa ion models making i impossible
o esea ch on he sensi i i ies o he s uc u e o in e ac ions [Jan05].
The esea ch de eloped on mul iagen sys ems in A i icial In elligence (AI) has
in luenced much he a chi ec u e o agen s p esen in ABM. Mul i-agen esea ch
s udies he adap i e beha iou o au onomous agen s in a conc e e en i onmen
[Jan05]. In elligen agen s a e able o ac lexibly and au onomously [Woo02].
This means ha agen s a e goal-di ec ed (sa is ying o maximizing hei u ili y),
eac i e (adap ing hemsel es o en i onmen al changes) and capable o in e ac ing
wi h o he agen s.
The use o ABM o esea ch and managemen is g owing apidly in a num-
be o ields [RLJ06]. This g ow h is due o he abili y o hese models o add ess
p oblems by he heo y o e olu ion [GRB+05] and s a egies [GR05]. Howe e ,
his kind o modelling is s ill an obs acle o esea che s since i equi es an in-
ensi e so wa e de elopmen in o de o be ca ied ou . Fo his eason, simula-
ion pla o ms ha e been de eloped o make expe imen s using an ABM app oach.
These lib a ies ollow a amewo k and lib a y pa adigm, p o iding a amewo k 1
along wi h a lib a y o so wa e which implemen s he amewo k and p o iding
simula ion ools. These pla o ms ha e succeeded since hey p o ide s anda dised
so wa e designs and ools enabling he simula ion o di e en kinds o models.
1a se o s anda d concep s o designing and desc ibing ABM
43
4. MODELLING AND SIMULATION FOR SMART GRIDS
Howe e , hese pla o ms ha e well-known limi a ions. F om [RLJ06], i e di -
e en pla o ms a e e iewed in he nex pa ag aphs: Ne Logo, Mason, Repas ,
Swa m o Objec i e-C, Swa m o Ja a. Fu he mo e, Anylogic and Flame a e
also e iewed as hey a e also popula solu ions o agen -based simula ion.
a) Ne Logo [TW04]. I is sugges ed o de elop models ha a e compa ible
wi h i s pa adigm o sho - e m, local in e ac ion o agen s and a g id en i onmen
which a e no ex emely complex. Highly ecommended as a ool o p o o yping
models ha can be, la e on, implemen ed in lowe -le el pla o ms. Howe e , i s
simpli ied p og amming en i onmen may make expe ienced p og amme s eel un-
com o able since all code mus be placed in jus one ile (which is con a y o good
p ac ices in objec -o ien a ion p og amming) and he lack o a s epwise debugge .
b) Mason [LCRPS04]. This is a good al e na i e o expe ienced p og amme s
who wo k on models ha a e compu a ionally in ensi e since hey p o ide a good
pe o mance and he bes execu ion ime o he i e pla o ms es ed. Things ha
can be imp o ed a e i s non-s anda dised e minology, i s incompa ible classes wi h
he schedule and i s lack o a e minal window o debugging pu poses.
c) Objec i e-C Swa m [MBLA96]. This e sion o Swa m is s able p o iding
a ai ly comple e se o ools, a clea concep ual basis and cle e design, whe e he
model can be sepa a ed om he in e aces. This ool is o ien ed o help he model
o ganisa ion by allowing he design and implemen a ion o modules in sepa a ed
swa ms, each o one owns objec s and schedules o hei ac ions. This helps o
manage he complexi y o he models which is in e es ing when dealing wi h high
complex sys ems. The d awbacks o his pla o m a e he lack o iendly de elop-
men ools, lack o ga bage collec ion, weak e o handling and low a ailabili y o
documen a ion.
d) Ja a Swa m [MBLA96]. They p o ide he Swa m implemen a ion o Ja a
use s bu i con ains signi ican d awbacks: clumsy wo ka ounds o implemen
Swa m ea u es in Ja a, di icul y on debugging e o s ha happen on Objec i e-C
lib a ies and slow execu ion speed, among o he s.
e) Repas [Col03]. Apa om implemen ing mos o Swa m’s unc ions, hey
ha e added capabili ies like ese and es a models om he g aphical in e ace
and he mul i- un expe imen manage . The execu ion speed is good when com-
pa ed o o he pla o ms. They also p o ide geog aphical and ne wo k suppo .
44
4.4 Swa m In elligence
Howe e , i p esen s weaknesses like he di icul y o ge s a ed on i , especially
o ama eu de elope s, and poo documen a ion.
) Anylogic [BF04]. AnyLogic is a mul i-pa adigm simula o suppo ing ABM
as well as Disc e e E en modeling. I p o ides suppo o de elop lowcha s,
Sys em Dynamics and s ock-and- low desc ip ions. AnyLogic can cap u e a bi -
a y complex logic, in elligen beha iou , spa ial awa eness and dynamically chan-
ging s uc u es. AnyLogic is objec o ien ed and based on he Ja a p og amming
language. To a ce ain deg ee his ensu es a compa ibili y and eusabili y o he
esul ing models [ZLB07].
g) Flame [HCS06]. FLAME (Flexible La ge-scale Agen -based Modelling En-
i onmen ) is an ABM amewo k which allows modelle s om a ious disciplines
like economics, biology and social sciences o easily w i e agen -based models and
simula e hem on pa allel ha dwa e a chi ec u es. The en i onmen allows o c ea e
agen -based models ha can be un in high pe o mance compu e s and g aphical
p ocessing uni s. The simula ion code is gene a ed by p ocessing a model de ini-
ion [KRH+10].
4.4 Swa m In elligence
The Swa m in elligence (SI) concep comes om he ields o AI, Dis ibu ed In-
elligence and Robo ics, whe e one o he main challenges is coo dina ing se e al
obo s. This concep o SI was i s ly in oduced in [BW89]. The au ho was in-
e es ed in how obo s p og ammed wi h simplis ic beha iou s may ou pu in elli-
gence as a esul o he eme gence coming om hei collec i e beha iou . The nex
pa ag aph p esen s how he au ho de ined hese sys ems in which se e al obo s
we e p esen :
“Sys ems o non-in elligen obo s exhibi ing collec i ely in elligen beha iou
e iden in he abili y o p oduce unp edic ably ‘speci ic’ ([e.g.] no in a s a is ical
sense) o de ed pa e ns o ma e in he ex e nal en i onmen .” [BW93]
Based on hese ideas, he au ho in oduced a new concep : SI. Then, Swa m
in elligen sys ems e e o hose sys ems in which each indi idual, o agen , has
been implemen ed wi h simple ules. The main cha ac e is ic o hese sys ems
45
4. MODELLING AND SIMULATION FOR SMART GRIDS
is ha hei eme gence is no p edic able. Once mo e, he au ho ’s de ini ion is
p esen ed o SI:
“Basically swa m in elligen sys ems a e unp edic able o hei de ini ion o
unp edic able (which is ho ough) and hey p oduce esul s ha a e imp obable so
a e in some way su p ising o unexpec ed.” [BW93]
Ano he impo an au ho in he ield o SI is E ic Bonabeau. In [BM01], he
explains SI by exposing main cha ac e is ics o li ing sys ems as o ins ance an
insec colony:
• Flexible: he colony can espond o in e nal pe u ba ions and ex e nal chal-
lenges
• Robus : asks a e comple ed e en i some indi iduals ail
• Decen alised: he e is no cen al con ol(le ) in he colony
• Sel -o ganised: pa hs o solu ions a e eme gen a he han p ede ined
In [BM01], SI is conside ed a mindse a he han a echnology ha uses a
bo om-up app oach o con ol and op imise dis ibu ed sys ems. This bo om-up
app oach ha p o ides an eme gen beha iou is suppo ed by he use o esilien ,
decen alised and sel -o ganised echniques.
In his hesis, he e alua ion o his kind o in elligence is made as a way o
design and pe o m DSM policies. I can be conside ed ha e e y single load o
he demand side is an an . Then, he sys em has a huge amoun o an s which can be
con olled a any ime o balance he powe g id. In he li e a u e e iew he e a e
di e en au ho s ha ha e app oached he DSM policy design and implemen a ion
h ough SI algo i hms.
SI ha e been p e iously used o con ol di e en aspec s in powe g ids. In
[YKF+00], he con ol o eac i e powe and ol age is made using Pa icle Swa m
Op imisa ion. In [Cao04], he powe g id is balanced h ough a collec ion o local
in e ac ions using An Colony Op imisa ion. An ecosys em o in elligen , au onom-
ous and coope a i e an s is p esen ed in [SD06]. These an s make decisions when
he sys em is unbalanced. An imp o ed e sion o he An Colony Op imisa ion is
used in [SS03]. In his e sion, mul iple colonies a e used o op imising he pe -
o mance o a conges ed ne wo k by ou ing ene gy ia se e al al e na i e pa hs.
46
5.3 Model D i en Enginee ing abs ac ion le els
Figu e 5.1: P oposal o MDE abs ac ion le els based on ou laye s.
In o de o be e unde s and hese abs ac concep s, an example ela ed wi h
ilms is explained (Figu e: 5.2). In his example, he ilm Casablanca is in he
lowes -le el o abs ac ion (M0). In his le el, he e a e objec s o he eali y, so
his ilm, Casablanca, e e s o a conc e e ins ance o he ilm (e.g. a conc e e
DVD). This DVD is an ins ance o a ilm which is de ined in M1. M1 may be
de o ed o ep esen he domain o he ilms. To do so, he ilm concep is de ined
he e. Howe e , his ilm is an ins ance o a concep which is coming om he
le el M2. In his le el, a me amodel is de ined in o de o allow de ining concep s
o he ilm domain. Fu he mo e, in M2, a class called A ibu e is de ined. This
class is o ien ed o p o ide he possibili y o gi ing a ibu es in he le el M1. This
is he case o he a ibu e yea in he concep ilm. Finally, a ibu e and concep
a e ins ances o classes, elemen ha is p o ided om M3 allowing o desc ibe and
de ine he me amodel in M2.
Ha ing hese ou le els is powe ul om he poin o iew o he de elopmen .
M0 and M1 le els do no need jus i ica ions as hey a e he common le els ha a e
suppo ed by all objec -o ien ed languages in which he class de ini ion would be
simila o he concep s p esen in M1 and he objec ins ances a e in M0.
In his example, M2 con ains classes ha a e necessa y o de ine elemen s in
M1. Thanks o hese classes, he ilm concep can be de ined. This le el is neces-
53

5. MODEL DRIVEN ENGINEERING
Figu e 5.2: Example o he use o MDE abs ac ion le els (using UML no a ion).
54
5.4 Model D i en Enginee ing applica ion cases
sa y no only o his eason, bu also o ha ing a o mal desc ip ion ha will be
used o gene a o s and ansla o s so ha hey can in e p e M1 models. The e-
o e, based on his M2, gene a o s and ansla o s will be able o p ocess all M1
models ha a e M2 compliance, gene a ing as many so wa e solu ions as di e en
M1 models a e de eloped.
A he same ime, M2 equi es a se o mechanisms o de ine all i s elemen s. To
his end, he me ame amodel, M3, p o ides one mechanism which allows de ining
classes in he M2 le el. This M3 le el is no mally called me ame amodel bu i is
usually a language ha allows de ining classes in M2.
Ano he example can be in o ma ion sys ems. Unde his con ex , M3 may
p o ide a se o mechanisms o de ine classes in M2. These classes de ined in M2
may desc ibe all needed elemen s o building in o ma ion sys ems. Then, M1 may
desc ibe a speci ic de ini ion o an in o ma ion sys em (e.g. in o ma ion sys em o
a speci ic ci y council) being suppo ed by M2. M0 may ep esen he in o ma-
ion sys em in exploi a ion. The e o e, many M1 models ep esen ing in o ma ion
sys ems o se e al o ganisa ions can be de eloped based on classes ha a e in M2.
Gene a o s and ansla o s can build a so wa e solu ion o each o hese M1 mod-
els.
5.4 Model D i en Enginee ing applica ion cases
The e a e many success ul MDE applica ion cases. Fo ins ance, Mo o ola has
been wo king using MDE [BLW05]. In one way o ano he , hey ha e been using
MDE o nea ly wo decades. They ha e ound ha h ough he coo dina ed and
con olled in oduc ion o MDE echniques, signi ican quali y and p oduc i i y
gains can be consis en ly achie ed, and he issues encoun e ed can be handled in a
sys ema ic way.
Ano he example is he applica ion o MDE in eBusinesses [He 09]. This e-
sea ch is o ien ed o help small o ganisa ions ha ha e limi ed esou ces o de ine
hei o ganisa ional and echnological p ojec ion. A conclusion o his esea ch
is ha his app oach is no only use ul o his kind o o ganisa ions bu also o
modelling se ice s uc u es o public adminis a ions.
55
5. MODEL DRIVEN ENGINEERING
MDE has also been ied in he ield o mobile applica ions de elopmen . In
his ield he e a e new challenges ha mus be aced, pa icula ly, he p edic ion
o pe o mance o a gi en design [TWDS09]. The au ho s ha e seen in MDE a
p omising app oach o add ess hese challenges. Using MDE i is possible o
de elope s o quickly unde s and he consequences o a chi ec u al decisions.
A esea ch made in [RMMG08] e alua es he use o MDE in designing adap -
i e mul i-agen sys ems. This is a opic ha is eally close o wha his hesis is
add essing. They conclude ha i wo ked as expec ed o an ad-hoc case bu ha
he e is s ill some wo k o be done o gene alising.
In [SCF+06], he au ho s ackle he main p oblems o au o-gene a ed use in-
e aces. To his end, MDE echniques a e e alua ed. Thei conclusion s a es ha
eusing MDE echnologies may be p omising.
56
CHAPTER
6
Da a analysis and decision
making suppo
Decision making p ocesses ake place in e e y company o o ganisa ion. In
hese ins i u ions, many decisions ha mus be made a e o ien ed o sol e he p ob-
lems o he domain in which hey a e wo king. Since hese decisions ha e epe -
cussions in each ins i u ion, hey mus be made wi h maximum accu acy so ha
bene i s can be maximized.
One o he main keys o succeeding in decision making p ocesses consis s in
ha ing as much in o ma ion as possible which conce n he domain o he decisions
o be made. Howe e , ge ing his in o ma ion may equen ly be impossible o oo
cos ly. Fo his eason, he use o s a egies o ga he in o ma ion o o educe he
unce ain y is impo an in o de o imp o e decision making p ocesses.
The e a e wo key concep ha mus be de ined: decision and decision making.
A decision is de ined as he choice o one among a numbe o al e na i es. Decision
Making e e s o he whole p ocess o making he choice [Boh03]
O e ime, di e en concep s ha e eme ged o suppo decision making p o-
cesses [HW05]. The e a e many o he classi ica ions o all hese concep s ha a e
57
6. DATA ANALYSIS AND DECISION MAKING SUPPORT
CONCEPT EVOLUTION
MIS
Managemen
In o ma ion
Sys ems
DSS
Decision
Suppo
Sys ems
EIS
Execu i e
In o ma ion
Sys em
DW
Da a
Wa ehouse
BI
Business
In elligence
1960 1970 1980 1990 2000 2010
Figu e 6.1: E olu ion o concep s ha ha e add essed he decision making suppo
[HW05].
ela ed o suppo decision making p ocesses. Howe e , checking he li e a u e e-
iew, i can be obse ed ha mos ele an li e a u e o each concep is mainly
concen a ed in he yea s ha igu e 6.1 p esen s.
In his chap e , wo o he concep s a e e iewed in a high le el o de ail: De-
cision Suppo Sys ems (DSS) and Business In elligence (BI). The i s has been
included because i is one o he esea ch opics ha has had mo e ele ance in he
pas o decision making p ocesses. The second one has been included oo because
i is he one ha is nowadays in ogue. Ne e heless, he o he h ee a e b ie ly
desc ibed in he nex pa ag aphs:
a) Managemen In o ma ion Sys em [Swa74]. The main idea is o ha e a da a
bank con aining all he ele an in o ma ion conce ning he company. To his end,
each execu i e o he company will be equipped o emo ely connec h ough e -
minal o a la ge scale compu e ha con ains his da a bank.
b) Execu i e In o ma ion Sys em[Sp 80]. EIS p o ides a se o capabili ies in-
cluding epo p epa a ion, inqui y capabili y, a modelling language, g aphic dis-
play commands, and a se o inancial and s a is ical analysis sub ou ines.
c) Da a Wa ehouse[Gup97]. A da a wa ehouse is a eposi o y o in eg a ed
in o ma ion a ailable o que ying and analysis [IK93, Wid95]. The in o ma ion is
s o ed in se s o iews de i ed om he da a e ie ed om he sou ces.
58

6.1 Decision Suppo Sys ems
Figu e 6.2: Decision making p ocess unde a DSS en i onmen .
6.1 Decision Suppo Sys ems
DSS a e compu e echnology solu ions ha a e used o suppo complex decision
making and p oblem sol ing [SWC+02]. Classic DSS ool design is comp ised
o componen s o (i) sophis ica ed da abase managemen capabili ies wi h access
o in e nal and ex e nal da a, in o ma ion, and knowledge, (ii) powe ul modelling
unc ions accessed by a model managemen sys em, and (iii) powe ul, ye simple,
use in e ace designs ha enable in e ac i e que ies, epo ing, and g aphing unc-
ions. Much esea ch and p ac ical design e o has been conduc ed in each o
hese domains.
The p ocess ha is usually ollowed o making decisions unde a DSS en i on-
men is shown in igu e 6.2 [SWC+02]. The main ocus is made on he model de-
elopmen and p oblem analysis. Once he p oblem is ecognised, his is desc ibed
in e ms ha make he c ea ion o models easie . Models a e c ea ed ep esen ing
al e na i e solu ions o he p oblem aised. Then, one o hese solu ions is selec ed
and implemen ed. This p ocess is i e a i e and has looping backs o p e ious s ages
acco ding o he be e ecogni ion o he p oblem, solu ions ha ail, e c.
This esea ch ield has e ol ed o he las 50 yea s p o iding di e en ap-
p oaches o DSS. The di e en app oaches o DSS de eloped du ing his pe iod can
59
6. DATA ANALYSIS AND DECISION MAKING SUPPORT
be classi ied in one o he ollowing g oups: model-d i en, da a-d i en, communica ion-
d i en, documen -d i en and knowledge-d i en [Pow07].
a) Model-d i en DSS. The i s app oaches o DSS we e model-d i en. Ex-
amples o hem can be seen in [MS84] and [FJ69]. Model-d i en DSS is ocused on
he accessibili y and manipula ion o inancial, op imisa ion and simula ion mod-
els. This is, simple models ha p o ide he mos elemen a y le els o unc ionali y.
Model-d i en DSS use limi ed da a and pa ame e s p o ided by decision make s in
o de o allow he analysis o a si ua ion by he decision make s, no being neces-
sa y la ge da a bases [Pow02]. IFPS (In e ac i e Financial Planning Sys em) is he
i s comme cial ool ha allows o de elop model-d i en DSS based on inancial
and quan i a i e models. This ool, IFPS, was de eloped by Ge a ld R. Wagne
and his s uden s a he Uni e si y o Texas in he 1970s. Ano he DSS ool based
on a model-d i en app oach was Expe Choice [exp]. VisiCalc [ is], a ool de-
eloped by Dan B ickling and Bob F anks on, p o ided he oppo uni y o he
analysis and decision suppo a a easonably low cos . This ool was he i s kille
applica ion o pe sonal compu e s and made possible he de elopmen o many
model-o ien ed, pe sonal DSS o manage s o use.
b) Da a-d i en DSS. Da a-d i en DSS is ocused on he access and manipula-
ion o empo al da a. This da a can be ei he in e nal o ex e nal om he poin
o iew o he company and some imes i is equi ed o deal wi h eal- ime da a.
In [CCS93], he au ho s conside ha Da a-D i en DSS wi h On-Line Analy ical
P ocessing (OLAP) p o ide he highes le el o unc ionali y and decision suppo
o analyse la ge collec ions o da a. In [Nyl99], he de elopmen o BI ela ed o
P oc e & Gamble’s e o was conside ed in 1985. They buil a DSS ha linked
in o ma ion o sales and e ail scanne da a. BI became popula as a e m ha was
coined and p omo ed by Howa d D esne o he Ga ne G oup in 1989. BI was
desc ibed as a se o concep s and me hods ha a e o ien ed o imp o e he decision
making o businesses by using ac -based suppo sys ems.
c) Communica ion-d i en DSS. The use o he ne wo k and communica ion
echnologies in o de o make easie he decision- ele an collabo a ion and com-
munica ion is known as Communica ion-d i en DSS. The dominan elemen o he
a chi ec u e in hese sys ems is he communica ion echnologies such as g oup-
wa e, ideo con e encing and based bulle in boa ds [Pow02]. Apa om hese
60
6.2 Business In elligence and On-Line Analy ical P ocessing
p ima y echnologies, he massi e expansion o he in e ne has made possible
he inc emen o echnologies o synch onous communica ion-d i en DSS such
as oice and ideo deli e ed h ough he in e ne .
d) Documen -d i en DSS. A documen -d i en DSS p o ides documen e ie al
and analysis using a compu e s o age in conjunc ion wi h p ocessing echnologies.
These compu e s o ages may con ain scanned documen s, hype ex documen s,
images, sounds and ideos. This ma e ial can be accessed by documen -d i en
DSS. An example o his kind o DSS is a sea ch engine [Pow02].
e) Knowledge-d i en DSS. Knowledge-d i en DSS solu ions p o ide ecom-
mended ac ions o manage s. These sys ems a e pe son-compu e sys ems wi h
specialized p oblem-sol ing expe ise. Such expe ise consis s o knowledge abou
a pa icula domain, unde s anding o p oblems wi hin ha domain and skills o
sol ing some o hese p oblems [Pow07]. These sys ems a e no mally implemen-
ed based on AI echniques.
6.2 Business In elligence and On-Line Analy ical P o-
cessing
BI is a se o me hodologies, p ocesses, a chi ec u es and echnologies ha com-
bine da a ga he ing, da a s o age, and knowledge managemen wi h analy ical ools
o p esen complex in e nal and compe i i e in o ma ion o planne s and decision
make s [EN08, Neg04]. The objec i e is, he e o e, o imp o e he imeliness and
quali y o inpu s o he decision p ocess by making he access o he in o ma ion
easie . These sys ems a e conside ed o be p oac i e and a e composed by he
ollowing essen ial componen s [LV03]:
• eal- ime da awa ehousing,
• da a mining,
• au oma ed anomaly and excep ion de ec ion,
• p oac i e ale ing wi h au oma ic ecipien de e mina ion,
• seamless ollow- h ough wo k low,
• au oma ic lea ning and e inemen ,
• geog aphic in o ma ion sys ems
61
6. DATA ANALYSIS AND DECISION MAKING SUPPORT
Figu e 6.3: BI da a amewo k.
• da a isualisa ion
The way in which BI ools wo k is h ough con e ing da a in o use ul in o m-
a ion which, using human analysis, is hen con e ed in o knowledge [Neg04]. A
main ask consis s in c ea ing o ecas s based on his o ical da a, pas and cu en
pe o mance, and es ima es ega ding u u e ends. Based on an analysis o he u-
u e, al e na i e scena ios can be designed in o de o e alua e hei impac (Wha i
analysis). Ano he necessa y ask is accessing he da a in o de o answe speci ic
ques ions. This is, ha he da a wa ehouse mus be lexible enough o answe -
ing his kind o ques ions. Fu he mo e, BI ools mus p o ide s a egy insigh s as
esul s o he que ies ha a e launched on he ool.
BI is de o ed o assis in s a egical and ope a ional decision making. An in e -
es ing summa y o his kind o decision makings is p o ided in [Wil02]:
• Co po a e pe o mance managemen
• Op imizing cus ome ela ions, moni o ing business ac i i y, and adi ional
decision suppo
• Packaged s andalone BI applica ions o speci ic ope a ions o s a egies
• Managemen epo ing o BI
One o he main issues wi hin BI is he managemen o semi-s uc u ed da a.
The asks ha a e de eloped by he analys s in o de o deal wi h bo h s uc u ed
and semi-s uc u ed da a is equi ed [RC03].
Semi-s uc u ed da a do no i in o ela ional o la iles, which is called s uc-
u ed da a. A su ey pe o med in [BA03] indica ed ha 60% o Chie In o ma-
ion O ice and Chie Technology O ice conside semi-s uc u ed da a as c i -
ical o imp o ing ope a ions and c ea ing new business models. Examples o
62
7.1 Hypo hesis #1
all sys em. Howe e , he way in which he o e all sys em will beha e canno be
in e ed om hese indi idual beha iou s. The e o e, as said be o e, his sys em
needs o be s udied as a complex sys em.
Fu he mo e, among all he elemen s, he e a e also agen s which a e cha ac-
e ised by being able o pe cei e he en i onmen and ac au onomously acco ding
o hei pa icula goals. Thus, powe g ids may be ep esen ed using an ABM
app oach. This app oach uses he same p inciples as complex sys ems, conside -
ing he inclusion o agen s as an impo an pa o he modelling and simula ion
p ocess.
In he s a e o he a , se e al agen -based pla o ms we e s udied in o de o
pe o m SG simula ions. As a esul , we ha e ound some limi a ions. Ini ially, he
lack o an explici seman ic ep esen a ion in models does no allow he euse, sha -
ing o combina ion. This is due o he ac ha each model has i s own seman ic, so
di e en de elope s can p oduce models ha canno be eused. The e o e, wo king
in a modula way is no acili a ed by hese pla o ms due o hei lack o seman ic
suppo .
Ano he limi a ion is he lack o suppo o de eloping la ge scale models. I is
possible o build la ge scale models o hese pla o ms, bu he de elopmen p ocess
could esemble he complexi y o he models. These pla o ms do no p o ide a
me hodology ha suppo s he de elopmen p ocess in la ge scale models. This is
o say, we ind he e is a lack o a chi ec u al suppo o de eloping la ge-scale
models.
Fu he mo e, hese pla o ms p o ide languages o desc ibing ha a e isola ed
om he p oblem domain. This means a seman ic gap be ween he concep s o
pla o m and modelle s. Ideally, modelle s would be mo e com o able exp essing
hei models in a language close o he p oblem domain.
The app oach o his hesis ocuses on complexi y. MDE has been explo ed as
a me hodology o de eloping an agen -based complex sys em o SGs. The ap-
plica ion o his me hodology in his ield has been esea ched in o de o ep esen
la ge scena ios h ough models. This hypo hesis is s a ed as ollows:
“MDE helps o de elop models o la ge powe g ids using an agen -based ap-
p oach wi h a high le el o disagg ega ion and can also gene a e simula o s acco d-
ing o hese models.”
69

7. MANAGING COMPLEXITY
7.2 Hypo hesis #2
La ge-scale simula ions p o ide a la ge amoun o da a ha needs o be analysed.
This analysis is necessa y since p oblems in he modelling p ocess can be de ec ed
and co ec ed, and consequences o DSM policies can be analysed. These con-
sequences will e u e o alida e he ideas de ined o a DSM policy and based on
his, new ideas can eme ge o imp o e i s design.
New app oaches o da a analysis and managemen de ine he p oblem o dealing
wi h da a as he challenge o he 3 Vs: olume, a ie y and eloci y [Lan01]. This
“3 Vs” challenge was c ea ed by Doug Laney, a Ga ne analys . “Volume” e e s
o he la ge amoun o da a, “ a ie y” o he da a he e ogenei y and “ eloci y” o
he need o ex ac ele an in o ma ion om he da a as as as possible.
The applica ion o BI me hodologies has been esea ched o add ess hese p ob-
lems o dealing wi h simula ion da a ou pu . The use o hese me hodologies o
analyse da a coming om la ge-scale simula ions may be help ul, om he poin
o iew o he design o DSM policies, o iden i y p oblems in he design o he
policy o “bugs” in i s implemen a ion o in he base scena io in which he policy
is es ed. This hypo hesis is s a ed as ollows:
“BI helps o analyse la ge amoun o he e ogeneous da a coming om simula-
ions.”
7.3 Hypo hesis #3
DSM policies aim o modi y he way in which ene gy is consumed in o de o im-
p o e he e iciency o powe g ids. Ne e heless, his modi ica ion o demand may
a ec he consume s by no mee ing hei needs. The e o e, hese policies, espe-
cially hose ha di ec ly ac on demand appliances, mus be designed o conside
consume needs in o de o ensu e a minimum quali y o se ice. Thus, he en i -
onmen in which policies a e ope a ing consis s o many di e en sel -in e es ed
agen s.
This p oblem can be o mula ed as an op imisa ion p oblem in which esou ces
mus be dis ibu ed e icien ly. In his sense, policies based on op imisa ion me h-
ods can be designed in o de o deal wi h eal p oblems on he demand side such
70
7.4 Resea ch conce ns
as he massi e in oduc ion o EVs in o he g id o he modi ica ion o he demand
side acco ding o g id s a es. SI o e s a p omising app oach o dealing wi h his
kind o op imisa ion p oblems. In his documen , he applica ion o his app oach
in he design o DSM policies is explo ed and e alua ed. This hypo hesis is s a ed
as ollows:
“SI helps o design DSM policies by dealing wi h he in e es s o many ac o s
in ol ed in he p ocess.”
7.4 Resea ch conce ns
The hypo heses p esen ed in ol e he applica ion o me hodologies. Thus, he main
esea ch conce n is how o alida e ha hese me hodologies a e going o be help ul
in sol ing he p oblems s a ed.
I is impo an o dis inguish he concep o “me hodology” om he concep o
“me hod”. In his sec ion, bo h concep s a e in oduced and ela ed o each o he o
cla i y he seman ic o he e ms.
On he one hand, a “me hodology” in enginee ing can be conside ed as a guideline
o sol ing a ce ain kind o p oblem. Me hodologies a e gene ally comp ised o
he ollowing ou elemen s: desc ip ion o he p oblem ha needs o be sol ed,
de ini ion o which echniques a e o be used and when hey will be used, gi ing
ad ice on p oduc quali y managemen as well as p o iding ools o acili a e he
p ocess [RH97].
On he o he hand, a me hod is a p ocedu e ha de ines a egula and sys ema ic
way o accomplishing some hing [How12].
Gene ally speaking, me hodologies do no desc ibe speci ic me hods. A me hod
ho oughly de ines he s eps ha ha e o be pe o med in acco dance o a me h-
odology guideline. The e o e, many me hods can be based on one me hodology
p o ided hey ollow he me hodology di ec i es [Vac12].
In gene al, me hodologies canno be alida ed h ough he scien i ic me hod.
This means, he ce ain y o he hypo heses canno be demons a ed wi h expe i-
men s since i is no possible o execu e he same p ojec wi h wo o mo e di e en
me hodologies in o de o compa e hem. Mo eo e , his conce n becomes mo e
de e minan since he execu ion o he p ojec depends on many o he a iables:
71
7. MANAGING COMPLEXITY
human knowledge, he na u e o he sys em, budge , ime, e c. The e o e, i is
almos impossible o isola e he me hodology a iable in he success o a p ojec .
Ne e heless, his wo k is an expe imen al esea ch wi h he goal o p o iding
e idence so as o he alidi y o hese me hodologies o sol ing he p oblems o
enginee ing SGs h ough case s udies. “Case s udy esea ch design” is p oposed as
a aluable and impo an empi ical app oach o alida ing a me hodology [LR04,
ZW97]. This is a use ul me hod o alida ing me hodologies by using hem in eal
wo ld si ua ions.
Gene ally, in expe imen al esea ch, i is necessa y o es ablish e idence o
causali y h ough in e nal alidi y [SAA+02]. Howe e , his is no su icien when
conduc ing expe imen s using enginee ing me hodologies. These expe imen s may
also s udy he ex e nal alidi y o hypo heses, which ensu es hei applica ion in a
b oade spec um o cases a he han only in expe imen al si ua ions.
An expe imen has in e nal alidi y i i demons a es a causal ela ion be ween
wo o mo e a iables [B e00]. A causal in e ence can be based on a ela ion when
h ee c i e ia a e sa is ied [SCC02]: empo al p ecedence (cause p ecedes e ec );
co a ia ion (cause and e ec a e ela ed) and nonspu iousness ( he e a e no mo e
explana ions o he obse ed co a ia ion).
In e nal alidi y is easy o demons a e h ough he esul s. I is also impo an
o demons a e ex e nal alidi y. Ex e nal alidi y is a gene alisa ion o causal
in e ences in scien i ic s udies [MJ12]. Tha is o say, esul s can be ex apola ed
om an expe imen o o he si ua ions.
A his poin , i is necessa y o de ine wha can be unde s ood by “ex e nal
alidi y” in he con ex o his esea ch. In his sense, wo di e en concep s o
ex e nal alidi y can be conside ed. In he i s , i can be said ha he hypo heses o
his esea ch ha e ex e nal alidi y i hey can be applied o mo e han one expe -
imen ela ed o SGs. The second, i could be ha he hypo heses o his esea ch
ha e ex e nal alidi y i hey can be applied o he enginee ing o o he ields apa
om SGs.
In his esea ch, a se o expe imen s has been de ined o alida e he hypo-
heses. Some o hese expe imen s ha e been join ly de ined wi h ou pa ne s
and colleagues om EIFER [ei ] and EDF [ed a]. These expe imen s analyse eal
p oblems in eal powe g ids. These con ibu ions o he expe imen al pa o his
72
7.4 Resea ch conce ns
wo k a e aluable because he alida ion has been made using expe imen s based
on eali y ins ead o only syn he ic p oblems.
Fu he mo e, hese expe imen s a e impo an because hey a e eal cases which
helps o demons a e bo h in e nal alidi y and ex e nal alidi y. The in e nal alid-
i y is demons a ed h ough he esul s o he execu ion o expe imen s. Conce ning
ex e nal alidi y, he i s de ini ion p esen ed is also demons a ed since se e al ex-
pe imen s ha e been conduc ed using he same me hodologies. The second de ini-
ion is no demons a ed in his esea ch bu i could be p oposed as u u e wo k.
73

CHAPTER
8
Sma G id modelling
As said be o e, a simula o is equi ed o each expe imen al s udy. Fo small
expe imen s, he cons uc ion o a simula o is usually simple and as . Howe e ,
la ge expe imen s equi e he cons uc ion o a mo e complex simula o since di -
e en beha iou s mus be implemen ed. Fu he mo e, in la ge expe imen s, he
ini ial expe imen condi ions a e changing cons an ly, so i is necessa y o modi y
he simula o in a as and easy way [PHS+08].
As he simula o cons uc ion may be an a duous ask du ing he expe imen al
s udy, gaining p oduc i i y in his ask is e y impo an . F om he poin o iew
o So wa e Enginee ing, his simula o mus be suppo ed by a good a chi ec u e.
This a chi ec u e should acili a e he simula o cons uc ion and allow he execu-
ion o con inuous changing o equi emen s ha ake place du ing he expe imen al
s udy.
Since he so wa e de elopmen o hese simula ions is ime-consuming, i is
necessa y o ind a way o imp o e de elopmen pe o mance. In he nex sec ions,
a amewo k o de elop simula o s based on MDE app oach is p esen ed. This
amewo k, known as Ta a [EKM+11, EHHK12, EHH13c], is in ended o speed up
he c ea ion o simula o s o domains in which a complex sys em app oach and a
disc e e iming can be applied. Among o he ad an ages, he use o MDE enhances
75
8. SMART GRID MODELLING
Figu e 8.1: Abs ac ion le els o modelling powe g ids.
he capabili y o componen euse and hides implemen a ion de ails p o iding a
high le el language o de elop simula o s.
The adap a ion o MDE o he ield o complex sys em simula ions is he base
ha suppo s he Ta a amewo k. The design o he abs ac ion applied o his
case consis s o h ee le els o abs ac ion. The i s le el, conside ed he lowes
abs ac ion le el, consis s o he conc e e elemen s ha can be ound in a speci ic
domain. The second and hi d le els p opose a way o abs ac he de ini ion o his
conc e e elemen .
In igu e 8.1 hese h ee le els a e ep esen ed o a conc e e case which is e-
la ed o he wo ld o powe g ids. In a i s le el, which is ela ed o he model de-
sc ip ion, conc e e elemen s a e loca ed. This is, a conc e e household, powe line
and cus ome . Howe e , hese conc e e elemen s can be abs ac ed in o a second
le el. The second le el, known as Me amodel, he concep o a household, a powe
line and a cus ome is ep esen ed wi hou conside ing speci ic de ails o a conc e e
ins ance. Based on his concep speci ica ion, he conc e e elemen s can be de ined
by pa ame ising he elemen s o he second le el. A hi d le el o abs ac ion sim-
pli ies he complex sys em wo ld in o h ee di e en ypes o elemen s: en i ies,
connec ions and agen s. The e o e, in his case, a building is conside ed as a en i y,
a powe line as a connec ion and a cus ome as an agen o he powe g id sys em.
76
8.1 Model D i en Enginee ing o modelling complex sys ems
8.1 Model D i en Enginee ing o modelling complex
sys ems
Ta a allows building models o complex sys ems whe e he o e all scene is de-
composed in o di e en componen s. Each componen o he model is s a ically
ep esen ed by means o a ibu es and a iables. In his way, i is modeled a s uc-
u al iew o he sys em. In o de o model he beha iou al iew o he sys em,
each componen may include se e al beha iou s ha desc ibe how his componen
changes o e ime and he way in which i in e ac s wi h o he componen s. Tha
is, he dynamic o he sys em.
This sepa a ion be ween s uc u al and beha iou al iew is a majo design con-
ce n in Ta a since i allows modeling a complex sys em in a modula app oach. A
he same ime, his design echnique is o ien ed o ace he challenge o building
la ge scale simula ions. Basically, when a simula ion is unning, all he beha iou s
a e concu en ly execu ing and modi ying he a iables o hei componen s.
8.1.1 A chi ec u e
The co e componen o Ta a a chi ec u e is he Me amodel (Figu e 8.2) which de-
sc ibes he ypes o ins ances ha could exis in he complex sys em which shall be
ep esen ed. The Me amodel is o ien ed o a speci ic domain and hus es ablishes
a common seman ic which allows modele s o sha e simula ions and models. The
Me amodel allows a ep esen a ion in e ms o en i ies, connec ions, agen s, a ib-
u es, a iables and con ex , among o he s.
The Me amodel can be ansla ed in o se e al o ma s: HTML, XSD and Ja a.
The i s one, HTML, allows he obse a ion o he Me amodel elemen s, as well
as hei p ope ies, in an easy and com o able way by using a b owse . The XSD
ansla ion e u ns a XML scheme model ha helps o alida e he model cons uc-
ion om a seman ic poin o iew and p o ide alid nex okens o add when
w i ing models. Finally, he mos impo an one, he Ja a classes ha implemen s
he Me amodel elemen s and hei ea u es. These Ja a classes a e used in com-
bina ion wi h he Simula o Engine and beha iou s ( eposi o y) in o de o build
simula o s.
77
8. SMART GRID MODELLING
Figu e 8.2: A chi ec u al diag am o he amewo k.
78
8.1 Model D i en Enginee ing o modelling complex sys ems
can widely a y om one o ano he . The e o e, a main ask o de eloping
simula ions consis s in p epa ing da a in o de o use hem in he model.
Depending on he da a o ma , cohe ence, comple eness, complexi y, e c. he
e o o de elop his ask may a y. The inal goal is o ha e he da a wi h
a o ma ha allows i o be pa sed by he P o ile . Howe e , some small
expe imen s may no ha e a complica ed p ocess o p epa ing he da a.
• Model c ea ion. This s ep is di ided in wo main pa s: c ea ion o he sim-
ula ion model and c ea ion/modi ica ion o model elemen s o beha iou s.
The i s one is manda o y o de eloping a simula ion since in his simu-
la ion model he scena io o he expe imen is desc ibed. The second one
depends on he expe imen equi emen s (e.g. a new elemen ha is no in
he Me amodel may be needed o a new beha iou , e c.). Based on he da a
o ma ed in he s ep be o e, he P o ile could help o gene a e he scena io,
especially, i i is a la ge one.
• Model simula ion and calib a ion. The model simula ion s ep inalises wi h
he esul s ga he ing. Howe e , his does no only consis o unning he
model in he simula o and wai ing o he esul s, bu also e i ying and al-
ida ing ha his p ocess is co ec (known as calib a ion). This calib a ion
p ocess conce ns he e i ica ion o he co ec wo king o each simula ion
elemen by checking ha he ou pu s hey ha e a e he expec ed ones. This
can be app oached in se e al ways: using es ing amewo ks, ou pu check-
ing by hand, e c. Fu he mo e, he expe imen equi emen s mus be con on-
ed wi h he simula ion ea u es, so ha , i is checked whe he he simula ion
ea u es ma ch he expec ed equi emen s o no .
• Resul analysis. Acco ding o he expe imen goals, he esul s mus be e al-
ua ed in o de o ob ain conclusions. Some imes, his e alua ion may in ol e
he simula ion o o he simula ion expe imen s in o de o check how he
sys em wo ks unde di e en condi ions. E en i he expe imen planning is
ho oughly de ailed, many condi ions can be in e es ing o be modi ied a e
wa ching he simula ion esul s so ha new simula ion expe imen s may be
ca ied ou .
85

8. SMART GRID MODELLING
8.2 Simula ion pe o mance
Mos o he ools p esen ed in his documen execu e he simula ion wi h a syn-
ch onised app oach. Synch onous simula ions ha e he ad an age o simple ime
managemen as all objec s o he modelled sys em a e unning in he same ime
ins an . I o ces objec s o always pe o m calcula ions, in e e y ime s ep. Some-
imes hese calcula ions a e unnecessa y due o he ac hey canno p o ide new
esul s. Fo example, a washing machine is usually wai ing o an agen o be u ned
on, conside ing his as an e en . La e on, i de elops some washing cycles whe e
he powe may a y along he ime. Whene e he washing machine s a e does no
change, calcula ions could be a oided.
In his sec ion, i is p oposed ha an asynch onous simula ion app oach is in-
cluded in Ta a which would allow objec s o de elop hei own ime as desi ed.
They could beha e bo h e en and ime-based acco ding o hei na u e. Fu he -
mo e, hey could use a iable s eps om one calcula ion o ano he . Fo example,
in an asynch onous simula ion whe e he powe consump ion o a washing ma-
chine is analysed, calcula ions would be done only when he washing machine s a e
changes. The ad an age wi h espec o a synch onous simula ion is clea since in
he synch onised case, calcula ions a e done e e y ime s ep.
In he con ex o disc e e e en simula ion he asynch onous concep has dual
conno a ion. One o hem consis s in a iable ime-inc emen p ocedu es as op-
posed o a “synch onous” o ixed ime-inc emen p ocedu es o simula ion con-
ol. This conno a ion is ela ed o he known concep Dis ibu ed Disc e e E en
Simula ions [Kau87, Mis86]. Fo ins ance, Simula [Poo87], a simula ion-o ien ed
p og amming language, is based on his asynch ony concep whe e he ime man-
agemen is mainly e en -based. This kind o asynch ony was al eady conside ed in
Ta a h ough using di e en ime s eps o each mode o beha iou [EKM+11]. On
he o he hand, he asynch ony can be unde s ood as a non-sequen ial p ocessing
whe e simula ion pa s may no be execu ed in he p ope empo al o de . Tha is
o say, la e pa s o he simula ion may be execu ed be o e p e ious ones [Gho84].
The las conno a ion is he one o which we subsc ibe in his documen . The ob-
jec i e is o apply he ime-managemen o each model elemen allowing hem o
be in di e en ime ins an s.
86
8.2 Simula ion pe o mance
8.2.1 Ta a asynch onous simula ion
This sec ion examines a new app oach o achie e asynch onous simula ions wi h
Ta a . This sec ion in oduces he concep s and cons uc ions ha Ta a a chi ec u e
includes o model powe g ids. Theses cons uc ions a e ocused on dependencies
be ween objec s ha a e massi e and e y ele an in a complex sys em simula ion.
In o de o p ope ly handle an asynch onous simula ion, i is impo an o unde -
s and he dynamics o coupled objec s. Fo he sake o cla i y, a aced execu ion
o objec s in e ac ion du ing an asynch onous simula ion is demons a ed.
8.2.2 Ta a sys em modelling
No mally, a Ta a beha iou is coupled wi h o he objec s, bo h o que ying hei
s a es o sending messages in o de o change hei s a es. In he Ta a model ep-
esen a ion, de ining beha iou which in e ac s wi h o he objec s is allowed.
This ep esen a ion app oach consis s o in e aces ha should be de ined in
he objec which could be ex e nally accessed. In Ta a , he e a e wo ypes o
in e aces:
1. e en in e aces ha handle messages and a e esponsible o modi ying he
objec in e nal a iables as eques ed, and
2. da a in e aces ha handle que ies and p o ide he alue o eques ed a ib-
u es
An example o hese ypes o in e aces is shown in he igu e 8.4. On he
one hand, he he mal beha iou wi hin a household has a da a dependence wi h
he empe a u e o he su ounding Ou doo . In his case, he Ou doo empe a u e
da a is eques ed by he associa ed objec h ough he ou doo da a in e ace. On
he o he hand, an sociological agen beha iou wan s o u n on he washing ma-
chine. Then, his sociological agen mus use he washing machine e en in e ace
o achie e his ask. The washing machine e en in e ace would change he wash-
ing machine mode o ``ON´´. The washing machine ope a ional beha iou would
calcula e he p ope powe consump ion based on his mode. La e on, when he
cycle ends, he ope a ional beha iou u ns o he washing machine.
87
8. SMART GRID MODELLING
Ou doo
Household
Washing
Machine
Ope a ional
Beha iou
Tempe a u e
Beha iou
Agen
Ac i i y
Beha iou
Ac i i y
Beha iou
u n on E en
In e ace
Da a
In e ace
ge (Tempe a u e)
Figu e 8.4: Dependencies examples be ween objec s.
8.2.3 A powe g id simula ion case
In o de o conside he main issues ha in ol e asynch onous simula ion a simula-
ion case is p oposed o show how objec s in e ac when wo king in di e en imes
(Figu e: 8.5).
The objec s wi hin his simula ion case a e an Ou doo , a Household, a Washing
Machine and a Radia o .
• The Ou doo is he objec ha ep esen s en i onmen al condi ions, in his
case, he empe a u e. The Ou doo empe a u e beha iou is esponsible o
se ing he empe a u e which can be loaded om an ex e nal da abase.
• The Household wo ks as a con aine o he appliances o a household, a
Washing Machine and Radia o in his case. The Household Beha iou is
conce ned wi h he he mal dynamics inside he household.
• The Elec ical de ices inside he Household a e a Radia o and a Washing
Machine. These de ices a e handled by an Agen .
88
8.2 Simula ion pe o mance
Ou doo
Household
Washing
Machine
Ope a ional
Beha iou
The mal
Beha iou
Agen
Ac i i y
Beha iou
Tempe a u e
Beha iou
u n on E en
In e ace
Da a
In e ace ge (Tempe a u e)
Radia o
Ope a ional
Beha iou
Da a
In e ace
Da a
In e ace ge (Tempe a u e)
ge (Powe )
Figu e 8.5: Model composi ion.
• Finally, he Agen ep esen s he people li ing in he Household and he as-
socia ed beha iou de ines he ac ions ha hese people a e pe o ming. Fo
example: a pe son u ning on he Washing Machine.
The coupling in his model is ep esen ed by he do ed lines in he igu e 8.5.
This coupling is always de ined om beha iou s o in e aces. The Agen depends
on he Washing Machine o change he ope a ion mode o his de ice. The Radi-
a o depends on he Household empe a u e, since he hea adia ion is calcula ed
based on he gap be ween he Radia o e e ence empe a u e and he Household
empe a u e. The Household has wo dependencies: wi h he Ou doo empe a u e
and wi h he Radia o powe , since he Household empe a u e is calcula ed by a
nume ical solu ion o a di e en ial equa ion which includes hese wo a iables.
No e ha , in his case, he e is a cyclic dependence be ween he Household and he
Radia o .
89
8. SMART GRID MODELLING
8.2.4 Asynch onous simula ion dynamics
A sys em simula ion equi es ime-managemen o ensu e ha empo al aspec s
a e co ec ly ep esen ed and emula ed. This empo al ep esen a ion only exis s
du ing he simula ion p ocess and is e e ed o as “Simula ion Time”. Simula ion
Time is ep esen ed as a imes amp, a long in ege whe e a uni co esponds o a
millisecond o eal ime.
The ime-managemen in a synch onous simula ion is cen alised while he
ime-managemen in an asynch onous simula ion is dis ibu ed. Tha is, an asyn-
ch onous simula ion in ol es ha e e y objec manages i s ime, so hey could ha e
di e en imes amps (Figu e: 8.6).
Ou doo
Household
Agen Washing
machine
Radia o
Simula ion Time
Ou doo
Household
Agen Washing
machine
Radia o
Simula ion Time
Simula ion Time
Simula ion Time
Simula ion Time
Simula ion Time
Figu e 8.6: Synch onous s Asynch onous simula ion.
In his simula ion pa adigm, when an objec is no coupled wi h o he objec s,
i s Simula ion Time de elops wi hou conside ing o he objec Simula ion Times.
In his simula ion case, he Ou doo is comple ely independen o o he objec s.
Howe e , when objec s a e coupled, he challenge consis s o co ec ly ep o-
ducing empo al ela ionships. The iden i ied empo al ela ionships a e as ollows:
1. Coupling wi h a da a in e ace
2. Cyclic coupling wi h da a in e aces
3. Coupling wi h an e en in e ace
90

8.2 Simula ion pe o mance
In he ollowing sec ions hese ela ionships a e discussed.
8.2.4.1 Coupling wi h a da a in e ace
Since an objec could access a a iable o an ex e nal objec which may be in a
di e en ime ins an , e e y objec mus keep he di e en s a es ha ha e been
calcula ed du ing he simula ion execu ion. So, when a a iable is modi ied, a s a e
snapsho is c ea ed in o de o keep he objec s a e in his ime ins an .
I an objec is que ying o a a iable alue in a ime ins an i, he e a e wo
cases: he objec Simula ion Time is delayed o ahead wi h espec o he ex e nal
objec Simula ion Time. In he i s case, he ex e nal objec is able o p o ide
he alue by e ie ing he las snapsho p e ious o his ime ins an ( i). In he
second case, he dependen objec mus wai un il he ex e nal objec eaches his
ime ins an ( i).
Figu e 8.7: The Household asks he Ou doo . Time is e ically ep esen ed.
In he igu e 8.7, he i s case is shown. The Household Simula ion Time is
iand he Ou doo Simula ion Time is j. Whene e iis lesse o equal han j,
he eques ed da a can be deli e ed since he da a has al eady been calcula ed and
s o ed.
Howe e , when he Household Simula ion Time ( i) is g ea e han he Ou doo
Simula ion Time ( j), he Household beha iou is blocked (Figu e: 8.8) un il jis
g ea e o equal han i(Figu e: 8.9) deli e ing he las Ou doo Tempe a u e alue
s o ed in he las calcula ed snapsho .
91
8. SMART GRID MODELLING
Figu e 8.8: The Household beha iou eques ge s blocked.
Figu e 8.9: The da a is deli e ed.
8.2.4.2 Cyclic coupling wi h da a in e aces
The cyclic dependence is a conc e e case o he da a dependence. Two objec s
depending on each o he whose Simula ion Times a e di e en , is handled wi h he
ollowing ules: he mos delayed one will always e ie e he equi ed da a while
he mos ad anced will be blocked un il he delayed eaches i s Simula ion Time
(Figu e: 8.10). The mu ual blocking is no possible since objec s e ie e he alue
o he cu en Simula ion Time o calcula e he nex Simula ion Time alue.
In he example shown in he igu e 8.10, he Household equi es he powe
consump ion o he Radia o in o de o calcula e he new empe a u e alue. On
he o he hand, he Radia o beha iou needs he Household empe a u e alue o
92
8.2 Simula ion pe o mance
Figu e 8.10: Radia o and Household cyclic dependence esolu ion.
modi y he Radia o s a e, since he e e ence empe a u e a he Radia o he mo-
s a se es as a con ol mechanism.
8.2.4.3 Coupling wi h an e en in e ace
The e en coupling means ha an objec ecei es ex e nal messages ha con ain
o de s o changing i s in e nal a iables. This is he case o objec s which a e
managed by people ha a e ep esen ed as Agen s in he model. The Agen in-
e ac s wi h hese objec s by sending a message using he objec e en in e ace.
When he message is ecei ed by he objec in e ace, he objec Simula ion Time
is de eloped and hen, a new snapsho s a e is c ea ed.
I could happen ha he Agen de elops i s Simula ion Time wi hou he in en-
ion o sending an o de o any objec . In his case, he Agen beha iou mus send
a ``No i ica ion Time Message´´ o he objec . In ac , when he Agen Simula ion
Time de elops, he Agen beha iou mus send a No i ica ion Time Message o all
objec s he Agen is con olling. This no i ica ion de e mines how long an objec
can de elop i s Simula ion Time. This ype o ela ionship means ha objec ’s Sim-
ula ion Time ha is con olled by an Agen , will ne e exceed he Agen Simula ion
Time.
Figu es 8.11-8.14 shows an e en ela ionship be ween a social Agen ha u ns
on he Washing Machine. In his example, he Washing Machine Simula ion Time
93
8. SMART GRID MODELLING
Figu e 8.11: The Agen sends a message o u n on he Washing Machine.
Figu e 8.12: The Washing Machine e en in e ace changes he objec mode o on.
is always behind he Agen Simula ion Time. In o he wo ds, he Agen Simula ion
Time se s a es ic ion o he Washing Machine Simula ion Time.
In he case o he Washing Machine, i s powe consump ion would be 0 a
he beginning o he simula ion as i ’s o . The e o e, a new snapsho is c ea ed
when he Agen u ns on he Washing Machine. F om ha momen , he Washing
Machine beha iou will calcula e he new powe consump ion wi h he es ic ion
ha he calcula ions de elopmen should no exceed he Agen Simula ion Time, in
case he Agen u ns o he Washing Machine.
94
CHAPTER
9
Simula ion esul s analysis
As said be o e, simula ions play a c ucial ole in he design o SG policies since
hey a e a way o es hem be o e hei launch. Howe e , he ou pu p o ided by
he simula ions mus be managed in a way ha allows he policy designe s o make
decisions. This sec ion explains he main conce ns when analysing esul s ob ained
in a SG simula ion. When acing a simula ion o SGs based on a complex sys em
app oach, he esul s analysis becomes a di icul s age since he amoun o en i ies
is la ge.
All sys ems con aining a la ge amoun o en i ies and ela ions in simula ion
p ocesses p o ide a la ge amoun o esul s. The way in which hese da a a e no -
mally expo ed is h ough da a iles. These da a iles a e usually designed acco ding
o he da a ha will be managed hus a oiding he possibili y o que ying his da a
beyond wha was decided o expo . The e o e, whene e we deem i con enien
o ex ac da a, which was no conside ed o expo a he design phase, a new
simula ion mus be con igu ed and execu ed.
In o de o exempli y his issue, a disagg ega ed model o a powe g id sys em
is used. This sys em only consis s o he demand side, which is disagg ega ed a
he de ice le el. I is p ecisely a his le el whe e we can ind a laye consis ing o
he e ogeneous elemen s, since he cha ac e is ics o ex ac om a adia o a e no
he same as he ones om a ele ision (TV). I we wan o p ese e all a iables ha
101

9. SIMULATION RESULTS ANALYSIS
Figu e 9.1: S uc u e example o expo simula ion esul s.
a e no common o e e y de ice, i will be necessa y o expo each de ice ype in o
a di e en da a shee (Figu e: 9.1). A his poin , once he da a expo a ion p ocess
has been de ined, we can s a hinking abou que ying i . The lis below s a es
some que y examples and how hey should be deal wi h acco ding o his da a
expo a ion s uc u e:
•Que ying he consump ion o all de ices. This que y is e y likely o be
equi ed. Acco ding o ou da a s uc u e, i s ly we calcula e he o al con-
sump ion a each de ice ype. This would in ol e opening as many iles as
de ice ypes and making he calcula ions o ob ain he o al consump ion pe
de ice ype. Secondly, hose columns which ha e he agg ega ed alue a
each de ice ype mus be mo ed in o a new shee whe e he inal calcula ion
would be pe o med ob aining he que y esul . The mo e de ice ypes he e
a e, he ickie his p ocess becomes.
•Que ying he consump ion o all de ices in a speci ic household. This
p ocess would consis in ga he ing he columns belonging o all he de ices
con ained in he household om he da a iles. Once hey a e all oge he in
a new shee , he que y esul can be ob ained by adding up.
•Que ying he consump ion o all de ices in a speci ic dis ic . The p o-
cess o ob ain his que y is eally icky. Fi s ly, all he de ices belonging
o a speci ic dis ic mus be lis ed. Nex , all he columns which e e o he
de ices consump ion mus be ga he ed om he de ice ype shee s ollowing
his lis . Finally, all ga he ed columns can be mo ed o a new shee whe e
he que y can be ob ained.
Taking hese examples in o accoun , i is possible o imagine how icky he
esul s managemen o mo e complica ed que ies can ge . P obably, some o hese
102
9.1 Business In elligence me hodologies o analysing da a
que ies a e easie o ob ain by ede ining he simula ion esul s o ma and unning
i again. Howe e , i would also be eally edious, and depending on he simula ion
kind, he esul s may di e om he p e ious simula ion and in he end i would be
necessa y o s a he esul analysis om he beginning.
All hese di icul ies in que ying he ou pu o a simula ion could in ol e ha
many o he que ies a e no made due o he ac ha hey in ol e a s ong and
ime consuming e o o pe o m hem. Un o una ely, his usually leads o ocus
on a small subse o a iables o he simula ion neglec ing much in o ma ion and
was ing oo much ime in pe o ming simple que ies.
The oo o he p oblem behind he esul analysis is ha such esul s ha e a
mul i-dimensional and a mul i-scale (namely empo al and spa ial) na u e which
canno be managed by using con en ional da a shee s. The example o he de-
mand disagg ega ion is mul i-dimensional and mul i-scale. Mul i-dimensional,
since each da ase ( o example, a powe measu e) is ela ed o a speci ic de ice,
loca ion (household, building, dis ic , e c.) and ime. Mul i-scale, since he in-
o ma ion can be agg ega ed a di e en ime scales (pe hou , pe day, pe mon h,
e c.) and a di e en spa ial le els (de ice, household, e c.).
9.1 Business In elligence me hodologies o analysing
da a
A amewo k, named Sumus, o applying BI me hodologies has been de eloped.
This amewo k o da a analysis is based on OLAP. OLAP is a solu ion used in
BI, he aim o which is o accele a e que ying la ge amoun o da a. OLAP is
based on cubes [CD97](Figu e: 9.2), a mul i-dimensional s uc u e whe e da a is
s o ed. These cubes enable he inse ion o da a, namely ac s, which a e e e ed
o se e al dimensions. Fo example, he measu e o powe aken om a washing
machine can be e e ed o he de ice, he household whe e he de ice is and he
ime. The e o e, in his case, he e would be h ee dimensions: de ices, households
and ime.
The s uc u e o a mul i-dimensional cube which add esses ou p oblem is
103
9. SIMULATION RESULTS ANALYSIS
Figu e 9.2: A mul i-dimensional cube o esiden ial consump ion.
Figu e 9.3: An OLAP Cube s uc u e.
p esen ed in he igu e 9.3. E e y cube consis s o dimensions, measu es and in-
dica o s. The lis below desc ibes e e y cube componen .
•Dimension: i es ablishes a way o access he da a inside he cube. E e y
single da a is ela ed o some elemen s such as when and whe e i happened.
Fo example, a da a o powe consump ion o a household would be ela ed
o he dimensions household and ime.
– Componen : i is an elemen which is ela ed o a dimension. Fo
example, a dimension which conce ns households would be illed by
componen s which a e households.
104
9.1 Business In elligence me hodologies o analysing da a
*Fea u e: i is a p ope y o he componen . In case he componen s
a e households, a possible ea u e could be he numbe o squa e
me e s he e is in each household.
– Taxonomy: i is a way o ca ego izing a dimension. The e a e di e en
ways o ca ego ize he componen s inside a dimension. Each o hese
ways is known as axonomy. In he example o he household dimen-
sion, a axonomy could be he size o he o ien a ion o he acade.
*Ca ego y: i is a se o componen s ha sa is y some speci ic con-
di ions. Fo ins ance, possible ca ego ies o he size axonomy
could be small, medium o big. The e o e, each o hese ca ego ies
would con ain a se o household componen s he ela ionship o
which is ha ing a simila size.
·Rule: i es ablishes he condi ion ha a componen mus mee
in o de o all in o he ca ego y ha owns he ule. In he
case o he small ca ego y, a possible ule could be: all he
household componen s he ea u e o which numbe o squa e
me e s is below 80m2
•Measu e: i p o ides a seman ic o he da a inse ed in he cube, e.g. he
powe o he household men ioned abo e is jus a numbe . Howe e , he
powe measu e is wha p o ides he seman ic o his numbe . A measu e is
usually ela ed o a me ic which enables he compa ison among measu es
ha a e in di e en cubes. In his case, he me ic o he powe measu e
would be Wa s.
•Indica o : i designa es he way in which a measu e o a se o measu es a e
agg ega ed. Fo example, he powe measu e could be agg ega ed using an
a e age unc ion. This way o agg ega ing measu es is known as indica o .
I is possible o ha e se e al indica o s o one measu e, i.e. he in eg al
ope a o o e he powe measu e would p o ide a second indica o o e his
measu e which could be designa ed as ene gy indica o .
•Fac : i ela es he measu es o a cube wi h he dimensions. A ac indica es
ha a ce ain combina ion o alues (measu es) ook place o a speci ic com-
bina ion o elemen s (componen s). In o he wo ds, a ac can be unde s ood
105
9. SIMULATION RESULTS ANALYSIS
Figu e 9.4: A ac consis s o con ex and s a e.
as a ela ion o a s a e o a con ex . The s a e is a se o measu es and he con-
ex consis s o componen s including ime. In igu e 9.4, he s a e con ains
20 (cen ig ades) and 135 (Wa s) as measu es. These measu es a e ela ed o
a con ex which indica es he ime and household whe e hose measu es we e
aken.
9.2 On-Line Analy ical P ocessing o Sma G ids
In his sec ion, all concep s exposed p e iously will be used in a p ac ical case.
Assuming ha a new SG policy is o be uned, se e al simula ions o powe g id
demand will be pe o med. To make decisions, hese simula ions mus ocus on he
powe demand and he empe a u e a he esiden ial sec o . The e o e, he scena io
o hose simula ions consis s o se e al dis ic s wi h households (Figu e: 9.5).
Each household con ains se e al de ices and calcula es he in e nal empe a u e.
To his end, se e al cubes ha e been designed so as o analyse he da a com-
ing om he simula ion: i s o all, he household cube which con ains he ac s
ega ding he empe a u e and, secondly, one cube pe de ice ype (TVs, Radia -
o s and Washing Machines, among o he s) which con ain ac s abou he de ices.
Since he e a e many kinds o de ices in a household, in his example we a e going
o ocus on wo o hem: TVs and adia o s.
The e a e wo dimensions in he household (HH) cube: one measu e and one
indica o (Figu e: 9.6). The Time dimension is common o all cubes and con igu es
a s anda d way o ca ego izing he imeline. Household dimension con ains he
106

9.2 On-Line Analy ical P ocessing o Sma G ids
Figu e 9.5: Scena io composi ion.
Figu e 9.6: Household cube.
households ans o med in o componen s which a e desc ibed by ea u es. The
empe a u e o he household is he only measu e ha his cube is going o s o e
and i will be agg ega ed using an a e age c i e ia acco ding o he designa ion o
he indica o .
The household dimension con ains a axonomy which conce ns he loca ions
(Figu e: 9.7). This axonomy is ca ego ized ollowing se e al le els: coun y, ci y
and dis ic . Fo ins ance, wo household componen s ha e been included, bo h o
which con ain a ea u e which is hei loca ion using UTM coo dina es. The e o e,
hese loca ion ea u es allow he dimension o iden i y which dis ic each house-
hold is loca ed in.
The TV and Radia o cases a e exposed in o de o demons a e why de ices
mus be disagg ega ed in o sepa a ed cubes. The main eason o his sepa a ion
is due o he ac ha bo h de ices do no sha e he same ea u es and, he e o e,
hei classi ica ion me hods a e di e en . This sepa a ion enhances he capaci y o
making que ies since i is possible o il e componen s by ea u es ha a e only
p esen in a speci ic kind o de ice.
107
9. SIMULATION RESULTS ANALYSIS
Figu e 9.7: Household dimension.
Figu e 9.8: TV cube.
The TV cube egis e s da a abou powe consump ion as well as he TV mode
(o , s andby and on) (Figu e: 9.8). E e y se o measu es (powe and mode) is
ela ed o h ee dimensions: ime, household and TV. Time and household dimen-
sions a e exac ly he same dimensions as he ones de ailed abo e. The TV di-
mension con ains in o ma ion abou he TVs in a componen o ma . Fu he mo e,
he e a e wo indica o s which a e esponsible o agg ega ing measu es: he mode
indica o , which pe o ms a calcula ion ha p o ides he pe cen age o TVs ha a e
u ned on, and he powe indica o , which agg ega es he powe measu es egis e ed
using an a e age o mula.
The TV dimension, like he household dimension, ocuses on speci ic ea u es
ela ed o TV componen s (Figu e: 9.9). In his case, he possibili y o il e ing
TVs using a echnological c i e ia is conside ed ele an . The e o e, wo ca ego ies
108
9.2 On-Line Analy ical P ocessing o Sma G ids
Figu e 9.9: TV dimension.
Figu e 9.10: Radia o cube.
ha e been c ea ed so as o sepa a e LED ele isions om LCD ele isions. This
in o ma ion will allow us o compa e he consump ion among he di e en TV
echnologies. Hence, TV componen s con ain he echnology ea u e which will be
used o calcula e whe he a TV belongs o he LED o LCD ca ego y by using he
ules ha a e ela ed o hese ca ego ies.
The adia o cube s o es measu es ela ed o bo h he powe consump ion and
he he mos a le el (Figu e: 9.10). These measu es a e ela ed o h ee dimensions,
as in he case o he TV cube. In his case, apa om ime and household dimen-
sions, a new dimension has been designed: adia o dimension. This dimension
con ains componen s ha ep esen adia o s and hei ea u es. In addi ion, he e
a e wo indica o s which agg ega e he measu es. On he one hand, he he mos a
indica o agg ega es he measu es s o ed using a g adien unc ion which shows
big changes in he he mos a le el in sho pe iods o ime. On he o he hand, he
powe indica o agg ega es he powe measu es using an a e age o mula like in
he TV cube.
109
9. SIMULATION RESULTS ANALYSIS
Figu e 9.11: Radia o dimension.
The adia o dimension ocuses on speci ic ea u es which conce n adia o
componen s (Figu e: 9.11). Since adia o s a e usually conside ed big consume s,
a axonomy o classi y hem in o wo g oups has been designed. Indeed, his ax-
onomy will allow us o ind ou he amoun o adia o componen s which a e in
wha we conside a small consume ca ego y (unde 1kW ins alled powe ) o a
big consume ca ego y (o e 1kW). Two componen s belong o his dimension and
con ain he ea u e ins alled powe which is used o pe o m he classi ica ion in
he ins alled powe axonomy.
9.2.1 N-Le el indica o s and Da a mining
So a , some mechanisms which allow us o ex ac in o ma ion based on he meas-
u es ha e been p esen ed: indica o s. These indica o s a e ega ded as i s le el
indica o s since hey a e jus based on measu es. Howe e , i is possible o de ine,
a second le el o indica o s which a e compu a ions ca ied ou based on p e ious
le el indica o s. This idea can be ex ended o he concep o N-Le el indica o s.
Th ough in oducing his concep , da a mining[RKU11] p ocedu es can be used in
o de o ind ou pa e ns.
An example o his is p esen ed in igu e 9.12. In his case, a da a mine has
been designed in o de o iden i y consump ion habi s which conce n adia o s. Us-
ing he he mos a indica o , which calcula es he g adien based on he he mos a
110
10.3 Mul i Objec i e Op imisa ion
Whe e id( )is he eloci y o he pa icle iin he dimension da ime .xid( )
is he posi ion o he pa icle iin he dimension da ime .c1and c2a e weigh
ac o s ha adjus he mo emen . pbes iis he bes posi ion achie ed by he pa icle
i.pbes gis he bes posi ion o he neighbou s o he pa icle i.u1and u2a e
andom ac o s in an in e al o [0,1] and wis he ine ial weigh .
This op imisa ion me hod is ini ialised wi h a andomly-gene a ed popula ion
o pa icles in he decision space which y o con e ge in o an op imal one by
ollowing he bes pa icles a e e y i e a ion. The algo i hm o his me hod is
p esen ed below:
I n i i a l i s e swa m .
I n i i a l i s e p b e s . Upda e g b e s . I n i i a l i s e e l o c i y .
i e a ionCoun e = 0
while (i e a ionCoun e <maxI e a ion ){
o ( each p a i c l e ) {
Pick andom u1 and u2
o ( each dimension ) {
Upda e p a i c l e e l o c i y and p o s i i o n
}
Upda e p b e s
}
Upda e g b e s
i e a ionCoun e ++
}
Repo e s u l s i n a c h i e
Lis ing 10.1: PSO code. Fo u he in o ma ion please check he bibliog aphy
10.3 Mul i Objec i e Op imisa ion
A mul i-objec i e op imisa ion p oblem can be exp essed as ollows:
min
x
~
F(x) = [F1(x), F2(x)...Fk(x)]T(10.3)
Subjec o:
gj(x)≤0, j = 1,2...m (10.4)
hl(x)=0, l = 1,2...e (10.5)
Whe e K is he numbe o objec i e unc ions and j and l a e, espec i ely, he
numbe o inequali ies and equali y cons ain s. x∈Enis he ec o o design
117

10. DEMAND SIDE MANAGEMENT POLICY DESIGN
a iables and ~
F(x)∈Ekis he ec o o objec i es, c i e ia, i ness o cos unc-
ions o be op imised. Any compa ison (≤,≥, e c.) among ec o s applies o ec-
o componen s [MA04]. O he p ima y concep s in mul i-objec i e op imisa ion
a e non-domina ed and domina ed poin s [S e89]. A ec o o objec i e unc ions
~
F(x∗)∈Zis non-domina ed i he e is no ano he ec o , ~
F(x)∈Z, such as
~
F(x)≤~
F(x∗)wi h a leas one Fi(x)< Fi(x∗). O he wise, ~
F(x∗)is domina ed.
[LTDZ02] p oposed a elaxed o m o dominance named -dominance. This
ac s as an a chi ing s a egy o ensu e bo h p ope ies o con e gence owa ds
he Pa e o-op imal se and p ope ies o di e si y among he solu ions ound. -
dominance is p oposed as an ex ension o he Pa e o-dominance ela ion so ha a
poin x∗no only domina es hose poin s xlowe o equal in all hei objec i es
~
F(x)≤~
F(x∗)and s ic ly lowe in a leas one objec i e Fi(x)< Fi(x∗), bu also
all poin s close enough o x∗(i.e., hose wi h a dis ance o x∗ ha is less han an
). This alue, , can be p o ided by he decision make o con ol he size o he
solu ion se [HDSQCCM07].
In mul i-objec i e op imisa ion p oblems, he op imum solu ion is, in gene al,
no as easy o o mula e as in single objec i e cases. Since ypically he e is no
single global solu ion, i is o en necessa y o de e mine a se o poin s ha all
ma ch a p ede e mined de ini ion o an op imum o e e y single p oblem [MA04].
In his sense, he p edominan concep in de ining an op imal poin is ha o Pa e o
op imali y [Pa 06].
A poin ~
x∗∈~
Xis Pa e o op imal i he e does no exis ano he poin , x∈X
such as ~
F(x)≤~
F(x∗)and Fi(x)< Fi(x∗) o a leas one unc ion. All Pa e o
op imal poin s lie in he bounda y o he easible c i e ion space Z[AP96]. A
Pa e o-op ima ha dly e e p o ides a single solu ion, bu a he a se o solu ions
called non-in e io o non-domina ed solu ions. The minima in he Pa e o sense a e
going o be in he bounda y o he objec i e egion, o in he locus o he angen
poin s o he objec i e unc ions, ha is in he egion in he space o alues o he
objec i e unc ions ec o [Coe98].
O en, algo i hms p o ide solu ions ha may no be Pa e o-op imal bu may
ul il o he c i e ia, which can be use ul o p ac ical applica ions. I is he case
o weakly Pa e o op imal. A poin ~
x∗∈~
Xis weakly Pa e o-op imal i he e is
no ano he poin , ~
x∗∈~
Xsuch as ~
F(x)≤~
F(x∗). Tha is, a poin is weakly
Pa e o op imal i he e is no o he poin ha imp o es all he objec i e unc ions
118
10.4 Mul i Objec i e Pa icle Swa m Op imisa ion
simul aneously.
10.4 Mul i Objec i e Pa icle Swa m Op imisa ion
The implemen a ion o he MOPSO used in hese s udies is desc ibed in [SC05].
This op imisa ion me hod is based on PSO. In his me hod, he i ness o a pa icle
is calcula ed conside ing se e al objec i es which a e de ined h ough i ness unc-
ions. This me hod uses he Pa e o-op imali y [SC05] app oach o add ess he
mul i-objec i e p oblem.
A se con aining he bes pa icles aised du ing he i e a ions is s o ed in wha
is named as a chi e, which e ie es he elemen s ha each he -dominance. This
dominance elaxes he weak-dominance cons ain s. The nex pseudo-code ex-
plains how he MOPSO wo ks:
I n i i a l i s e swa m .
I n i i a l i s e l e a d e s . Send l e a d e s o a ch i e . c owding ( l e a d e s ) .
i e a ionCoun e = 0
while (i e a ionCoun e <maxI e a ion ){
o ( each p a i c l e ) {
S e l e c l e a d e . F l i g h . Mu a ion . E a l u a i o n . Upda e p be s .
}
Upda e l e a de s , Send l e a d e s o a c h i e . c owding ( l e a d e s ) .
i e a ionCoun e ++
}
Repo e s u l s i n a c h i e
Lis ing 10.2: MOPSO code. Fo u he in o ma ion please check he bibliog aphy
Ini ially he swa m is popula ed h ough he c ea ion o he pa icles o indi-
iduals. A pa icle con ains a sys em con igu a ion which is se up andomly. This
con igu a ion is he pa icle gene ic code. Be o e s a ing he i e a i e p ocess,
he leade s a e calcula ed using he i ness unc ions which e alua e he op imali y
o he pa icles. Those ha each he -dominance c i e ia will be s o ed in he
a chi e. Tha is, he algo i hm includes a c owding p ocess ha is used o es ablish
a second disc imina ion c i e ion (addi ional o Pa e o dominance) [CL02, SC05].
Along he execu ion his a chi e may be illed up and, in hese cases, he a chi e
dele es pa icles based on he c owding ac o c i e ia. A e e y me hod i e a ion,
each pa icle is mo ed acco ding o i s p e ious posi ion and speed which a e cal-
cula ed as ollows:
119
10. DEMAND SIDE MANAGEMENT POLICY DESIGN
• The speed is calcula ed based on he p e ious pa icle speed, he dis ance o
i s bes posi ion and he dis ance o i s leade . This me hod spli s he sea ch
space in hype cubes, so ha he pa icle leade will be he one ha has he
bes i ness wi hin he hype cube.
• The posi ion is calcula ed adding he p e ious pa icle posi ion o he speed
calcula ed abo e.
A e he pa icle mo emen , he pa icle could be mu a ed wi h a p obabili y
calcula ed as 1 / gene ic code size (mu a ion a e). I a pa icle is going o be
mu a ed, his mu a ion may be uni o m o non-uni o m. The uni o m mu a ion
allows he pa icle o explo e he sea ch space. Howe e , a non-uni o m mu a ion
allows he pa icle o exploi he sea ch space whe e e i is. La e on, he pa icle
i ness is e alua ed and he bes posi ion o he pa icle is changed whene e he
new posi ion is be e han he bes achie ed in he pas .
Once all he pa icles ha e been p ocessed, he whole popula ion is e alua ed in
o de o upda e he leade s o he a chi e. I his a chi e eaches he size limi , some
o he pa icles inside i will be dele ed ollowing he same c i e ia as desc ibed
abo e. A e his s ep, he nex i e a ion will be execu ed. The numbe o i e a ions
mus be se be o ehand.
The eason why his me hod was selec ed ins ead o o he s such as Gene ics
Algo i hms is ha MOPSO has a good esul quali y and ime esponse ela ion.
The ime esponse is impo an since he o de s could a ise ega ding he cu en
p oduc ion-consump ion balance in a eal- ime sys em whe e he speediness o ap-
plying he measu emen s is c i ical.
120
Pa IV
Expe imen a ion
121

CHAPTER
11
Expe imen a ion
conside a ions
This pa o he documen is de o ed o show case s udies ha ha e been ca ied
ou in his esea ch. As p e iously said in chap e 7, he hypo heses ha ha e
been s a ed in his documen canno be e i ied, bu alida ed. Tha is, i is no
possible o e i y hem as o do so i would be necessa y o ca y ou all SG s udies.
Ne e heless, hey can be alida ed h ough he expe imen a ion wi h case s udies.
In his pa , hese case s udies a e p esen ed and explained, p o iding e idence ha
alida es he s a ed hypo heses.
In his chap e , some syn he ic case s udies a e p esen ed. The s udies ha e
been designed o show how he ools desc ibed in he hypo heses pa wo k. Tha
is, how a me amodel is de ined, how a simula ion model is de ined, how o c ea e
la ge-scale scena ios, e c. Each o he chap e s in his pa e e s o a speci ic case
s udy. These case s udies a e o ien ed o sol e eal p oblems and, he e o e, hey
can be conside ed as impo an e idences ha alida e he hypo heses.
123
11. EXPERIMENTATION CONSIDERATIONS
11.1 Modelling in Ta a
The wo k, published in [EHH13c] 1, p esen ed in his sec ion conce ns p ospec -
i e expe imen s ha we e de eloped o s udying he human lows on a shopping
cen e om he poin o iew o he amoun o people. In o de o un hese expe i-
men s, he p oposed amewo k will be cus omised. Fo his he de elopmen o he
me amodel, he beha io s and models in which he expe imen s will be desc ibed
is necessa y.
Howe e , as he aim o his case s udy is no o p o ide accu a e esul s, bu o
es he amewo k o complex sys em simula ion, a non-con as ed in o ma ion
has been used o simula ing he human lows. The in o ma ion we ha e designed
behind he human lows going shopping is exp essed in he lis below:
• F om 9.30 o 10.00, a ound 60-70 wo ke s a i e o he Shopping Cen e,
lea ing i be ween 16:00 and 16:50.
• F om 15.30 o 16.00, a ound 80-90 wo ke s a i e o he Shopping Cen e,
lea ing i be ween 22:00 and 22:50.
• F om 10.00 o 16.00, a ound 1000-2000 cus ome s will a i e o he Shop-
ping Cen e, lea ing i be ween 20 and 180 minu es la e .
• F om 16.00 o 22.00 a ound 2000-4000 cus ome s will a i e o he Shopping
Cen e, lea ing i be ween 20 and 180 minu es la e . I such lea ing ime
eaches he 22.00 limi , he lea ing ime will be 22.00
The nex sub-sec ions ocus on he s eps o de eloping a simula o om sc a ch
o unning his kind o expe imen . The i s s ep is he me amodel de elopmen
whe e he elemen s ha a e going o be used in he simula ion models a e desc ibed.
La e on, he beha iou s which add ess he way o ac ing o each model elemen
a e de eloped. Once he me amodel and beha iou s a e implemen ed, he P o ile
ool is used o gene a ing a huge scena io which con ains many agen s and wo
shopping cen es. Once he model is eady, he simula ion is execu ed and he
esul s coming om i a e analysed.
1People ha pa icipa ed in his expe imen : Jos´
e´
E o a, Jos´
e Juan He n´
andez and Ma io
He n´
andez.
124
11.1 Modelling in Ta a
Figu e 11.1: Case s udy’s me amodel.
11.1.1 Me amodel de elopmen
Fo his expe imen , he me amodel de elopmen is simple since i is only com-
posed by h ee elemen s. No e ha he me amodel design is one o he mos c i ical
p ocesses since i s good design will allow lexibili y o adding o modi ying ele-
men s wi hou a ec ing he whole s uc u e. In igu e 11.1, he me amodel o his
complex sys em is p esen ed.
F om he scene pa , he e is only one elemen : he Shopping cen e; he opo-
logy is emp y in his case and he popula ion has wo elemen s: selle and buye
agen s. F om he poin o iew o he expe imen con ex , bo h do he same: ge
in and ge ou om he Shopping cen e. Howe e , hey a e sepa a ed in o de o
p o ide concep ual cla i y. When unning expe imen s, i is impo an o keep he
concep s which exis in eali y in he me amodel. The nex XML codes ep esen
he desc ip ion o e e y me amodel elemen :
<class name=” Sho ppingCen e ” p a e n =” L oc a io n ”>
< ea u e name=” a d d e s s ” y pe =” s i n g ” />
< a i a b l e name=” pe sonCoun ” ype =” i n ” i n i i a l − a lue =”0” />
</class>
Lis ing 11.1: Desc ip ion o he Shopping Cen e me amodel class. This desc ip ion
con ains he add ess whe e he Shopping Cen e is loca ed. Fu he mo e, i con ains a
a iable which wo ks as a coun e o he numbe o people inside.
<class name=” S e l l e ” p a e n =” Bus ine ss ”>
< a i a b l e name=” s h o p p i n g C e n e I d ” yp e =” s i n g ” e q u i e d =” u e ”
>Id o he shopping c e n e whe e h e s e l l e wo ks </
a i a b l e>
<con ex name=” s ho p pi ngC en e ” yp e =” ShoppingCen e ”>Shopping
Ce n e whe e h e s e l l e wo ks </con ex >
125
11. EXPERIMENTATION CONSIDERATIONS
</class>
Lis ing 11.2: Desc ip ion o he Selle agen me amodel class. When he Selle is
ins an ia ed in he simula ion model, he shopping cen e id is equi ed in o de o
ela e he agen o whe e i wo ks. The ini /con igu e code will look o his id in o de
o se in he con ex a ibu e he shopping cen e p o ided.
<class name=” Buye ” p a e n =” B us in es s ”>
<con ex name=” s h o p p i n g C e n e L i s ” yp e =” Sho pp in gC en e ”
eplica ed=” ue”>A l i s c o n a i n i n g he Shopping C en e s
whe e he buye u s u a l l y go </con ex >
</class>
Lis ing 11.3: Desc ip ion o he Buye agen me amodel class. In his case, he buye
con ex is a lis o shopping cen es whe e he buye usually goes. Tha ’s he eason
why an id is no necessa y o be p o ided since his con ex lis will be illed by a
disco e y p ocess whe e he shopping cen es will be associa ed o his agen ollowing
a c i e ia (e.g. p oximi y, p ices, e c)
11.1.2 Simula ion de elopmen
The simula ion de elopmen is guided by he p esen ed Simula ion Li e Cycle
which conce ned ou main s eps. A his momen , all he ideas abou how he
expe imen should be de eloped mus be clea in o de o decide how o p oceed
wi h he da a, which elemen s a e necessa y o model, e c.
11.1.2.1 Da a p ocessing
Since he p oposed expe imen s in ol e he simula ion o many elemen s, he use
o a ool o au oma ing and gene a ing simula ion models is necessa y. In his case,
he P o ile is ed by a da a s o e whe e he s a is ical in o ma ion conce ning he
human lows on he shopping cen e is s o ed. Based on his in o ma ion, a model
ha ep esen s a conc e e scena io o he people lows is gene a ed. A his poin ,
and h ough he use o he P o ile ool, he scena io can be modi ied as needed
acco ding o he expe imen s design in o de o es how he people lows would
eac conce ning a iables a ia ion.
The able in o which he hypo heses we e ansla ed con ains he in o ma ion
abou he popula ion o he simula ion model (Table: 11.1). Ge ing in o he de ails
126
19.2 Case s udy
Table 19.1: Vehicles
B and Model Capaci y Range
Mega e-Ci y 9kWh 100km
Re a L-Ion 11kWh 120km
Think Ci y 25kWh 200km
Mi subishi i-Mie 16kWh 130km
Ci oen C-Ze o 16kWh 130km
Renaul Fluence-ZE 22kWh 160km
Nissan Lea 24kWh 160km
Tesla Roads e 42 42kWh 257km
Tesla Roads e 70 70kWh 483km
19.2 Case s udy
This sec ion is de o ed o explain how models ha e been de eloped. The EVs
ha e been modelled o ep esen any EV in he ma ke . I is easily pa ame e ised
by simply p o iding he ba e y capaci y (in kWh), he au onomy (in km) and he
s a ing s a e o cha ge ( om 0% o 100%). In his simula ion, ehicles in able
19.1 ha e been aken in o accoun .
These ehicles a e discha ged based on he mileage hey co e . This means
ha he ene gy ha has o be discha ged om he ba e y is calcula ed based on
he kilome es ha ha e been d i en and he au onomy o he ehicle. Vehicles a e
cha ged acco ding o he s anda d which is a 3,700 wa s.
This elec ical ehicle model needs o be commanded by an agen model in
o de o pe o m he ips (e.g. going o wo k). The e o e, an agen has been
de eloped o implemen he beha iou o he d i e who needs anspo a ion. This
agen has been implemen ed in acco dance o o dina y li es yles, since i was no
possible o ob ain in o ma ion o usage pa e ns o ehicles. These agen s execu e
ou ac i i ies which conce n he use o he ehicle: going o wo k, e u ning om
wo k, going o shops, e u ning om shops. An example o how hese agen s a e
scheduled is p esen ed below:
• Ac i i y 1. Id: go wo k. Time: 7:00. De ia ion: 500s. Pa e n: Monday,
Tuesday, Wednesday, Thu sday, F iday.
229

19. AGENT-BASED MODELLING FOR DESIGNING AN EV CHARGING
DISTRIBUTION SYSTEMS: A CASE STUDY IN SALVADOR OF BAHIA
• Ac i i y 2. Id: back om wo k. Time: 15:00. De ia ion: 500s. Pa e n:
Monday, Tuesday, Wednesday, Thu sday, F iday.
• Ac i i y 3. Id: go shop. Time: 18:00. De ia ion: 500s. Pa e n: Tuesday,
Thu sday, Sunday.
• Ac i i y 4. Id: back om shop. Time: 21:00. De ia ion: 500s. Pa e n:
Tuesday, Thu sday, Sunday.
In addi ion o his in o ma ion, agen s a e also p o ided wi h he dis ance o
wo k and o he shops. These dis ances a e calcula ed based on a no mal dis ibu-
ion cen ed in he a e age mileage d i en by people in Sal ado de Bahia. A e
each ac i i y, he agen s will plug in he EVs so ha hey can s a cha ging hei
ba e ies. Howe e , depending on he s a egy, he connec ion o he ehicles o he
g id will no necessa ily in ol e i s cha ging, as his will be decided by he s a egy.
This means ha in he simula ion in which he e is no a s a egy, EVs will s a o
cha ge as soon as hey a e plugged in. In he simula ion in which he e is a s a egy,
EV cha ging will s a in acco dance o he policies o he s a egy.
19.3 Resul s
Keeping he same dis ibu ion in as uc u es, he cha ging po en ial o he subs a-
ion o elec ical ehicles can be de ined as he di e ence be ween he maximum
deli e able powe o ehicles connec ed o cha ging s a ions and he cu en en-
e gy demand: Pp( ) = Pm( )−Pd( ). Ob iously, his po en ial is no he same
in all powe g id subs a ions, since bo h ac o s, Pm and Pd, may a y om one
subs a ion o ano he as well as seasonally.
In he p oposed cha ging model ha his pape explo es, one o mo e subs a ion
eede s will be enabled o allow he managemen o all ehicles based on hese
wo ac o s. Ideally, he ene gy used o hei cha ging should be he cheapes .
Mo eo e , sys em ope a ion should be acili a ed so ha gene a o s a e swi ched on
and o as li le as possible. To his end, he objec i e should pu sue he la ening
o he demand cu e.
230
19.3 Resul s
0
2
4
6
8
10
12
14
16
18
0:00 2:00 4:00 6:00 8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00
MW
Figu e 19.1: EVs consump ion when hey s a cha ging as soon as hey a e plugged
19.3.1 No DSM policy
In his expe imen , he model p e iously p esen ed will be simula ed conside ing
ha EVs s a cha ging as soon as hey a e plugged in. In he igu e 19.1, he
consump ion o EVs o a weekday is p esen ed. The highes peak o his load
cu e can be ound a 7pm app oxima ely. This peak is due o he ac ha almos
e e yone is a home a his ime, wi h hei EV plugged in, wi h he esul ha ha
hei cha ging eaches he g ea es le el o o e lapping. This o e lapping e ec is
also high in he mo ning and a e noon, coinciding wi h commu e “ ush hou s”.
In he igu e 19.2, wo load cu es a e p esen ed: he o iginal load cu e o
2014 (g ey) and o 2030 in which EVs a e included (black). In his cha , i can be
seen he inc emen ha in ol es he EVs in he o al consump ion. As a emainde ,
his signi ican impac is caused by 5,253 EVs in a powe g id se ing an a ea o
142,000 (in 2014) and 149,000 (p edic ed o 2030) inhabi an s. I any o he h ee
p edic ions ha a e used in his s udy inc eases (popula ion, ehicle owne ship a e
o EVs pene a ion a e), would esul in he consump ion inc emen being highe .
I can be obse ed ha he impac o in oducing 5,253 EVs can each 8 MW,
which is signi ican conside ing ha he maximum consump ion o his a ea o
he analysed day is 103 MW. This app oxima ely means an 8% inc emen . Along
wi h his inc emen , we mus also conside how he es o he ene gy demand
no p o oked by he EVs in oduc ion has inc eased as well. This leads o a o al
inc emen o app oxima ely 15MW. This would mean ha his subs a ion would
ha e o be e-sized.
19.3.2 RTP-based DSM policy
Knowing he alley exis en in he ea ly mo ning, RTP policy has been p og ammed
o ou pu a e y cheap p ice o he ene gy ha is consumed om 1am o 8am. In
hese expe imen s, all ehicles ha e been uned o only accep his p ice. In he
231
19. AGENT-BASED MODELLING FOR DESIGNING AN EV CHARGING
DISTRIBUTION SYSTEMS: A CASE STUDY IN SALVADOR OF BAHIA
0
20
40
60
80
100
120
0:00 2:00 4:00 6:00 8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00
MW
Figu e 19.2: Compa ison o he o iginal load cu e o 2014 (g ey) wi h he o al
consump ion including EVs cha ging o 2030 (black)
0
2
4
6
8
10
12
14
16
18
0:00 2:00 4:00 6:00 8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00
MW
Figu e 19.3: EVs consump ion acco ding o he RTP based policy which makes
cheape cha ging om 1am o 8am
igu e 19.3, he consump ion o he EVs is p esen ed unde his new policy. As i
can be obse ed, since many ehicles ha e no been cha ged du ing he p e ious
days, mos o hem s a cha ging hei ba e ies a 1am, hus p oducing a high
demand peak o 16MW. This consump ion becomes 0 a 7am app oxima ely.
In he igu e 19.4, again, wo load cu es a e p esen ed: he o iginal load cu e
o 2014 (g ey) and o 2030 in which EVs a e included (black). This ime, he peak
ha al eady exis ed in he load cu e o 2014 has only inc eased as consequence
o he popula ion inc ease o 2030. This ime, EV cha ging has been shi ed o an
o -peak pe iod which is in he ea ly mo ning. In his new in e al, EVs cha ging
does no inc ease a all he o e all maximum peak o he demand allowing o a
g ea e in oduc ion o EVs i i was necessa y. No wi hs anding, he e is an ab up
inc ease o he consump ion a 1am since almos all EVs s a cha ging a ha ime
due o he low p ices. This is a non-desi ed e ec o he g id s abili y and should
be u he s udied in o de o smoo h hei cha ging s a ing p ocess.
232
19.4 Discussion
0
20
40
60
80
100
120
0:00 2:00 4:00 6:00 8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00
MW
Figu e 19.4: Compa ison be ween o iginal load cu e o 2014 (g ey) wi h he o al
consump ion including EVs cha ging (black) using he RTP-based policy o 2030
19.4 Discussion
This wo k has been o ien ed o de elop he model o Sal ado o Bahia’s powe
g id using an Agen -based modelling app oach. This wo k has allowed o alida e
he abili y o his modelling app oach o:
1. Model indi idual decision making.
2. Conside local cons ain s.
3. Model powe g ids as social echnical sys em.
4. S udy eme gen synch onisa ion and coupling e ec s when many agen s co-
incide.
The majo ad an age is ha his model is able o simula e human decisions
and ac ions ha would a ec he unc ioning o he powe g id. In addi ion, such
changes in he g id would in u n in luence human decisions and ac ions. Once
his model is de eloped, u u e p ojec s will be o ien ed o design and assess DSM
policies ha would help educe in es men s on g id in as uc u e. On he o he
hand, he agen decision model could also be imp o ed. Fo example, agen s could
make decisions based on esponses o changes in he sys em, which will in u n
change he con ex o u u e decisions; o agen s could beha e in a he e ogeneous
way, hus maximising a ce ain p o i , ei he a ull cha ging o he ehicle o sa e
money.
233

Pa V
Conclusions
235
CHAPTER
20
Resul s
Resea ch wo k has been conduc ed o deal wi h he p oblems o designing and
e alua ing managemen policies ela ed o he SG pa adigm. These policies a e
aimed a di e en s akeholde s o bene i hei in e es s. Fo ins ance, inal cus-
ome ’s in e es s is he educ ion o he elec ici y cos s and he imp o emen o
he ene gy e iciency. The in e es s o ope a o s and e aile s is o imp o e hei
p o i by educing ope a ion and p oduc ion cos s. Globally, people wan o ha e a
mo e e icien powe g id in which CO2emissions a e educed wi h a highe RES
pene a ion.
The mo i a ion o conduc ing esea ch in he ield o Powe G ids is due o i s
p oposed e olu ion o SGs. In his e olu ion, IT seems o ha e an impo an ole.
This e olu ion is p oposed because he e is a signi ican need o he imp o emen
o powe g ids, such as: educ ion o he dependency on ossil- uel based ene gy
p oduc ion; ma ke conce ns such as uel p ices ola ili y; and educ ion o GHG
emissions.
SGs aim o ace he nex challenges: highe in oduc ion o RES; mo e e i-
ciency and lexibili y; and less dependency on ossil uels. Mo eo e , hey a e
cha ac e ised by he in oduc ion o au oma ion a dis ibu ion le el. I is expec-
ed ha his e olu ion will o e come he challenge o he massi e in oduc ion o
elec ical ehicles, dis ibu ed gene a ion and RES [HHM11]. The SG pa adigm
237
20. RESULTS
is an e en mo e impo an challenge o isola ed powe g ids as hey ha e limi ed
esou ces on he p oduc ion side o balance o e and demand.
This is he case o he Cana y Islands. In his e i o y, he e a e excellen
en i onmen al condi ions o inc easing he exploi a ion o RES. The Cana y Is-
lands ha e 2,500-3,000 hou s/yea o sola adia ion p oducing an a e age o 5-
6kWh/m2pe day (sou ce: Cana y Ins i u e o Technology). Fu he mo e, he e
a e 3,000-4,500 hou s/yea o wind wi h a speed a e age o 7-8m/s. This p oduces
625MWh pe day ha ing ins alled 75MW o wind ene gy. In his sense, he esul s
o his esea ch can be di ec ly applied o imp o ing powe g ids in he Cana y
Islands.
The s udy o hese sma managemen policies equi es models ep esen ing
he sys em a he lowes le el o de ail allowing o analyse side-e ec s. Agg ega ed
models a e no able o ep esen e en s on he demand side ha should be con-
side ed in designing hese policies. The e o e, Powe G id models should be dis-
agg ega ed o ep esen all loads ha can be managed. In hese models, complexi y
a ises since he e a e many he e ogeneous componen s which a e in e ac ing a di -
e en scales.
The eme gen beha iou o he sys em canno be in e ed wi h adi ional mod-
els, because complexi y canno be ep esen ed. SGs canno be s udied using adi-
ional app oaches since hese a e no able o cap u e all o he sys em’s p ope ies.
The bes way o s udy complex models is h ough So wa e ools. Howe e , i SGs
a e no suppo ed by he equi ed o malisms and me hodologies, so wa e ech-
nologies by hemsel es a e no su icien o pe o m hese s udies. This means,
he p oblem is no choosing he so wa e ool, bu applying he igh o malisms
and me hodologies. Fo hese easons, se e al au ho s ha e sugges ed he use o
complex sys em based o malisms o ep esen SGs. A Complex Sys ems app oach
acili a es he ep esen a ion o SGs a he lowes le el o de ail.
Unde he complex sys em app oach, only so wa e ools ha suppo his o m-
alism can be conside ed o s udy SGs. Ne e heless, he e a e s uc u al conce ns
ha mus be aken in o accoun o suppo he enginee ing o la ge-scale complex
sys ems. In he case o SGs, models may ha e millions o componen s. To model
hem, se e al complex sys em simula ion so wa e ools ha e been e iewed in his
documen . In all hese ools, a lack o a me hodological o ien a ion o ace he
modelling o la ge-scale complex sys ems was ound.
238
20.7 Publica ions
• Towa ds an in e disciplina y app oach o he simula ion o u u e SG a chi-
ec u es om a complex sys ems science poin o iew
–Au ho s: Viejo, P., K eme s, E., He n´
andez, M., He n´
andez, J., ´
E o a,
J., Langlois, P., Daude, E., Gonz´
alez de Du ana, J.M., & Ba ambones,
O.
–Cong ess: Eu opean Con e ence on Complex Sys ems 2011
–Place: Vienna, Aus ia
–Da e: Sep embe , 2011
• A la ge-scale elec ical g id simula ion o massi e in eg a ion o dis ibu ed
pho o ol aic ene gy sou ces
–Au ho s: ´
E o a, J., K eme s, E., He n´
andez, M., & He n´
andez, J. J.
–Cong ess: 6 h Eu opean Con e ence on PV-Hyb ids and Mini-G ids
(OTTI)
–Place: Chambe y, F ance
–Da e: Ap il, 2012
• Modelling li es yle aspec s in luencing he esiden ial load-cu e
–Au ho s: Hause , W., ´
E o a, J., & K eme s, E.
–Cong ess: 26 h Eu opean Con e ence on Modelling and simula ion
(ECMS 2012)
–Place: Koblenz, Ge many
–Da e: May, 2012
• Asynch onous Sma G id Simula ions
–Au ho s: ´
E o a, J., He n´
andez, J. J., & He n´
andez, M.
–Cong ess: UCNC’12 (Uncon en ional Compu a ion and Na u al Com-
pu a ion): CoSMoS - P oceedings o he 2012 Wo kshop on Complex
Sys ems Modelling and Simula ion
–Place: O leans, F ance
–Da e: Sep embe , 2012
• Decision suppo o Complex Sys ems: a Sma G id case
–Au ho s: ´
E o a, J., He n´
andez, J. J., & He n´
andez, M.
245

20. RESULTS
–Cong ess: UCNC’13 (Uncon en ional Compu a ion and Na u al Com-
pu a ion): CoSMoS - P oceedings o he 2013 Wo kshop on Complex
Sys ems Modelling and Simula ion
–Place: Milan, I aly
–Da e: July, 2013
• C i icali y in complex socio echnical sys ems: an empi ical app oach
–Au ho s: Viejo, P., K eme s, E., ´
E o a, J., He n´
andez, J. J., He n´
andez,
M., Ba ambones, O., & Gonz´
alez de Du ana, J. M.
–Cong ess: Eu opean Con e ence on Complex Sys ems 2013 (ECCS’13)
–Place: Ba celona, Spain
–Da e: Sep embe , 2013
• Asynch onous app oach o simula ions in Sma G id
–Au ho s: ´
E o a, J., He n´
andez, J. J., & He n´
andez, M.
–Cong ess: Eu opean Simula ion and Modelling Con e ence 2013 (ESM’13)
–Place: Lancas e , England
–Da e: Oc obe , 2013
• Ta a : A amewo k o de eloping simula o s based on Model D i en En-
ginee ing
–Au ho s: ´
E o a, J., He n´
andez, J. J., & He n´
andez, M.
–Cong ess: Eu opean Simula ion and Modelling Con e ence 2013 (ESM’13)
–Place: Lancas e , England
–Da e: Oc obe , 2013
• Model D i en Enginee ing o da a mine s simula ion
–Au ho s: ´
E o a, J., He n´
andez, J. J., & He n´
andez, M.
–Cong ess: IEEE In e na ional Con e ence on Da a Mining 2013 (ICDM’13)
- In e na ional Wo kshop on Domain D i en Da a Mining
–Place: Dallas, Uni ed S a es o Ame ica
–Da e: Decembe , 2013
• Agen -based modelling o designing an EV cha ging dis ibu ion sys ems: a
case s udy in Sal ado o Bahia
–Au ho s: ´
E o a, J., He n´
andez, J. J., Ba bosa, D. & Naza eno, P.
246
20.7 Publica ions
–Cong ess: Eu opean Simula ion and Modelling Con e ence 2014 (ESM’14)
–Place: Po o, Po ugal
–Da e: Oc obe , 2014
Nex lis con ains he publica ions eleased in jou nals:
• Ad an ages o Model D i en Enginee ing o s udying complex sys ems
–Au ho s: ´
E o a, J., He n´
andez, J. J., & He n´
andez, M.
–Jou nal: Na u al Compu ing
–DOI: 10.1007/s11047-014-9469-y
• C i icali y in complex socio echnical sys ems: an empi ical app oach o elec-
ical g ids 1
–Au ho s: Viejo, P., K eme s, E., ´
E o a, J., He n´
andez, J.J., He n´
andez,
M., Ba ambones, O, Gonz´
alez de Du ana, J.M.
–Jou nal: Applied ene gy
1This pape has been submi ed and now i is awai ing o he accep ance
247
CHAPTER
21
Discussion
In his chap e he scien i ic and echnical con ibu ions o his esea ch a e
desc ibed. “Scien i ic con ibu ions” e e o he c ea ion o knowledge whe eas
echnical con ibu ions e e o he c ea ion o new p ocesses o deal wi h speci ic
p oblems.
The pu pose o his esea ch is o explo e and desc ibe he applica ion o al eady
exis ing me hodologies o iden i y p oblems in SGs. This explo a ion has also
allowed an ou look on new esea ch a ge s ha could be s udied in u u e esea ch.
21.1 Con ibu ions
21.1.1 On Applying he complex sys em app oach in Sma G ids
Acco ding o he well-honou ed philosophe o science Thomas Kuhn, in he sci-
en i ic de elopmen o a discipline he e a e h ee main s ages [Kuh12]. These
h ee main s ages a e: “p escience”, “no mal science” and “ e olu iona y science”
(Figu e: 21.1). The “p escience” s age is cha ac e ised by ha ing nume ous incom-
pa ible and incomple e heo ies. The consensus o a p escien i ic communi y in
e ms o me hods, e minologies, expe imen s, e c. leads o he “no mal science”.
In his s age, new pa adigms can be concei ed o deal wi h anomalies eaching he
“ e olu iona y science” s age.
249
21. DISCUSSION
Figu e 21.1: Kuhn’s cycle.
In he case o he SG opic, i can be conside ed o be cu en ly in he second
s age: “no mal science”. Wi hin he s udy o his no mal science, se e al lim-
i a ions ha e been ound when SGs a e being modelled (appea ance o anom-
alies). The complex sys em app oach has been p oposed o o e come hem (new
pa adigms).
The e a e se e al s udies in he documen a ion ha asse ha powe g ids can-
no be s udied using adi ional o malisms. In hese s udies, he hypo hesis s a es
ha complex sys em app oach could o e come he limi a ion o he adi ional
o malisms. This hypo hesis canno be absolu ely e i ied since i canno be p o ed
o ca y ou all possible case s udies. Howe e , wha he hypo hesis s a es can be
ep oduced and e u ed.
In his esea ch, his hypo hesis has been ep oduced h ough he expe imen a-
ion o case s udies. These case s udies p o ide addi ional e idence ha his o m-
alism is alid [CGLP94, PKZK08, EKM+11]. Se e al case s udies ha e been ca -
ied ou o analyse SGs ollowing complex sys em app oach. I has been ound ha
his o malism is use ul o s udying SG policies wi h a maximum le el o disag-
g ega ion. When a new policy is being p oposed, he beha iou o a powe g id is
no known. In hese cases, he sys em can be modelled h ough i s componen s’
beha iou s and simula ed o ob ain he eme gen beha iou . A maximum le el o
disagg ega ion also allows o e-agg ega e he componen s’ beha iou s and analyse
250

21.1 Con ibu ions
hem a di e en le els (e.g. appliance, household, dis ic , coun y le el).
Ne e heless, in his esea ch, i has been ecognised ha complex sys em ap-
p oach by i sel may no be enough o deal wi h he s udy o SGs. O he issues
appea when using his o malism: complexi y o modelling sys ems, complexi y
o analysing da a and complexi y o designing s a egies. An impo an con ibu-
ion is he iden i ica ion o complemen a y me hodologies o o malisms ha could
be use ul o o e coming hese issues. In pa icula , he applica ion o MDE, BI
and SI should be conside ed.
The alidi y o his con ibu ion has been demons a ed h ough he execu ion o
case s udies. As in he case o complex sys em app oach, hese hypo heses canno
be e i ied. In o de o e i y hem absolu ely, hese hypo heses should be applied
o pe o m all SG case s udies, which is impossible. Ne e heless, in his esea ch,
se e al case s udies ha e been ca ied ou o p o ide e idence o i s alidi y.
21.1.2 On Modelling Sma G ids
This esea ch has ound ha he modelling o SGs equi es he coexis ence o di -
e en na u es in hei componen s. When SGs a e being modelled, i is necessa y
o desc ibe componen s ha may ha e di e en na u es such as: elec ical, he mo-
dynamical, me eo ological o sociological, among o he s.
Ob iously, elec ical componen s mus be ep esen ed since hei beha iou is
closely ela ed o he human beha iou s (e.g. when hese componen s a e swi ched
on and o ). Ac ually, he way in which he humans use hese elec ical compon-
en s will de e mine he sys em’s eme gen beha iou . The in e ac ion hey do o e
his componen s in ol e consump ions ha a ec how he sys em wo ks. Thus,
many elec ical models mus be modelled in o de o ep esen p oduc ion, dis i-
bu ion, demand, e c. wi h he deg ee o de ail ha is necessa y o es ing he SG
implemen a ion.
In addi ion, he modynamical componen s mus also be modelled since i is
necessa y o ep esen he he mal ans e ences be ween he mal elec ical com-
ponen s (e.g. e ige a o s, domes ic wa e hea e s, e c.). The modelling o hese
elec ical componen s mus include he modynamical beha iou ha calcula es he
he mal gains de i a i e om hei consump ions. Mo eo e , in as uc u e as
households may include a he mal beha iou as well. This beha iou would cal-
251
21. DISCUSSION
cula e he in as uc u e empe a u e based on he ex e nal empe a u e and he
he mal gains de i a i e om he he mal elec ical loads. In his manne , he mal
elec ical loads can calcula e hei in e nal empe a u e based on hei consump ion
and he household empe a u e.
The calcula ion o he in e nal empe a u e o a household depends on he ex-
e nal empe a u e, as p e iously s a ed. The e o e, i is necessa y o ep esen
he me eo ological condi ions o he en i onmen whe e he household is loca ed.
Thus, ano he kind o beha iou which is necessa y o model powe g ids is iden i-
ied. Fu he mo e, me eo ological condi ions mus be ep esen ed i he e a e RES
in he powe g id unde s udy. Renewable ene gy echnologies, such as PV cells
and wind u bines, equi e he ep esen a ion o en i onmen al condi ions. These
condi ions a e used by he models o hese ypes o echnologies o calcula e he
ene gy p oduced.
Fu he mo e, he ep esen a ion o SGs also use he modelling o human in e -
ac ions. Human beings a e use s o Powe G ids and he way hey use ene gy mus
be ep esen ed. This is a e y impo an conside a ion because o he signi ican
impac human in e ac ion has on he powe g id. As men ioned p e iously, in s a e
o he a , powe demand a ies geog aphically, seasonally, cul u ally, sociologic-
ally, e c. All hese a ia ions ha e o do wi h he way in which humans in e ac
wi h he powe g id.
In conclusion, he s udy o he applica ion o SG policies equi es he modelling
o many componen s wi h he e ogeneous na u es. The beha iou s ha ha e been
p e iously men ioned a e no only di e en because hey ha e di e en na u es:
elec ical, he mal, social, e c. bu also because hey equi e di e en execu ion
pa adigms: con inuous (e.g. he mal) o disc e e (e.g. washing machine).
21.1.3 On De eloping Complex Models
In his esea ch, MDE has been ound o be a complemen a y me hodology o
modelling SGs wi h a complex sys em app oach. When i is necessa y o in eg a e
di e en modelling echniques, s uc u al guidelines a e needed o deal wi h his
complexi y. In hese cases, MDE is use ul o managing he model de ini ion and
he simula o cons uc ion.
In his esea ch, he applica ion o MDE has been alida ed by ca ying ou he
252
21.1 Con ibu ions
case s udies. Mo eo e , his app oach is being used by o he esea che s. Cu -
en ly, EIFER and EDF a e de eloping hei own case s udies wi h his me hod-
ology. F om a scien i ic poin o iew, i is known ha o he labo a o ies a e e-
sea ching in he hypo hesis. To his day, he hypo hesis has no been e u ed.
The mos impo an implica ion o applying MDE in his ield is echnical:
MDE helps o build up and main ain la ge-scale simula ion models. Modelle s a e
assis ed in he cons uc ion p ocess by sepa a ing conce ns: he scena io desc ip-
ion (simula ion model) and he desc ip ion o indi idual componen s (beha iou s).
In his way, a se o s eps a e es ablished o guide he modelle o he goal:
• De eloping he simula ion model.
• Desc ibing a componen in he me amodel i i has no been p e iously de ined
he e.
• De eloping necessa y beha iou s no ye modelled.
• Building a simula o based on his simula ion model and simula e i .
Ano he MDE implica ion is i s adap abili y o changes. A single componen
o beha iou can be easily eplaced o added o an exis ing model scena io. Flex-
ibili y is an impo an ea u e o enginee ing SGs. In he design o a new policy,
i is necessa y o es i in di e en scena ios. The lexibili y p o ided by MDE
makes his possible. These scena ios may no only use di e en beha iou s o he
componen s bu may also o ganise componen s in a di e en way. In addi ion, he
same scena io can be used o es ing di e en policies.
Since MDE equi es he de ini ion o a me amodel, a echnical con ibu ion o
his esea ch is he de elopmen o a me amodel o SGs. In his sense, a me amodel
can be unde s ood as he c ea ion o an ag eemen con ac be ween many model-
le s in o de o make collabo a ion easie . The me amodel has se e al impo an
implica ions:
• Modelle s can sha e hei models whene e hey ha e been buil unde he
same me amodel.
• Modelle s can con ibu e o he me amodel by ex ending i s de ini ion o c e-
a ing new beha iou s ha can be used by o he s.
• Modelle s do no need o desc ibe he componen s o he eali y e e y ime
hey s a a new simula ion since hey ake ad an age o he al eady de ined
me amodel.
253
21. DISCUSSION
• Modelle s can use p e ious wo k de eloped by o he modelle s.
• Modelle s can de elop hei models as e hanks o all o hese abo e-men ioned
ac o s.
In his way, la ge simula ion models can be buil by de eloping he simula ion
model; and, in case i is needed, de eloping new componen s and beha iou s. In
his sense, modelle s a e guided comple ely so hey know wha hey ha e o do o
de elop simula ion models. The e o e, he complexi y o he scena io only a ec s
he simula ion model and no he o e all amewo k.
Fu he mo e, ools ha e been de eloped o deal wi h he complexi y o he ep-
esen a ion o la ge-scale sys em in his simula ion model. The ool named P o ile
was a esponse o he need o many modelle s had o he cons uc ion o la ge-
scale scena ios. In his sense, his ool handles he complexi y o building hese
simula ion models. Based on inpu da a p o ided by he modelle , his ool c e-
a es he scena io. Ano he impo an ool is he Simula o Builde which builds a
simula o based on a model ha desc ibes he scena io.
21.1.4 On Analysing Simula ion Resul s
This esea ch demons a es ha he applica ion o BI me hodologies is help ul.
These me hodologies ha e been used o analyse da a coming om he simula ion
o a complex sys em.
When SGs a e s udied as complex sys ems, he huge quan i y o esul s coming
om hese simula ions can be managed using BI me hodologies. BI me hodologies
can be used o disco e laws and egula i ies ha p o ide a be e knowledge o he
eme gen beha iou in SGs. These me hodologies allow science o be done o e
simula ed expe imen s. Based on he ou pu a ailable om he simula ions, i is
possible o ind ules, laws, egula i ies, e c. au oma ically.
In his esea ch, BI me hodologies a e used on he ou pu o he case s udies
p esen ed. The use o hese me hodologies is eally impo an in o de o know
how he eme gen beha iou in SGs is eached; and how i can be in luenced.
A e pe o ming all o he case s udies p esen ed in his documen , i has been
shown ha BI can be used in his con ex wi h na u alness. The mechanisms
p o ided by BI a e su icien o ep esen da a coming om SG simula ions and
allow i s exploi a ion.
254
21.3 Fu u e wo k
a e p og essi ely in oduced; o RES quo a inc eases. Fu he mo e, case s udies in
which bo h scena ios and policies can e ol e can be conside ed.
These kind o scena ios may be impo an o es ing he applicabili y o SG
policies in he long- e m. In o de o suppo hem, new mechanisms mus be de-
signed wi hin his ecosys em o me hodologies and o malisms.
21.3.3 On Disco e ing Pa e ns
As said be o e, BI me hodologies can be used o ind pa e ns in he simula ions o
SGs. The disco e y o pa e ns may ha e many di e en usages: ob aining a be e
knowledge o how eme gence is eached, simpli ying a complex model, eeding
back in o he design o SG policies, e c.
In his ield, much wo k can be ca ied ou as a con inua ion o his esea ch.
Da a mining echniques can be explo ed in o de o au oma ically de ec pa e ns in
se ies o da a. Howe e , his is no an easy ask, as he e is a main unde lying con-
ce n ha mus be conside ed when explo ing hese echniques: he huge quan i y
o da a.
As s a ed in he hypo heses pa o he documen , an impo an conce n o da a
managemen is he eloci y. Huge quan i y o da a is con adic o y o he eloci y.
This is especially icky when dealing wi h da a mining echniques ha equi e
a ho ough analysis o he da a. Fo his eason, his esea ch should consis in:
in es iga ing da a mining p ocedu es o ind pa e ns; and inding p ocedu es o
accele a e he p ocess.
One o he applica ions ha his esea ch may ha e is he simpli ica ion o com-
plex models. This would open ano he in e es ing u u e p ojec ha could consis
in au oma ing he model simpli ica ion p ocess. Tha means, he use o me hodolo-
gies, o malisms o echniques o iden i y beha iou al ea u es in complex models
ha can cha ac e ise simpli ied models.
To his end, esul s coming om he disco e y o pa e ns can be used in his
esea ch. A e , so wa e ools could be able o pa se da a coming om complex
models in o de o c ea e simpli ied models. The beha iou s p o ided by he sim-
pli ied model mus be ce i ied o be wo king a ce ain in e als ha ensu e meas-
u able e o s. Fo ins ance, hese simpli ied models could be used as submodels
wi hin bigge simula ions. The e o e, smalle complex simula ions can be used o
261

21. DISCUSSION
enginee bigge ones in which he compu a ional cos s can be educed.
21.3.4 On O he Fields
In he “hypo heses pa ” o his documen , he de ini ion o ex e nal alidi y in he
con ex o his esea ch was discussed. Howe e , a hypo hesis has ex e nal alidi y
i i applies o se e al case s udies o SG policies. This ex e nal alidi y has been
checked in his documen as hese me hodologies ha e been success ully applied
in all p esen ed case s udies.
Ne e heless, he second de ini ion p o ided o he ex e nal alidi y is no deal
wi hin his esea ch. This de ini ion conce ned he use o hese me hodologies in
o he complex sys ems apa om SGs. This is ac ually an ex ension o he scope
o he hypo heses de ined in his wo k.
I is easonable o belie e ha i hei applica ion in he ield o SGs has been
use ul, hei applica ion in o he ields can be help ul as well. Howe e , his should
be checked and is p oposed o u u e esea ch. In his sense, he hypo hesis o be
esea ched could be: “MDE, BI and SI can be help ul me hodologies o enginee -
ing complex sys ems o any na u e”. As in he p e ious hypo heses, his canno be
ully e i ied because i is no possible o pe o m all o he case s udies in all com-
plex sys ems. Howe e , a subse o hem can be ca ied ou , which could p o ide
new ideas o equi emen s.
21.3.5 On Expe imen ing wi h O e iew, Design concep s and
De ails p o ocol
Ano he line ha can be explo ed is he in eg a ion o he O e iew, Design con-
cep s and De ail (ODD) p o ocol in ou app oach o model agen -based models
wi h MDE. This p o ocol is a p oposed s anda d o desc ibing agen -based models
[GBB+06]. The e o e, ou app oach could include he desc ip ion o models ha
a e complian wi h ODD.
This p o ocol consis s o h ee blocks (O e iew, Design concep s and De ails),
which a e subdi ided in o se en elemen s: Pu pose, S a e Va iables and Scales,
P ocess O e iew and scheduling, Design concep s, Ini ializa ion, Inpu , and Sub-
models. The e o e, agen -based models can be desc ibed acco ding o hese se en
262
21.3 Fu u e wo k
elemen s. The nex lis summa ises he con en s o place a each o hese elemen s:
• Pu pose: desc ip ion o he scope o he model. Clea , concise and speci ic
o mula ion o he model’s pu pose. Explana ion o he need o building a
complex model.
• S a e a iables and scales: s a e a iables e e o low-le el a iables ha
cha ac e ise low-le el en i ies o he model. The scales e e o he simula ion
con igu a ion (e.g. leng h o ime s eps, size o habi a cells).
• P ocess O e iew and scheduling: he p ocess o e iew desc ibes en i on-
men al and indi idual p ocesses ha a e buil in o he model (e.g. ood p o-
duc ion, g ow h, mo emen ). The schedule has o do wi h how hese p o-
cesses occu and hei o de .
• Design concep s: hese concep s p o ide a common amewo k o designing
and communica ing indi idual-based models.
• Ini ializa ion: de ines how he en i onmen and he indi iduals ha e o be
c ea ed.
• Inpu : condi ions o he en i onmen ha change o e space and ime.
• Submodel: p esen a ion and explana ion o all he p ocesses lis ed abo e in
a de ailed manne , including he pa ame e isa ion o he model.
263
Bibliog aphy
[ADG08] A A enas and A Diaz-Guile a. Syncho niza ion in complex ne -
wo ks. Physics Repo s, 469:93–153, 2008. 191
[ADHM] Robe A i , Roge Dugan, Robe Hen y, and Tom McDe mo .
OpenDSS. 35, 37
[AEE08] ADEME, ENERTECH, and EDF. Campagne de mesu es des ap-
pa eils de p oduc ion de oid e des appa eils de la age dans
100 logemen s. Technical epo , ADEME and ENERTECH and
EDF, 2008. 150
[AER] ADEME, EDF, and R´
egion R´
eunion. Millene , une longueu
d’a ance pou la R´
eunion. 9
[AES07] M.H. Albadi and EF El-Saadany. Demand esponse in elec ici y
ma ke s: An o e iew. In Powe Enginee ing Socie y Gene al
Mee ing, 2007. IEEE, pages 1–5. IEEE, 2007. 30
[AG12] Hazem Ahmed and Janice Glasgow. Swa m in elligence: Con-
cep s, models and applica ions. School O Compu ing, Queens
Uni e si y Technical Repo , 2012. 47
[All09] P Allen. Complexi y, Inno a ion and O ganiza ional E olu ion.
2009. 189
[Ami01] Massoud Amin. Towa d sel -healing ene gy in as uc u e sys-
ems. Compu e Applica ions in Powe , IEEE, 14(1):20–28,
2001. 4, 19
265
BIBLIOGRAPHY
[And09] Da id L Ande son. An e alua ion o cu en and u u e cos s o
li hium-ion ba e ies o use in elec i ied ehicle powe ains.
PhD hesis, Duke Uni e si y, 2009. 5
[AP96] Timo hy Wa d A han and Panos Y Papalamb os. A no e on
weigh ed c i e ia me hods o comp omise solu ions in mul i-
objec i e op imiza ion. Enginee ing Op imiza ion, 27(2):155–
176, 1996. 118
[B+09] Albe -L´
aszl´
o Ba ab´
asi e al. Scale- ee ne wo ks: a decade and
beyond. science, 325(5939):412, 2009. 40
[BA03] Robe Blumbe g and Shaku A e. The p oblem wi h uns uc-
u ed da a. DM REVIEW, 13:42–49, 2003. 62
[BB02] Tim M Blackwell and Pe e Ben ley. Imp o ised music wi h
swa ms. In Compu a ional In elligence, P oceedings o he Wo ld
on Cong ess on, olume 2, pages 1462–1467. IEEE, 2002. 48
[BBS+09] P asanna Balap akash, Mau o Bi a a i, Thomas S ¨
u zle, Zhi
Yuan, and Ma co Do igo. Es ima ion-based an colony op im-
iza ion and local sea ch o he p obabilis ic a eling salesman
p oblem. Swa m In elligence, 3(3):223–242, 2009. 47
[BC05] C Bueno and J A Ca a. Technical–economic analysis o wind-
powe ed pumped hyd os o age sys ems. Pa II: model applica-
ion o he island o El Hie o. Sola ene gy, 78(3):396–405,
2005. 5
[BF04] A Bo shche and A Filippo . Anylogic-mul i-pa adigm simula-
ion o business, enginee ing and esea ch. In The 6 h IIE Annual
Simula ion Solu ions Con e ence, olume 150, 2004. 45
[BI04] J.P. Ba on and D.G. In ield. Ene gy s o age and i s use wi h
in e mi en enewable ene gy. Ene gy Con e sion, IEEE T ans-
ac ions on, 19(2):441–448, June 2004. 28
266

BIBLIOGRAPHY
[Blu05] Ch is ian Blum. Beam-aco-hyb idizing an colony op imiza ion
wi h beam sea ch: An applica ion o open shop scheduling. Com-
pu e s & Ope a ions Resea ch, 32(6):1565–1591, 2005. 47
[Blu08] Ch is ian Blum. Beam-aco o simple assembly line balancing.
INFORMS Jou nal on Compu ing, 20(4):618–627, 2008. 47
[BLW05] Paul Bake , Shiou Loh, and F ank Weil. Model-d i en enginee -
ing in a la ge indus ial con ex -mo o ola case s udy. In Model
D i en Enginee ing Languages and Sys ems, pages 476–491.
Sp inge , 2005. 55
[BM01] E ic Bonabeau and Ch is ophe Meye . Swa m in elligence. Ha -
a d Business Re iew, 79(5):106–114, 2001. 46
[BMW10] PE Bake , C Mi chell, and B Woodman. Elec ici y ma ke design
o a low-ca bon u u e. London, UKERC, 2010. 24
[Boh03] Ma ko Bohanec. Decision suppo . Sp inge , 2003. 57
[BPSK02] Michael Bu le , Luigia Pe e, Kaisa Se e, and S ua Ken . Model
D i en Enginee ing. In In eg a ed o mal me hods, olume 2335
o Lec u e No es in Compu e Science, pages 286–298. Sp inge
Be lin / Heidelbe g, 2002. 50, 52
[B e00] Ma ilynn B B ewe . Resea ch design and issues o alidi y.
Handbook o esea ch me hods in social and pe sonali y psycho-
logy, pages 3–16, 2000. 72
[BRS+10] S e e Bou Ghosn, P akash Rangana han, Saeed Salem, Jingpeng
Tang, Da in Loege ing, and Kendall E. Nyga d. Agen -O ien ed
Designs o a Sel Healing Sma G id. In 2010 Fi s IEEE In-
e na ional Con e ence on Sma G id Communica ions, pages
461–466. IEEE, Oc obe 2010. 36
[Bun11] Bundes e band de Ene gie- und Wasse wi scha . BDEW -
S anda dlas p o ile (SLP), 2011. 143
267
BIBLIOGRAPHY
[BVB08] Ch is ian Blum, Ma eu Y´
aba Vall`
es, and Ma ia J Blesa. An an
colony op imiza ion algo i hm o dna sequencing by hyb idiz-
a ion. Compu e s & Ope a ions Resea ch, 35(11):3620–3635,
2008. 47
[BW89] Ge a do Beni and Jing Wang. Swa m In elligence in Cellula Ro-
bo ic Sys ems, P oceed. In NATO Ad anced Wo kshop on Robo s
and Biological Sys ems, Tuscany, I aly, 1989. 45
[BW93] Ge a do Beni and Jing Wang. Swa m in elligence in cellula o-
bo ic sys ems. In Robo s and Biological Sys ems: Towa ds a New
Bionics?, pages 703–712. Sp inge , 1993. 45, 46
[BY97] Yanee Ba -Yam. Dynamics o complex sys ems, olume 213.
Addison-Wesley Reading, MA, 1997. 39
[C+94] CDIF Technical Commi ee e al. Cdi amewo k o modeling
and ex ensibili y. Elec onic Indus ies Assocaia ion, EIA/IS-107,
1994. 52
[C+06] Eu opean Commission e al. Eu opean echnology pla o m
sma g ids: ision and s a egy o eu ope’s elec ici y ne wo ks
o he u u e. O ice o O icial Publica ions o he Eu opean
Communi ies, 2006. 22
[Cao04] Junwei Cao. Sel -o ganizing agen s o g id load balancing. In
P oceedings o he 5 h IEEE/ACM In e na ional Wo kshop on
G id Compu ing, pages 388–395. IEEE Compu e Socie y, 2004.
46
[CCS93] Edga F Codd, Sha on B Codd, and Clynch T Salley. P o iding
olap (on-line analy ical p ocessing) o use -analys s: An i man-
da e. Codd and Da e, 32, 1993. 60
[CD97] Su aji Chaudhu i and Umeshwa Dayal. An O e iew o Da a
Wa ehousing and OLAP Technology. ACM Sigmod Reco d,
26(1), 1997. 103
268
BIBLIOGRAPHY
[CD10] E Chappin and G.P. Dijkema. Agen -based modelling o ene gy
in aes uc u e ansi ions. In e na ional jou nal o C i ical In-
aes uc u es, 6:106–130, 2010. 190
[CDMT94] Albe o Colo ni, Ma co Do igo, Vi o io Maniezzo, and Ma co
T ubian. An sys em o job-shop scheduling. Belgian
Jou nal o Ope a ions Resea ch, S a is ics and Compu e Sci-
ence, 34(1):39–53, 1994. 47
[CGLP94] A. Capasso, W. G a ie i, R. Lamedica, and A. P udenzi. A
bo om-up app oach o esiden ial load modeling. Powe Sys-
ems, IEEE T ansac ions on, 9(2):957–964, 1994. 8, 32, 33, 250
[CL02] C A Coello Coello and M S Lechuga. MOPSO: a p oposal o
mul iple objec i e pa icle swa m op imiza ion. In E olu iona y
Compu a ion, 2002. CEC ’02. P oceedings o he 2002 Cong ess
on, olume 2, pages 1051–1056, 2002. 119, 164, 168
[Cla11] C´
ed ic Clas es. Sma g ids: Ano he s ep owa ds compe i ion,
ene gy secu i y and clima e change objec i es. Ene gy Policy,
39(9):5399–5408, Sep embe 2011. 3, 20
[CMB10] A.J. J. Conejo, J.M. M. Mo ales, and L. Ba ingo. Real-Time
Demand Response Model. Sma G id, IEEE T ansac ions on,
1(3):236–242, 2010. 33
[CMG11] Timo hy Coope , Gillian Mead, and Phil G ¨
unewald. The Fu u e
o Ene gy S o age: S akeholde Pe spec i es and Policy Implic-
a ions. Technical epo , UK Ene gy Resea ch Cen e, Impe ial
College, 2011. 23
[CnA00] C.A. Ca˜
niza es and F.L. Al a ado. UWPFLOW P og am, Uni-
e si y o Wa e loo, 2000. 35, 36
[Coa99] Me a Da a Coali ion. Open in o ma ion model. A URL:
h p://www. mdcin o. com/OIM/MDCOIM10. pd , 1999. 52
269
BIBLIOGRAPHY
[Coe98] Ca los A. Coello Coello. A comp ehensi e su ey o
e olu iona y-based mul iobjec i e op imiza ion echniques.
Knowledge and In o ma ion Sys ems, 1:269–308, 1998. 118
[Col03] Nick Collie . Repas : An ex ensible amewo k o agen sim-
ula ion. The Uni e si y o Chicago’s Social Science Resea ch,
36:2003, 2003. 44
[CSMJ13] T inidad Cha din-Segui, Sebas ien Ma hiou, and Y es Jeanjean.
The Millene p ojec : Fi e hund ed RES & one housand i y
DSM ins alla ions o households in F ench islands, agg ega ed
o g id suppo . In 2013 IEEE G enoble Con e ence, pages 1–6.
IEEE, June 2013. 9
[Dep] Depa men o Ene gy, USA. h p://ene gy.go /oe/ echnology-
de elopmen /sma -g id. 21
[DGS07] Joyce Da gay, De mo Ga ely, and Ma in Somme . Vehicle own-
e ship and income g ow h, wo ldwide: 1960-2030. The Ene gy
Jou nal, pages 143–170, 2007. 228
[DMCM91] Ma co Do igo, Vi o io Maniezzo, Albe o Colo ni, and Vi o io
Maniezzo. Posi i e eedback as a sea ch s a egy, 1991. 47
[Do 92] Ma co Do igo. Op imiza ion, lea ning and na u al algo i hms.
Ph. D. Thesis, Poli ecnico di Milano, I aly, 1992. 47
[Dou06] Douglas C. Schmid . Gues Edi o ’s In oduc ion: Model-D i en
Enginee ing. Compu e , 39:25–31, Feb ua y 2006. 49, 50, 51
[Du 10] D Du e . Las modellie ung zu Un e suchung on Las man-
agemen - Maßnahmen in Elek izi ¨
a sne zen. PhD hesis, Final
P ojec Deg ee, Ka ls uhe Ins i u e o Technology, 2010. 182
[DW07] Pa ick De ine-W igh . Ene gy ci izenship: psychological as-
pec s o e olu ion in sus ainable ene gy echnologies. People
cen ed eco-design: consume adop ion o low and ze o ca bon
p oduc s and sys ems, page 63, 2007. 24
270
BIBLIOGRAPHY
Na ional Ins i u e o S anda ds and Technology, page 33, 2012.
21, 25
[Joh07] Neil F Johnson. Two’s company, h ee is complexi y: a simple
guide o he science o all sciences. Onewo ld Pubns L d, 2007.
67
[Kau87] F ed J Kaudel. A li e a u e su ey on dis ibu ed disc e e e en
simula ion. SIGSIM Simul. Dig., 18(2):11–21, 1987. 86
[KC00] Joshua D Knowles and Da id W Co ne. App oxima ing he
nondomina ed on using he pa e o a chi ed e olu ion s a egy.
E olu iona y compu a ion, 8(2):149–172, 2000. 168
[KdD12a] En ique K eme s and J.M. Gonzalez de Du ana. Eme gen syn-
ch onisa ion p ope ies o a e ige a o demand side manage-
men sys em. Applied Ene gy, 2012. 191, 220, 222
[KdD12b] En ique K eme s and J.M. Gonzalez de Du ana. Towa ds com-
plex sys em design and managemen in he enginee ing domain -
The Sma g id Challenge. In Eu opean Con e ence on Complex
Sys ems 2012, 2012. 191, 194, 220, 223
[KdH09] S. Ka nouskos and T.N. de Holanda. Simula ion o a sma g id
ci y wi h so wa e agen s. In Compu e Modeling and Simula ion,
2009. EMS ’09. Thi d UKSim Eu opean Symposium on, pages
424–429, No 2009. 39
[KE95] J Kennedy and R Ebe ha . Pa icle swa m op imiza ion. In
Neu al Ne wo ks, 1995. P oceedings., IEEE In e na ional Con-
e ence on, olume 4, pages 1942 –1948 ol.4, 1995. 47, 116,
164
[KHTS09] P. Ko dik, M. Hasaj, M. T dy, and M. Sno ek. Building Au o-
ma ion Simula o and Con ol S a egy o In elligen and Ene gy
E icien Home. In Compu e Modeling and Simula ion, 2009.
EMS’09. Thi d UKSim Eu opean Symposium on, pages 154–159,
2009. 33
277

BIBLIOGRAPHY
[ki ] KIT web. h p://www.ki .edu/. 139
[KKE01] James F Kennedy, James Kennedy, and Russell C Ebe ha .
Swa m in elligence. Mo gan Kau mann, 2001. 162
[KR02] Ralph Kimball and Ma gy Ross. The Da a Wa ehouse Toolki :
The Comple e Guide o Dimensional Modeling. John Wiley &
Sons, Inc., New Yo k, NY, USA, 2nd edi ion, 2002. 63
[K e13] En ique Albe o K eme s. Modelling and Simula ion o Elec-
ical Ene gy Sys ems h ough a Complex Sys ems App oach us-
ing Agen -Based Models. KIT Scien i ic Publishing, Ka ls uhe,
2013. 38, 40, 192
[KRH+10] Ma iam Ki an, Paul Richmond, Mike Holcombe, Lee Shawn
Chin, Da id Wo h, and Ch is G eenough. Flame: simula ing
la ge popula ions o agen s on pa allel ha dwa e a chi ec u es.
In P oceedings o he 9 h In e na ional Con e ence on Au onom-
ous Agen s and Mul iagen Sys ems: olume 1-Volume 1, pages
1633–1636. In e na ional Founda ion o Au onomous Agen s
and Mul iagen Sys ems, 2010. 45
[KSE07] Oli e Ko b, Thomas S ¨
u zle, and Thomas E Exne . An an
colony op imiza ion app oach o lexible p o ein–ligand docking.
Swa m In elligence, 1(2):115–134, 2007. 47
[Kuh12] Thomas S Kuhn. The s uc u e o scien i ic e olu ions. Uni e -
si y o Chicago p ess, 2012. 249
[KV10] E K eme s and Pablo Viejo. A complex sys em modelling ap-
p oach o descen alized simula ion o elec ical mic ogi ds. In
15 h IEEE In e na ional Con e ence on Engine ing o Complex
Compu e Sys ems, 2010. 190
[Lan90] C.G. Lang on. Compu a ion a he edge o chaos: Pahse
ansi ions and eme gen compu a ion. Physica D: Nonlinea
Ph enomena, 42:12–37, 1990. 188
278
BIBLIOGRAPHY
[Lan01] Doug Laney. 3d da a managemen : Con olling da a olume,
eloci y and a ie y. META G oup Resea ch No e, 6, 2001. 70
[LB08] Lo enz Lu zenhise and Syl ia Bende . The “A e age Ame ican”
unmasked: Social s uc u e and di e ences in household ene gy
use and ca bon emissions. In 2008 ACEEE Summe S udy on
Ene gy E iciency in Buildings, 2008. 150
[LC11] Fe y Liene and Ma hew Ca son. Sma me e oll-ou o he
domes ic sec o (GB), 2011. 23
[LCRPS04] Sean Luke, Claudio Cio i-Re illa, Li iu Panai , and Kei h Sulli-
an. Mason: A new mul i-agen simula ion oolki . In P oceed-
ings o he 2004 Swa mFes Wo kshop, olume 8, 2004. 44
[Le 98] Simon A Le in. Ecosys ems and he biosphe e as complex ad-
ap i e sys ems. Ecosys ems, 1(5):431–436, 1998. 67
[LG03] Ya’an Liu and Xiaohong Guan. Pu chase alloca ion and demand
bidding in elec ic powe ma ke s. Powe Sys ems, IEEE T ans-
ac ions on, 18(1):106–112, Feb 2003. 31
[LGRT11] Ma in Lukasiewycz, Michael Glaß, Felix Reimann, and J¨
u gen
Teich. Op 4J - A Modula F amewo k o Me a-heu is ic Op im-
iza ion. In P oceedings o he Gene ic and E olu iona y Com-
pu ing Con e ence (GECCO 2011), pages 1723–1730, Dublin,
I eland, 2011. 164
[LR04] Seok Won Lee and Da id C Rine. Case s udy me hodology de-
signed esea ch in so wa e enginee ing me hodology alida ion.
In SEKE, pages 117–122, 2004. 72
[LTDZ02] Ma co Laumanns, Lo ha Thiele, Kalyanmoy Deb, and Eck-
a Zi zle . Combining con e gence and di e si y in e olu-
iona y mul iobjec i e op imiza ion. E olu iona y compu a ion,
10(3):263–282, 2002. 118
279
BIBLIOGRAPHY
[LV03] Jus in Langse h and Ni hi Vi a a . Why p oac i e business in el-
ligence is a hallma k o he eal- ime en e p ise: Ou wa d bound.
In elligen En e p ise, 5(18):34–41, 2003. 61
[LVC09] MD Lemmon, G Venka a amanan, and P Chapman. Using mi-
c og ids as a pa h owa ds sma g ids. In Posi ion pape p esen-
ed a NSF wo kshop in cybe physical ene gy sys ems, Bal-
imo e, MD, 2009. 36
[MA04] R.T. Ma le and J.S. A o a. Su ey o mul i-objec i e op imiza-
ion me hods o enginee ing. S uc Mul idisc Op im, 26, 2004.
118
[Mas13] Gilbe M Mas e s. Renewable and e icien elec ic powe sys-
ems. John Wiley & Sons, 2013. 5, 28
[MBLA96] Nelson Mina , Roge Bu kha , Ch is Lang on, and Mano
Askenazi. The swa m simula ion sys em: A oolki o build-
ing mul i-agen simula ions. Technical epo , San a Fe Ins i u e
San a Fe, 1996. 44
[MCH+94] Melanie Mi chell, James P. C u ch ield, Pe e T. H abe , In G.
Cowan, D. Pines, D. Melzne (edi o s, and Complexi y Me a-
pho s. Dynamics, compu a ion, and he “edge o chaos”: A
e-examina ion. In Complexi y:Me apho s, Models, and Reali y,
pages 497–513. Addison-Wesley, 1994. 188
[ME04] Leon Messe schmid and And ies Pe us Engelb ech . Lea ning
o play games using a pso-based compe i i e lea ning app oach.
E olu iona y Compu a ion, IEEE T ansac ions on, 8(3):280–
288, 2004. 48
[ME10] An hony R Me ke and Randy L Ekl. Secu i y echnology
o sma g id ne wo ks. Sma G id, IEEE T ansac ions on,
1(1):99–107, 2010. 19
[Mey03] Niels I Meye . Eu opean schemes o p omo ing enewables in
libe alised ma ke s. Ene gy policy, 31(7):665–676, 2003. 18
280
BIBLIOGRAPHY
[MH11] Janine Mo ley and Mike Hazas. The signi icance o di e ence:
Unde s anding a ia ion in household ene gy consump ion. In
ECEEE 2011 summe s udy. Ene gy e iciency i s : he ounda-
ion o a low ca bon socie y, 2011. 150
[MHC93] Melanie Mi chell, Pe e T. H abe , and James P. C u ch ield. Re-
isi ing he edge o chaos: E ol ing cellula au oma a o pe o m
compu a ions. Complex Sys ems, 7:89–130, 1993. 188
[Mik01] Donald C Mikulecky. The eme gence o complexi y: science
coming o age o science g owing old? Compu e s & chemis y,
25(4):341–348, 2001. 68
[Mil05] F. Milano. An Open Sou ce Powe Sys em Analysis Toolbox.
IEEE T ansac ion on Powe Sys em, ol. 20, no.3, 2005. 35, 37
[Mis86] Jayade Mis a. Dis ibu ed disc e e-e en simula ion. ACM Com-
pu . Su ., 18(1):39–65, 1986. 86
[MJ12] Ma k L Mi chell and Janina M Jolley. Resea ch design explained.
Cengage Lea ning, 2012. 72
[MOF02] OMG MOF. Omg me a objec acili y (mo ) speci ica ion 1. 4,
2002. 52
[Mom09] James A Momoh. Sma g id design o e icien and lexible
powe ne wo ks ope a ion and con ol. In Powe Sys ems Con-
e ence and Exposi ion, 2009. PSCE’09. IEEE/PES, pages 1–8.
IEEE, 2009. 18, 19
[MP10] M Moglia and P Pe ez. Modelling an u ban wa e sys em on he
edge o chaos. En i onmen al Modelling and So wa e, 25:1528–
1538, 2010. 189
[MPBL10] A. Mon i, F. Ponci, A. Benigni, and J. Liu. Dis ibu ed in el-
ligence o sma g id con ol. In Nonsinusoidal Cu en s and
Compensa ion (ISNCC), 2010 In e na ional School on, pages
46–58, 2010. 32
281
BIBLIOGRAPHY
[MS84] James L McKenney and Mo on M Sco . Managemen decision
sys ems: compu e -based suppo o decision making. Ha a d
Business School P ess, 1984. 60
[Neg04] Solomon Negash. Business in elligence. Communica ions o he
Associa ion o In o ma ion Sys ems, 13(1):177–195, 2004. 61,
62
[NN12] G egoi e Nicolis and Ca hy Nicolis. Founda ions o Complex
Sys ems: Eme gence, In o ma ion and P edici ion. Wo ld Sci-
en i ic, 2012. 41
[NNdGW09] DT Nguyen, M. Negne i sky, M. de G oo , and C. Wang. De-
mand esponse in he e ail ma ke : Bene i s and challenges.
In Powe Enginee ing Con e ence, 2009. AUPEC 2009. Aus-
alasian Uni e si ies, pages 1–6. IEEE, 2009. 5, 30
[Nwa02] C. Nwankpa. Vol age S abili y Toolbox, e sion 2, Cen e o
Elec ic Powe Enginee ing, D exel Uni e si y, 2002. 35, 37
[Nyl99] A Nylund. T acing he bi amily ee. Knowledge Managemen ,
1999. 60
[OEC] OECD. OECD G een G ow h S udies: Ene gy. h p:
//www.oecd.o g/g eeng ow h/g eening-ene gy/
49157219.pd . 4, 20
[O ] O ice o Elec ici y Deli e y and Ene gy Reliabili y.
Sma G id. h p://ene gy.go /oe/se ices/
echnology-de elopmen /sma -g id. 4, 20
[OLA97] OLAP Council. OLAP De ini ions, 1997. 63
[O g11] O ganisa ion o Economic Co-ope a ion and De elopmen .
G een G ow h S udies: Ene gy, 2011. 3
[O 05] Gunna O e. En wicklung und Tes eine in eg a i en Typolo-
gie de Lebens ¨
uh ung ¨
u die Bundes epublik Deu schland. Con-
282

BIBLIOGRAPHY
s uc ion and es o an in eg a i e li es yle- ypology o ge many.
Zei sch i ¨
u Soziologie, 34(6):442–467, Decembe 2005. 150
[PAB12] Co alie Pe e mann, Sou ian Ben Amo , and Alain Bui. A com-
plex sys em app oach o a eliable sma g id modeling. In KES,
pages 149–158, 2012. 7
[Pac] Paci ic No hwes Na ional Labo a o y. G idLAB-D. 38
[Pa 06] V Pa e o. Manuale di economica poli ica, socie a edi ice lib a ia.
Manual o poli ical economy, 1971, 1906. 118
[PD11] P. Palensky and D. Die ich. Demand side managemen : Demand
esponse, in elligen ene gy sys ems, and sma loads. Indus ial
In o ma ics, IEEE T ansac ions on, 7(3):381–388, Aug 2011. 29,
31, 33
[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 2009 IEEE/PES Powe Sys ems Con e ence and Exposi ion,
pages 1–8. IEEE, Ma ch 2009. 8, 191
[PHS+08] Fiona A. C. Polack, Tim Ho e d, Adam T. Sampson, Susan
S epney, and Jon Timmis. Complex sys ems models: Enginee -
ing simula ions. In ALi e XI, Winches e , UK, Augus 2008, pages
482–489. MIT P ess, 2008. 33, 75
[PKZK08] P. Palensky, F. Kupzog, A. A. Zaidi, and Zhou Kai. Modeling
domes ic housing loads o demand esponse. In Indus ial Elec-
onics, 2008. IECON 2008. 34 h Annual Con e ence o IEEE,
pages 2742–2747, 2008. 8, 32, 38, 42, 145, 250
[PL06] Jukka Paa e o and Pe e Lund. A model o gene a ing house-
hold elec ici y load p o iles. In e na ional Jou nal o Ene gy
Resea ch, 30(5):273–290, 2006. 150
[PM98] Ka he ine N P obs and Michael H McGo e n. Long- e m s ew-
a dship and he nuclea weapons complex: The challenge ahead.
Resou ces o he Fu u e, 1998. 19
283
BIBLIOGRAPHY
[Pol08] Michael G Polli . The u u e o elec ici y (and gas) egula ion
in a low-ca bon policy wo ld. Ene gy Jou nal, 2008. 18
[Poo87] R J Pooley. An in oduc ion o p og amming in SIMULA. Black-
well Scien i ic Publica ions, L d., Ox o d, UK, UK, 1987. 86
[Poo01] John D. Poole. Model-D i en A chi ec u e: Vision, S anda ds
and Eme ging Technologies. In Wo kshop on Me amodeling and
Adap i e Objec Models, ECOOP, 2001. 50
[Pow] Powe ech. TSAT - T ansien Secu i y Assessmen Tool. 37
[Pow99] D.J. Powe . Decision Suppo Sys ems Glossa y. DSS Resou ces,
1999. 63
[Pow02] Daniel Powe . Decision suppo sys ems: concep s and esou ces
o manage s. G eenwood Publishing G oup, 2002. 60, 61
[Pow07] Daniel J Powe . A b ie his o y o decision suppo sys ems.
DSSResou ces. COM, Wo ld Wide Web, h p://DSSResou ces.
COM/his o y/dsshis o y. h ml, e sion, 4, 2007. 60, 61
[PRS08] D Pudjian o, C Ramsay, and G S bac. Mic og ids and i ual
powe plan s: concep s o suppo he in eg a ion o dis ibu ed
ene gy esou ces. P oceedings o he Ins i u ion o Mechanical
Enginee s, Pa A: Jou nal o Powe and Ene gy, 222(7):731–
741, 2008. 24
[RBK11] G zego z Rozenbe g, Thomas Bck, and Joos N Kok. Handbook
o na u al compu ing. Sp inge Publishing Company, Inco po -
a ed, 2011. 12
[RC03] Ken Rudin and Da id C essy. Will he eal analy ic applica ion
please s and up? DM REVIEW, 13:30–41, 2003. 62
[RCO09] N. Ruiz, I. Cobelo, and J. Oya zabal. A di ec load con ol
model o i ual powe plan managemen . Powe Sys ems, IEEE
T ansac ions on, 24(2):959–966, May 2009. 33
284
BIBLIOGRAPHY
[RCV07] M Rosas-Casals and S Val e de. Topological ulne abili y o
he Eu opean powe g id unde e o s and a acks. In e na ional
Jou nal o Bi u ca ions and Chaos, 17:2465–2475, 2007. 191
[REE] Red El´
ec ica de Espa˜
na. h p://www. ee.es/es/. 29
[RH97] Wendy Robson and P en ice Hall. S a egic managemen and
in o ma ion sys ems: an in eg a ed app oach, 1997. 71
[RKU11] Anand Rajama an, Ju e Kesko ec, and Je ey D. Ullman. Min-
ing o Massi e Da ase s. Camb idge Uni e si y P ess, 2011. 110
[RLJ06] S. F. Railsback, S. L. Ly inen, and S. K. Jackson. Agen -based
Simula ion Pla o ms: Re iew and De elopmen Recommenda-
ions. SIMULATION, 82(9):609–623, Sep embe 2006. 43, 44
[RMMG08] Syl ain Rougemaille, F ´
ed´
e ic Migeon, Ch is ine Mau el, and
Ma ie-Pie e Gleizes. Model d i en enginee ing o designing
adap i e mul i-agen s sys ems. In Enginee ing Socie ies in he
Agen s Wo ld VIII, pages 318–332. Sp inge , 2008. 56
[Ros77] Robe Rosen. Complexi y as a sys em p ope y. In e na ional
Jou nal O Gene al Sys em, 3(4):227–232, 1977. 68
[RPT07] Sai u Rahman, Manisa Pipa anasompo n, and Yonael Teklu. In-
elligen dis ibu ed au onomous powe sys ems (idaps). In Powe
Enginee ing Socie y Gene al Mee ing, 2007. IEEE, pages 1–8.
IEEE, 2007. 21
[RVRJ11] Sa apali Ramchu n, Pe uk ishnen Vy elingum, Alex Roge s,
and Nick Jennings. Agen -Based Con ol o Decen alised De-
mand Side Managemen in he Sma G id. In The 10 h In-
e na ional Con e ence on Au onomous Agen s and Mul iagen
Sys ems-Volume 1, pages 5–12, Taipei, Taiwan, 2011. 8, 32, 33
[RWE11] RWE Rhein-Ruh Ve eilne z. Las p o ile, 2011. 143
[SAA+02] Dag IK Sjobe g, Ben e Anda, E ik A isholm, To e Dyba, Magne
Jo gensen, Amela Ka ahasano ic, Espen F imann Ko en, and
285
BIBLIOGRAPHY
Ma ek Vok´
ac. Conduc ing ealis ic expe imen s in so wa e en-
ginee ing. In Empi ical So wa e Enginee ing, 2002. P oceed-
ings. 2002 In e na ional Symposium n, pages 17–26. IEEE, 2002.
72, 243
[SC05] Ma ga i a Reyes Sie a and Ca los A Coello Coello. Imp o ing
PSO-based mul i-objec i e op imiza ion using c owding, mu a-
ion and ? -dominance. In In EMO?2005, pages 505?519. LNCS
3410, pages 505–519. Sp inge -Ve lag, 2005. 119, 164, 168
[SCC02] William R.. Shadish, Thomas D Cook, and Donald Thomas
Campbell. Expe imen al and quasi-expe imen al designs o gen-
e alized causal in e ence. Wadswo h Cengage lea ning, 2002.
72
[SCF+06] Jean-S´
ebas ien So e , Ga¨
elle Cal a y, Jean-Ma ie Fa e, Jo¨
elle
Cou az, Alexand e Demeu e, and Lionel Balme. Towa ds model
d i en enginee ing o plas ic use in e aces. In Sa elli e E en s a
he MoDELS 2005 Con e ence, pages 191–200. Sp inge , 2006.
56
[Sch10] Thomas Schwa z. Lebenss ile und W¨
ahle e hal en in S u ga .
E gebnisse de S u ga e Lebenss ilbe agung 2008. S a is ik
und In o ma ionsmanagemen , July 2010. 151, 152
[SD06] Mohsen Amini Salehi and Hossein Delda i. G id load balancing
using an echo sys em o in elligen an s. In Pa allel and Dis ib-
u ed Compu ing and Ne wo ks, pages 47–52, 2006. 46
[Sei03] Edwin Seidewi z. Wha models mean. IEEE so wa e, 20(5):26–
32, 2003. 51
[SGRM07] F ank Sensuß, Massimo Genoese, Ma io Ragwi z, and Dominik
M¨
os . Agen -based simula ion o elec ici y ma ke s-a li e a u e
e iew. Ene gy S udies Re iew, 15(2):2, 2007. 36, 191
286