Introduc)on*–*Chris*Develder*
! PhD,%Ghent%University,%2003%• “Design%and%analysis%of%op<cal%packet%switching%networks”%
! Professor%at%Ghent%University%since%Oct.%2007%• Research(Interests:%smart*grids%(op<miza<on/scheduling%algorithms%for%DSM/DR;%data%analy<cs),%informa)on*retrieval/extrac)on*(e.g.,%knowledge%base%popula<on,%event%rela<ons%in%news%archives);%op)cal*networks*(dimensioning,%resilience%schemes,%ILP)%%
• Visi<ng%researcher%at%UC%Davis,%CA,%USA,%JulVOct.%2007%%(op<cal%grids)%• Visi<ng%researcher%at%Columbia%Univ.,%NY,%USA,%2013V14%%(IR/IE)%
! Industry%Experience:%network%planning/design%tools%• OPNET%Technologies%(now%part%of%Riverbed),%2004V05%
! More%info:%h_p://users.atlan<s.ugent.be/cdvelder%%
%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Smart*grid*algorithms:*Knowing*and*controlling*power*consump)on*
Chris%Develder,%Kevin%Mets,%Ma_hias%Strobbe%%%%Ghent%University%–%iMinds%Dept.%of%Informa<on%Technology%–%IBCN%
Smart*Grids*
Distributed*genera)on*(large*scale)*Green*energy*sources*(fluctua)ng)*
Distributed*genera)on*(small*scale)*
Local*energy*storage*
PHEV*charging*
(residen)al)*
PHEV*charging*(car*parks)*
Demand*side*management*
ICT*infrastructure*
New*services*&*business*models*Fault*detec)on?*Restora)on?*
Data*processing?*Privacy,*security?*Pricing*schemes?*
…*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Power*grid*structure*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Transmission*network*(operated*by*TSO)*
Distribu)on*network*(operated*by*DSO)*
Outline*
1. Introduc<on%Part*I:*Algorithms*for*DSM/DR*2. Example%1:%Peak%shaving%%3. Example%2:%Wind%balancing%4. Tools%to%study%smart%grid%cases%Part*II:*Data*analy)cs*5. Clustering%smart%metering%data%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
K.(Mets,(R.(D'hulst(and(C.(Develder,("Comparison+of+intelligent+charging+algorithms+for+electric+vehicles+to+reduce+peak+load+and+demand+variability+in+a+distribu9on+grid",(J.(Commun.(Netw.,(Vol.(14,(No.(6,(Dec.(2012,(pp.(672M681.(doi:10.1109/JCN.2012.00033((
Example*case*study:*EV*charging*
! Research%ques<ons:%1. Impact%of%(uncontrolled)%EV%charging%in%a%residen<al%environment?%2. Minimal%impact%on%load%peaks%we%could%theore<cally%achieve?%3. How%can%we%minimize%the%impact%of%EV%charging%in%prac<ce?%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
! Sample%analysis%for%150%homes,%x%%of%them%own%a%PHEV%! BAU%=%maximally%charge%upon%arrival%at%home%
Impact*of*EV*charging*
0%
50%
100%
150%
200%
250%
12:00%
13:30%
15:00%
16:30%
18:00%
19:30%
21:00%
22:30%
00:00%
01:30%
03:00%
04:30%
06:00%
07:30%
09:00%
10:30%
12:00%
Total*loa
d*(kWh)*
Time*(h)*
No%PHEVs%BAUV10%%BAUV30%%BAUV60%%
0%%
20%%
40%%
60%%
80%%
100%%
10%% 30%% 60%%PHEV*Penetra)on*(%)*
Addi<onal%Power%Consump<on%Addi<onal%Peak%Load%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
! Objec<ves:%• Reduce%peak%load%• Fla_en%(total)%load%profile%(=%reduce%<meVvariability)%
• Avoid%voltage%viola<ons%
Controlling*EV*charging?*
Time Slot
Charging Rate (W)
Charging*schedule*
0%
50%
100%
150%
200%
250%
12:00%
14:15%
16:30%
18:45%
21:00%
23:15%
01:30%
03:45%
06:00%
08:15%
10:30%
Total*loa
d*(kWh)*
Time*(h)*
0%
50%
100%
150%
200%
250%
12:00%
14:15%
16:30%
18:45%
21:00%
23:15%
01:30%
03:45%
06:00%
08:15%
10:30%
Total*loa
d*(kWh)*
Time*(h)*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Quadra)c*Programming*(QP)*! Offline%algorithm%! Planning%window%! “Benchmark”%! Three%approaches:%
• Local%• Itera<ve%• Global%
Mul)YAgent*System*(MAS)*! Online%algorithm%! No%planning%window%"%current%<me%slot%info%only%(but%EV%bidding%changes%when%charging%deadline%approaches)%
! “Realis<c”%! Single%approach%
Smart*charging*algorithms*
Reference*scenario:%Uncontrolled%charging%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Smart*charging:*QP*
BAU%(uncontrolled)%
Local*control*(QP)%Global*control*(QP),*
Market*MAS%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Case*study*
! 63%Households%• Randomly%distributed%over%3%phases%
• Spread%over%3%feeders%
! Electrical%vehicles%• PHEV:%15%kWh%ba_ery%• Full%EV:%25%kWh%ba_ery%• Randomized%arrivals%(~5pm)%and%departures%(~6am)%
Scenario* PHEV*3.6*kW*
PHEV*7.4*kW*
EV*3.6*kW*
EV*7.4*kW*
Light% 4% 3% 2% 1%
Medium% 10% 10% 5% 4%
Heavy% 17% 16% 7% 7%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Results*(1)*–*Load*profiles*
0%
50%
100%
150%
200%
250%
300%
12:00% 15:00% 18:00% 21:00% 0:00% 3:00% 6:00% 9:00% 12:00%
Power*con
sum)o
n*(kW)*
Time*(hh:mm)*
Heavy*
0%
20%
40%
60%
80%
100%
120%
12:00% 15:00% 18:00% 21:00% 0:00% 3:00% 6:00% 9:00% 12:00%
Power*con
sum)o
n*(kW)* Light* Uncontrolled% Local%
Global% MAS%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Results*(2)*–*Load*peaks*&*variability**
QP1 = local QP2 = iterative QP3 = global
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Not%solved%en<rely!%(No(explicit(part(of(objec>ve(func>on!)%%
Note: 10 slots ~ 3.4% of the time
Results*(3)*–*Voltage*devia)ons*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Outline*
1. Introduc<on%Part*I:*Algorithms*for*DSM/DR*2. Example%1:%Peak%shaving%%3. Example%2:%Wind%balancing%4. Tools%to%study%smart%grid%cases%Part*II:*Data*analy)cs*5. Clustering%smart%metering%data%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Distributed*genera)on*(DG)*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Distributed*genera)on*(DG)*
! Mo<va<on%for%DG%• Use%renewable%energy%sources%(RES)%⇒%reduc)on*of*CO2*
• Energy*efficiency;%e.g.,%Combined%Heat%and%Power%(CHP)%• Genera<on%close%to%loads%• Deregula<on:%Open%access%to%distribu<on%network%• Subsidies%for%RES%• …%
! Technologies%• Wind%turbines%• Photovoltaic%systems%• CHP%(based%on%fossil%fuels%or%RES)%• Hydropower%• Biomass%• …%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Technical*impact*of*DG?*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Wind*turbines*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
! Horizontal%axis%%• Upwind%vs%downwind%• Needs%to%be%pointed%into%the%wind%• High%rota<onal%speed%(10V22%rpm)%• Needs%a%lot%of%space%(cf.%60V90m%high;%blades%20V40m)%
! Ver<cal%axis%• Omnidirec<onal%• No%need%to%point%to%wind%• Lower%rota<onal%speed%• Can%be%closer%together%
((((E.g.,(hZp://www.inflowMfp7.eu/((Savonius(Darrieus(
A*typical*wind*profile*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Case*Study*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
K.(Mets,(F.(De(Turck(and(C.(Develder,("Distributed+smart+charging+of+electric+vehicles+for+balancing+wind+energy",(in(Proc.(3rd(IEEE(Int.(Conf.(Smart(Grid(Communica>ons((SmartGridComm(2012),(Tainan(City,(Taiwan,(5M8(Nov.(2012,(pp.(133M138.(doi:10.1109/SmartGridComm.2012.6485972(
High*peak*loads*
Wind*balancing*with*EV*charging*
Supply/demand*imbalance*! Inefficient%use%of%RES%! Imbalance%costs%! High%peak%loads%
Undesirable!%
0%
10000%
20000%
30000%
40000%
50000%
60000%
70000%
0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00%
Power*(W
)*
Time*Uncontrolled% Wind%energy%
Day%2% Day*3* Day*4* Day*5* Day*6* Day*7*Day%1%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Distributed*control*
Balance%Responsible%Party%
EV% EV%
Coordinator Coordinator%
Exchange%of%control%messages%to%itera<vely%nego<ate%
charging%plans%for%a%specific%period%of%<me%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Electric*vehicle*model*
! Minimize%disu<lity:%• Charging%schedule%variables:%xt
k%=%%charging%rate%for%user*k*at%)me*t• Spread%demand%over%<me,%preferably%at%the%“preferred%charging%rate”%(pk),%
which%is%the%maximum%supported%charging%rate%in%our%case%• Model%behavior/preferences%of%the%subscriber%(βk)%
• Charging%schedule%for%a%window%of%T%<me%slots:%Minimize%disu<lity%
! Respect%energy%Requirement:%
• Vehicle%can%only%be%charged%between%arrival%<me%Sk and%departure%<me%Tk
(1)%
(2)%
(3)%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Balance*Responsible*Party*(BRP)*Model*
! Imbalance%Costs%• Minimize%imbalance%costs:%%%Penalty%cost%if%supply%≠%demand%• Supply:%%Wind%energy%(wt)%• Demand:%%Total%of%all%electric%vehicles%(dt)%• Tuning%parameter:%α• Cost%func<on:%
! For%a%planning%window%of%T%<me%slots,%minimize:%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Centralized*Op)miza)on*Model*
! Based%on%social%welfare%maximiza<on%• Minimize%imbalance%costs%• Minimize%user%disu<lity%
! Objec<ve:%
! Global%constraints:%
! Local%constraints:%• BRP:%supply%<%limit%• EV:%energy%&%<me%constraints%
Drawbacks:%1) Privacy:%sharing%of%
cost%&%disu<lity%func<ons,%arrival/departure%info,%…%
2) Scalability+
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
! Move%demandVsupply%constraint%into%objec<ve,%w/%Lagrange%mul<plier%λt
! No<ce:%Objec<ve%func<on%is%separable%into%K+1%problems%that%can%be%solved%in%parallel%(assuming(λt(are(given)%
! Itera<vely%update%pricing%vector…%%
Distributed*op)miza)on*model*
1%BRP%problem%
K%subscriber%problems%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
original(objec>ve( constraint(
Distributed*op)miza)on*model*scheme:*
1. Coordinator%distributes%virtual%prices%2. BRP%solves%local%problem%3. Subscribers%solve%local%problem%4. Coordinator%collects%schedules:%
• BRP:%%• EVs:%
5. Coordinator%updates%virtual%prices:%%
6. Repeat%un<l%demand%=%supply%
in%parallel%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Case*study:*Assump)ons*
! Wind%energy%supply%≈%EV%energy%consump<on%! Energy%supply%=%6.8%MWh%
! 100%Electric%vehicles%! Ba_ery%capacity:%10%kWh%ba_ery%! Maximum%charge%power:%3.68%kW%! Arrivals%&%departures:%sta<s<cal%model%! Charging%at%home%scenario%
! Time%! Simulate%4%weeks%! Time%slots%of%15%minutes%! Planning%window%of%24%hours%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Case*study:*Algorithms*
! Uncontrolled*business*as*usual*(BAU)*• EV%starts%charging%upon%arrival%• EV%stops%charging%when%stateVofVcharge%is%100%%• No%control%or%coordina<on%
%
! Distributed*algorithm*• Executed%at%the%start%of%each%<me%slot%
! “Ideal*world”*benchmark*• Offline%allVknowing%algorithm%determines%schedules%for%ALL%sessions%• No%EV%disu<lity%func<on%"%maximum%flexibility%• Objec<ve:%
min
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Results:*Uncontrolled*BAU*vs.*Distributed*
0%
10000%
20000%
30000%
40000%
50000%
60000%
70000%
0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00%
Power*(W
)*
Time*
Uncontrolled% Wind%energy%
Day%2% Day*3* Day*4* Day*5* Day*6* Day*7*Day%1%
0%
5000%
10000%
15000%
20000%
25000%
30000%
0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00%
Power*(W
)*
Time*
Distributed% Wind%energy%
Day%2% Day*3* Day*4* Day*5* Day*6* Day*7*Day%1%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Results:*Distributed*vs.*Benchmark*
0%
5000%
10000%
15000%
20000%
25000%
30000%
0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00% 6:00%12:00%18:00%0:00%
Power*(W
)*
Time*
Benchmark% Distributed% Wind%energy%
Day%2% Day*3* Day*4* Day*5* Day*6* Day*7*Day%1%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Results:*Energy*Mix*
! Total%energy%consump<on%≈%6.8%MWh%! Substan<al%increase%in%the%use%of%renewable%energy%! Reduced%CO2%emissions%
2.72%
4.62% 4.99%
4.07%
2.18% 1.80%
0.00%
1.00%
2.00%
3.00%
4.00%
5.00%
6.00%
7.00%
8.00%
Uncontrolled% Distributed% Benchmark%
Energy*su
pply*(M
Wh)*
Charging*strategy*
Renewables% Non%renewables%
1450%
798%
633%
0%
200%
400%
600%
800%
1000%
1200%
1400%
1600%
Uncontrolled% Distributed% Benchmark%
CO2*em
ission
s*(kg)*
Charging*strategy*
Renewables:%%%%%%%%%%%%%7.4%CO2%g/kWh%Non%Renewables:%351.0%%CO2%g/kWh%
−45%*
40%% 68%% 73%%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
−56%*
Contribu)on*from*RES* Reduc)on*of*CO2*emissions*
Conclusions*
! Objec)ve:%balance%wind%energy%supply%with%electric%vehicle%charging%demand%
! Method:%Distributed%coordina<on%algorithm%where%par<cipants%exchange%virtual%prices%and%energy%schedules%
! Performance:%Distributed%coordina<on%significantly%be_er%than%BAU,%close%to%“ideal%world”%benchmark%• Increased%usage%of%renewable%energy%sources%• Reduc<on%of%CO2%emissions%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Outline*
1. Introduc<on%Part*I:*Algorithms*for*DSM/DR*2. Example%1:%Peak%shaving%%3. Example%2:%Wind%balancing%4. Tools%to%study%smart%grid%cases%Part*II:*Data*analy)cs*5. Clustering%smart%metering%data%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
K.(Mets,(J.(Aparicio(and(C.(Develder,("Combining+power+and+communica9on+network+simula9on+for+costAeffec9ve+smart+grid+analysis",(IEEE(Commun.(Surveys(Tutorials,(Vol.(PP,(2014,(pp.(1M26.(doi:10.1109/SURV.2014.021414.00116(
Problem*Statement*
! Simulators%are%already%used%in%the%two%domains:%• Communica)on%network%engineering%• Power%engineering%
%! In%a%coYsimula)on*approach,%power%&%communica<on%are%loosely%coupled%• Requires%careful%synchronisa<on%• Drawback:%no%integra<on%of%tools%
ns-2 / ns-3
OpenDSS
OMNeT++
Matlab tools
…
…
Power Grid Simulation
Communication Network Simulation
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Challenge*for*coYsimula)on:*Synchronisa)on*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Figure: [H.Lin et al., 2012]
Integrated%(combined)%power%grid%and%communica<on%network%simula<on%"%Large%scale%smart%grid%simula<ons%
Our*proposed*solu)on*
Power Grid Simulation
Communication Network Simulation
Middleware Layer
Application Layer
DSM/DR% AMI% PMU% …
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Outline*
1. Introduc<on%Part*I:*Algorithms*for*DSM/DR*2. Example%1:%Peak%shaving%%3. Example%2:%Wind%balancing%4. Tools%to%study%smart%grid%cases%Part*II:*Data*analy)cs*5. Clustering%smart%metering%data%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
K.(Mets,(F.(Depuydt.(and(C.(Develder,("TwoAstage+load+paGern+clustering+using+fast+wavelet+transforma9on",(IEEE(Trans.(Smart(Grid,(2015,((to(appear(
Clustering*smart*metering*data*
! Goal:*Iden<fy%different%types%of%daily%power%consump<on%<me%series*1. Single%household:%dis<nct%types%of%daily%load%pa_erns?%2. Over%whole%popula<on:%dis<nct%groups%of%users?%
! Why?*• Demand%analysis%(na<onVwide,%distribu<on%substa<ons,%…%single%houses)%• Customer%segmenta<on,%tariffs,%billing…%• Power%system%planning%• Load%forecas<ng%• Demand%response%programs%• …%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
TwoYstage*load*paoern*clustering*
< >
(i) typical patterns per user (ii) overall load pattern clustering
user 1...
user N
...
. . .
. . .
global cluster 1
global cluster 2
global cluster 3
global cluster K
n 1Kn 11, ... ,
< >n NKn N1, ... ,
cluster selectrepresentative
pattern
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Can%run%in%parallel,%simultaneously%for%all%users%
Representa<ve%pa_ern%=%real%pa_ern%closest%to%center%
Core*ideas*
! Hierarchical%scheme%! Wavelet%transforma<on:%
• Dimensionality%reduc<on%• Invariance/tolerance%to%<me%shi|ing%
! GVmeans%(instead%of%kVmeans)%[Hamerly2003]%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
128%features/day% 128%features/day% 7Y8*features/day*
x( f(
96%features/day%
Upsampling%(Linear%
Interpola<on)%
Fast%Wavelet%Transform%(Haar%wavelet)%
Energy%per%detail%scale%
xu( xw(
G.(Hamerly,(C.(Elkan,(“Learning(the(k(in(kMmeans”,(NIPS(2003(
Cluster 1 Cluster 2 Cluster 3
0
1
2
3
4
00:00 06:00 12:00 18:00 00:0000:00 06:00 12:00 18:00 00:0000:00 06:00 12:00 18:00 00:00Time
Powe
r con
sum
ptio
n (k
Wh)
meanrepresentativereal
Sample*result:*Single*user*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
For%alpha%=%0.01%%"%low%number%of%clusters%Note:(representa>ve(≠(arithme>c(mean(
Time*vs*wavelet*domain:*Number%of%clusters%
alpha = 15.00% alpha = 10.00% alpha = 5.00%
alpha = 2.50% alpha = 1.00% alpha = 0.01%0
10
20
30
0
10
20
30
40
50
0
10
20
30
40
0
10
20
30
40
50
0
20
40
0
25
50
75
0 25 50 75 100 0 25 50 75 100 0 25 50 75 100cluster size
num
ber o
f clu
ster
s
wavelet featurestime series
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Time%shi|ing%"%distance%metric%differences%in%<me%greater%than%in%wavelet%feature%domain%
Smaller%alpha%"%convergence%to%%the%same%clusters%
Time*vs*wavelet*domain:*Cluster%quality%(in%<me%domain)%
J MIA CDI
SMI DBI WCBCR
0.5
1.0
1.5
2.0
2.5
0.050
0.075
0.100
0.125
0.150
0.175
1
2
3
4
0.0
0.2
0.4
0.6
0.8
0.0
0.1
0.2
0.3
0.4
0
500
1000
1500
0 25 50 75 100 0 25 50 75 100 0 25 50 75 100Number of clusters
Met
ric v
alue
time serieswavelet features
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Time%shi|ing:%distance%metric%differences%in%<me%domain%greater%than%in%wavelet%feature%domain%"%Metrics%calculated%in%<me%domain%worsen%
Conclusions*
! Totally%unsupervised%clustering%process%• No%a%priori%defini<on%of%‘typical%day’,%groupings%into%weekday/weekend%…%• Cluster%quality%not%%
! Note%on%scalability:%• Stage%1%=%executed%per%user%(in%parallel)%• Stage%2%=%number%of%profiles%to%cluster%is%limited,%by%reducing%‘representa<ve’%profile%%
• Vector%space%dimensionality%is%reduced%by%FWT%(96%"%7%or%8%features)%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
WrapYup*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Summary*
! Challenge:%deal%with%renewable%sources%! Demand%response%algorithms:%ini<al%feasibility%studies%
• How%close%to%“best”%possible?%scalable?%• What%are%achievable%benefits?%
! Get%insight%in%consump<on/produc<on:%e.g.,%clustering%as%first%step%! Quan<fy%flexibility%=%amount%of%“shi|able”%power%+%what%<me?%
! What’s%next?%• Can%we%learn/predict%flexibility,%e.g.,%from%smart%metering%data?%• Can%we%infer%user*behavior,%and%from%there%(contextVaware)%preferences?%• Evaluate%business*case*of%flexibility?%• Convincingly%demonstrate%flexibility%exploita<on%in%the%real%world?%
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
E.g.,%refine%“disu<lity”%from%user;%“imbalance”%from%business%perspec<ve;%evaluate%using%real(is<c)%data…%
Thank*you*…*any*ques)ons?*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Thank*you*…*any*ques)ons?*
C.(Develder,("Smart(grid(algorithms:(Knowing(and(controlling(power(consump>on",(Invited(talk(at(UCI,(5(Jun.(2015(
Chris%Develder%[email protected]%%Ghent%University%–%iMinds%%
%