Embed Size (px)
Citation preview
Urban energy micro-simulation
Professor Darren Robinson
Chair in Building and Urban Physics
Deputy Head of Department of Architecture and Built Environment
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 2.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Conception optimisée du bâtiment par la SIMUlation et le Retour d'EXpérience, 00010 (2012) DOI:10.1051/iesc/2012simurex00010 © Owned by the authors, published by EDP Sciences, 2012
University of Nottingham (UK).
Article published online by EDP Sciences and available at http://www.iesc-proceedings.org or http://dx.doi.org/10.1051/iesc/2012simurex00010
1
Urban energy micro-simulation
Professor Darren Robinson
Chair in Building and Urban Physics
Deputy Head of Department of Architecture and Built Environment
University of Nottingham (UK).
2
Macro-simulation (1970s +)
source
sub-system
reservoir
producer
consumer
3
Typological sampling: EEP (1998 +)
Image procesing of DEMs (late 90s)…
Passive zone Non passive zone
Orientation Urban horizon angle LT-Urban
(1999-2000)
MATLAB / LT coupling: processing
Urban 3-D
database (DEM) Image
processing LT Model
Energy
consumption
Energy use @ 50% GR Energy use @ optimal GR
Difference Optimal GR
8
Micro-simulation: SUNtool (2002 +)
Energy, people and cities • EU Target: 80% GHG reductions by 2050; cities will be crucial
• Cities are complex self-organising systems: firms + individuals
• Buildings & infrastructural networks: increased interconnectivity
• But to transform cities in this complex landscape we need:
– Integrated models where behaviour is central
• Investment probability
• (Activity-dependent) behaviours
– Better understanding of how this behaviour can be influenced
– Better energy conservation: reinforces behaviour sensitivity
• Approach: Multi-Agent Stochastic Simulation (MASS)
Energy, people & cities: bigger picture
10
CitySim (2006+)
• CitySim: a detailed decision support tool to support
sustainable urban planning and design
• Micro-simulation for flexibility; breadth of scenarios
• Based on modelling of urban energy and matter flows
• Accounting for:
– Occupants’ behaviour
– Urban climate
– Synergetic energy & matter exchanges
• Applicable at the range of scales
• Productive and intuitive
Robinson et al, BS 2009; Robinson, 2011.
11
CitySim workflow
1) Create or import 3D
model and its clones
2) Describe envelope
composition
3) Describe occupancy
and equipment
schedules
4) Describe HVAC and ECS
systems
5) Simulate and analyse
12
Scene file (XML)
Climate file (ascii)
Rad model
HVAC models
CitySim scene creation
Rad model pre-process
ECS models
Constructions
ECS characteristics
Fuel characteristics
HVAC characteristics
Behavioural profiles
Write results
t+1?
Thermal model
CitySim solver: current version
• XML data exchange: GUI to solver
• ASCII data exchange: solver to GUI
• Solver coded in C++ Behavioural model
Radiation Model (SRA)
• Step 1 – Calculate energy distribution
throughout a discretised sky vault.
• Step 2 – Determine patch / obstruction view
factors and identify dominant obstruction
• Step 3 – Determine solar view factors
• Step 4 – Build matrices and determine energy
contribution from sun, sky and reflections
30
35
40
45
50
55
60
W/m2 SrAnnual Mean Diffuse Radiance
Robinson and Stone (2004), Solar Energy 77(3)
Robinson and Stone (2005) BSER&T 25(4).
Radiance
Annual irradiation,
MWh/m2
SRA for 10mx10m facade discretisation:
6500 surfaces
Difference plot: green is within 10%
SRA verification
Napier-Shaw
Medal,
CIBSE 2007
Daylight: SRA
H
2H 3H
Robinson and Stone (2006), Solar Energy 80(3)
Simple RC network thermal model
Kämpf and Robinson (2007), Energy & Buildings 39(4).
4150 4200 4250 4300
Time h
12.5
17.5
20
22.5
25
27.5
Temperature °CRoom 5
Outside
ESP r
Two Nodes
200 400 600 800 1000
ESP r Wh
200
400
600
800
1000
Two Nodes WhRoom 6
100 200 300 400 500
ESP r Wh
100
200
300
400
500
Two Nodes WhRoom 8
500 1000 1500 2000 2500 3000
ESP r Wh
500
1000
1500
2000
2500
3000
Two Nodes WhRoom 4
100 200 300 400 500 600
ESP r Wh
100
200
300
400
500
600
Two Nodes WhRoom 5
3850 3900 3950 4000 4050 4100
Time h
10
15
20
25
Temperature °CRoom 11
Outside
ESP r
Two Nodes
Good dynamic behaviour compared to ESP-r
Ideal Heating and Cooling Loads reasonable dispersion
Heating Ventilation and Air Conditioning Model
Mixture of Ideal Gases (Air and Vapour)
• enthalpy changes
HVAC systems
Water and/or Air
Co-generation of
heat and
electricity
Energy Conversion Systems
Solar collectors (PV + thermal)
Wind turbines
Boilers and co-generation systems
Heat pumps (air + ground: hoz / vert)
Sensible + latent heat storage
Occupants’ presence Energy & Buildings 40(2), 2007
Use of electrical lighting (associated electricity
consumption and internal heat gains)
Use of appliances (associated electricity and hot and cold water
consumption, internal heat gains and production of grey- and waste-water)
Production of internal metabolic heat
gains and “pollutants”
Position of blinds (solar gains;
daylight)
Use of windows and doors (ventilation rate and associated
heat gains and losses)
Key behavioural models (2001 - 2011)
Falsecoloured images
Tables, line graphs & results export
23
From deterministic to multi-agent
stochastic simulation
24
Stochastic simulation
We want stochastic models that will account for:
- the variety of behaviours (investments, occupants’ presence,
appliance use, comfort adaptations: personal & envelope)
- the variation over time of these behaviours,
- the variation between individuals of these behaviours.
To test scenarios to improve urban system performance
And to provide more precise inputs to our simulations.
- better load profiles for the sizing and control of energy
conversion systems and the design of supply networks:
building-embedded and district-wide.
25
Proposed MAS platform
1) Create synthetic
population of N agents
4) Long workplace absences
& destination
5) Presence chains (arrivals
/ departures)
2) Assign parameters to
each agent of population
3) Assign appliances to
household / workplace
Option 1: Occupants’ (short term) presence
Basic hypotheses:
• Independence of occupants
• Markov condition:
tTiXjXP ijtt :1
Page, Robinson et al (2007), Energy & Buildings 40(2).
From the profile of probability of presence:
Parameter of mobility:
Determination of probabilities of transition:
And closing the loop:
)()(1)()()1( 0111 tTtPtTtPtP
)()(
)()(:)(
1100
1001
tTtT
tTtTt
)1()(1
1)(01
tPtPtT
)(
)1()1()(
1
1
)(
)(1
)(
)1(
)(
)(10111
tP
tPtPtP
tP
tP
tP
tPT
tP
tPT
1110 1 TT 0100 1 TT
Occupant presence: theory
)()(1
)()(1
1
0
10
1111
0101
1
tTtT
tTtTX
X
t
t
0
1
0
Cumulated probability function
Dra
w a
nu
mb
er
ni
un
ifo
rmly
be
twe
en
0 a
nd
1.
01T
)1( 01T
34.001 T
n1 = 0.71
1
n2 = 0.18
n3 = 0.52
Occupant presence: IFM
In addition to this we simply need to account for long
absences… this we do as a pre-process
Weekly presence profile
Duration of a single presence
Cumulative presence
32
Option 2: Transport model
33
Proposed MAS platform
1) Create synthetic
population of N agents
4) Long workplace absences
& destination
5) Presence chains (arrivals
/ departures)
6) Presence-dependent
activities
2) Assign parameters to
each agent of population
3) Assign appliances to
household / workplace
Distribution of activities
• Activity probability
distributions pi(t) for a
typical day.
• Modelling approach 1:
Bernoulli process: activity
is drawn from the current
distribution. Requires
Nact [52] x Nts [24] =
1248 parameters
Hour
Dis
trib
utio
n o
f a
ctivitie
s
5 10 15 20
0.0
0.2
0.4
0.6
0.8
1.0
Personal Care
Household Activities
Helping HH MembersHelping Non-HH Memb.
Work-RelatedEducation
Consumer Purchases
Prof. & Pers. Care Serv.Household Services
Civic Obligations
Eating & Drinking
Socializing-RelaxingSports-Recr.
Religious Act.Volunteer Act.
Tel. Calls
TravelingMissing data
Hour
Dis
trib
utio
n o
f a
ctivitie
s
5 10 15 20
0.0
0.2
0.4
0.6
0.8
1.0
Transition probabilities
The probability of changing activity varies according to time of day and the activity.
0 5 10 15 20
0.0
00
.02
0.0
40
.06
0.0
80
.10
Hour
Tra
nsitio
n p
rob
ab
ility
1 2 3 4 5 6 7 8 9 11 13 15 17
0.0
00
.05
0.1
00
.15
0.2
00
.25
0.3
0
Activity type
Tra
nsitio
n p
rob
ab
ility
Ptrans(Act) Ptrans(t)
0 5 10 20
0.0
00.0
40.0
8Personal Care
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
60.1
20.1
8
Household Activities
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
80.1
20.1
6
Helping HH Members
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.1
0.3
0.5
Helping Non-HH Memb.
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
20.0
4
Work-Related
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
20.0
6
Education
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.1
0.3
0.5
Consumer Purchases
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
00.0
60.1
2
Prof. & Pers. Care Serv.
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
00.1
50.3
0
Household Services
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
0.2
0.4
0.6
Civic Obligations
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.1
10.1
40.1
7
Eating & Drinking
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
30.0
50.0
7
Socializing-Relaxing
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
40.0
80.1
2
Sports-Recr.
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
00.1
00.2
00.3
0
Religious Act.
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
40.0
8
Volunteer Act.
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.1
00.1
50.2
00.2
5
Tel. Calls
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.1
60.2
20.2
8
Traveling
Hour
Tra
nsiti
on p
robability
0 5 10 20
0.0
50.0
80.1
1
Missing data
Hour
Tra
nsiti
on p
robability
Ptrans(Activity, t)
Modelling approaches
• Modelling approach 2:
– Monte-Carlo simulation of activity changes with
ptrans(t), ptrans(Act) or ptrans(t, Act)
– Choice of new activities based on their probability
distribution
– Nts [Ptrans(t)] + Nact x Nts = 1272 parameters
• Modelling approach 3:
– Use of time-dependent transition probabilities pij(t)
– Nact2 x Nts = 64,896 parameters
Durations of activities
• The distribution of the durations of activities is easily modelled
by standard functions, eg. Exponential (1 parameter), Weibull
(2 parameters),…
• To efficiently model dynamics:
– When an activity starts, predict its
duration.
– When this time elapses, choose
a new activity, either from pi(t) or
pij(t).
– Requires (min): 2 x Nact [Weibull]
+ Nact x Nts = 1352 parameters.
Activity duration (min)
Fre
qu
en
cy
0 200 400 600 800 1000 1200 1400
01
00
02
00
03
00
04
00
0
Activity duration (min)
Fre
qu
en
cy
0 50 100 150 200
02
00
04
00
06
00
08
00
0
Sleeping Eating & drinking
Fitted duration distributions
40 Efficient aggregate model: Pi(t) + Wiebull
41
Proposed MAS platform
1) Create synthetic
population of N agents
4) Long workplace absences
& destination
5) Presence chains (arrivals
/ departures)
6) Presence-dependent
activities 7) Activity location
8) Thermal / gaseous
emissions
9) Environmental solver
10) Behavioural actions on:
• windows
• shading devices
• lights
• personal characteristics
• appliances
•HVAC systems
S(t) Action(t)
Action(t)
2) Assign parameters to
each agent of population
3) Assign appliances to
household / workplace
Windows: three different approaches
• Bernoulli process (using observed window states).
• Discrete-time Markov process (using observed actions on windows).
• Continuous-time random
process (using observed
delays between actions).
OpenClosed
OpenClosed
OpenClosed
Arrival
During presence
Departure
P01,dep P10,dep
P01,arr P10,arr
P01
P10
Actions on windows may be modelled in at least three ways:
OpenClosed
OpenClosed
OpenClosed
Arrival
During presence
Departure
P01,dep P10,dep
P01,arr P10,arr
Tclosed
Topen
Haldi and Robinson (2009), Building and Environment, 44(12).
Best Paper: Building and Environment, 2009
1): A Bernoulli process with static probability distributions
• Probability p that we observe the window to be open; which may depend
on one or several predictors x = (x1,…,xp).
• Binomial family of generalised linear
models, using the logit link:
)...exp(1
)...exp(,...,(
11
111
pp
ppp
xbxba
xbxbaxxp
Indoor temperature (°C)P
rop
ort
ion
of w
ind
ow
s o
pe
n
15 20 25 30
0.0
0.2
0.4
0.6
0.8
1.0
• Select the best univariate model, followed by examination of the usefulness of adding further variables (forward selection).
• Disadvantage: Poor simulation of the
dynamics of opening and closing
behaviour (reset at each time step).
2): Markov process predicting actions
• Here we consider the current state of the system using a Markov process (transitions).
OpenClosed
OpenClosed
OpenClosed
Arrival
During presence
Departure
P01,dep P10,dep
P01,arr P10,arr
P01
P10
• Occupancy transitions are a
crucial factor.
• Realistic dynamic description
of the states of windows.
Actions on arrival
Closed
OpenClosed
P01,arr1-P01,arr
Open
OpenClosed
1-P10,arrP10,arr
)0505.0286.097.3exp(1
)0505.0286.097.3exp(),(10
outin
outinoutinP
)45.0862.10433.0312.088.13exp(1
)45.0862.10433.0312.088.13exp(),(
,
,
01Rprevabsoutin
Rprevabsoutinoutin
fT
fTP
Indoor temperature is the most influential
environmental variable in both cases.
15 20 25 30
-10
01
02
03
04
0
Indoor temperature (°C)
Ou
tdo
or
tem
pe
ratu
re (
°C)
0
0.001
0.0025
0.005
0.01
0.025
0.05
0.1
0.25
0.5
1
0.0
5
0.1
0
.15
0.2
0.2
5
0.3
0.3
5
0.4
0.4
5
0.5
0
.55
0.6
0.6
5
0.7
0.7
5
0.8
15 20 25 30
-10
01
02
03
04
0
Indoor temperature (°C)O
utd
oo
r te
mp
era
ture
(°C
)
0
0.001
0.0025
0.005
0.01
0.025
0.05
0.1
0.25
0.5
1
0.0
5
0.1
0.1
5
0.2
0.2
5
0.3
0.3
5
0.4
0
.45
0.5
0.5
5
0.6
0.6
5
0.7
0
.75
0.8
Actions during presence
)0016.00779.00481.064.1exp(1
)0016.00779.00481.064.1exp(),(10
presoutin
presoutinoutin
T
TP
OpenClosed
P01
P10
1-P01 1-P10
OpenClosed
P01
P10
1-P01 1-P10
)336.00009.00271.0281.023.12exp(1
)336.00009.00271.0281.023.12exp(),(01
Rpresoutin
Rpresoutinoutin
fT
fTP
For openings, indoor temperature is the most
influential predictor, while it is outdoor
temperature for closings.
15 20 25 30
-10
01
02
03
04
0
Indoor temperature (°C)
Ou
tdo
or
tem
pe
ratu
re (
°C)
0
0.001
0.0025
0.005
0.01
0.025
0.05
0.1
0.25
0.5
1
0.0
2
0.0
4 0
.06
0.0
8
0.1
0.1
2
0.1
4
0.1
6
0.1
8
0.2
0.2
2
0.2
4
0.2
6
0.2
8
0.3
0
.32
15 20 25 30
-10
01
02
03
04
0
Indoor temperature (°C)O
utd
oo
r te
mp
era
ture
(°C
)
0
0.001
0.0025
0.005
0.01
0.025
0.05
0.1
0.25
0.5
1
0.005
0.01
0.015
0.02
0.025
0.03 0.035
0.04
0.045
0.05 0.055
0.06 0.065
0.07
0.075
0.08
0.085 0.09
0.095
Actions at departure
OpenClosed
OpenClosed
P01,dep
Open
OpenClosed
P10,dep
)83.084.01371.075.8exp(1
)83.084.01371.075.8exp(),,(
,,
,,
,01GFnextabsdmout
GFnextabsdmout
GFnextabsoutff
ffffP
15 20 25 30
01
02
03
0
Indoor temperature (°C)
Da
ily m
ea
n o
utd
oo
r te
mp
era
ture
(°C
)
0
0.001
0.0025
0.005
0.01
0.025
0.05
0.1
0.25
0.5
1
0.02
0.04
0.06
0.08 0.1
0.12
0.14
0.16 0.18
0.2 0.22
15 20 25 30
-50
51
01
52
02
53
0
Indoor temperature (°C)
Da
ily m
ea
n o
utd
oo
r te
mp
era
ture
(°C
)
0
0.001
0.0025
0.005
0.01
0.025
0.05
0.1
0.25
0.5
1
0.0
5
0.1
0.1
5
0.2
0.2
5
0.3
0.3
5
0.4
0.4
5
0.5
0.5
5
0.6
0.6
5
0.7
0.7
5 0
.8
0.8
5
0.9
)923.0614.10911.0213.054.8exp(1
)923.0614.10911.0213.054.8exp(),,,(
,,
,,
,,10GFnextabsdmoutin
GFnextabsdmoutin
GFnextabsdmoutinff
ffffP
Outdoor temperature (expectations) is the
most influential predictor, along with some
other related variables.
3): Continuous-time process
• Treating windows as a Markov process leads to redundant Monte-Carlo simulation (low transition probabilities).
• Alternative approach: directly simulate the durations that states will remain unchanged.
• More efficient, independent of time step and offers more general results.
• Survival analysis : we can fit parametric distributions, such as Weibull distribution:
Delays until opening
• Both indoor and outdoor conditions influence the durations for which windows are left closed.
• Weibull distribution with shape a<1 with scale l(in, out).
0 20 40 60 80 100 120
0.6
0.7
0.8
0.9
1.0
1.1
Follow-up time (min.)
Su
rviv
al cu
rve
-4°C
-3°C
-2°C
-1°C
0°C
1°C
2°C
3°C
4°C
5°C
6°C
7°C
8°C
9°C
10°C
11°C
12°C
13°C
14°C
15°C
16°C
17°C
18°C
19°C
20°C
21°C
22°C
23°C
24°C
25°C
26°C
27°C
28°C
29°C
30°C
31°C
32°C
0 20 40 60 80 100 120
0.6
0.7
0.8
0.9
1.0
1.1
Follow-up time (min.)
Su
rviv
al cu
rve
19°C
19.5°C
20°C
20.5°C
21°C
21.5°C
22°C
22.5°C
23°C
23.5°C
24°C
24.5°C
25°C
25.5°C
26°C
26.5°C
27°C
27.5°C
28°C
28.5°C
29°C
29.5°C
30°C
0 20 40 60 80 100 120
0.6
0.7
0.8
0.9
1.0
1.1
Follow-up time (min.)
Fitte
d s
urv
iva
l fu
nctio
n
19°C
19.5°C
20°C
20.5°C
21°C
21.5°C
22°C
22.5°C
23°C
23.5°C
24°C
24.5°C
25°C
25.5°C
26°C
26.5°C
27°C
27.5°C
28°C
28.5°C
29°C
29.5°C
30°C
Delays until closing
• Opening durations vary mainly with outdoor temperature (more differentiated curves).
• Weibull distribution with shape a<1 with scale l(out).
0 20 40 60 80 100 120
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Follow-up time (min.)
Su
rviv
al cu
rve
-4°C
-3°C
-2°C
-1°C
0°C
1°C
2°C
3°C
4°C
5°C
6°C
7°C
8°C
9°C
10°C
11°C
12°C
13°C
14°C
15°C
16°C
17°C
18°C
19°C
20°C
21°C
22°C
23°C
24°C
25°C
26°C
27°C
28°C
29°C
30°C
31°C
32°C
0 20 40 60 80 100 120
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Follow-up time (min.)
Su
rviv
al cu
rve
19°C
19.5°C
20°C
20.5°C
21°C
21.5°C
22°C
22.5°C
23°C
23.5°C
24°C
24.5°C
25°C
25.5°C
26°C
26.5°C
27°C
27.5°C
28°C
28.5°C
29°C
29.5°C
30°C
0 20 40 60 80 100 120
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Follow-up time (min.)
Fitte
d s
urv
iva
l fu
nctio
n
-4°C
-3°C
-2°C
-1°C
0°C
1°C
2°C
3°C
4°C
5°C
6°C
7°C
8°C
9°C
10°C
11°C
12°C
13°C
14°C
15°C
16°C
17°C
18°C
19°C
20°C
21°C
22°C
23°C
24°C
25°C
26°C
27°C
28°C
29°C
30°C
31°C
32°C
Hybrid algorithm
Recommended model:
• Discrete-time Markov
process for opening
actions and a Weibull
distribution for opening
times.
• Efficient modelling of
airflows on a time step
independent basis.
Thermal solver
Air flow model
Check occupancy
status
Check windowstatus
Check windowstatus
Check windowstatus
Opening time< time step ?
Occupancy model(pre-process)
Climate data:outdoor
temperature, rain
Windowstatus
DepartureIntermediateArrival
Open Closed Open Closed Open Closed
Indoortemperature
Drawopening time
Decrementopening time
YES
Opening if P < r
01,arr
Opening if P < r
01,int
Opening if P < r
01,dep
Closing if P < r
10,dep
WINDOW OPEN
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
1-Specificity (False positive rate)
Se
nsitiv
ity (
Tru
e p
ositiv
e r
ate
)
Markov Markov (1 var.)
Markov (close at long dep.) Markov (close at all dep.)
Weibull dist.
Markov (Weibull when open) Logit dist.
Humphreys alg. 2007 Humphreys alg. 2008
• The hybrid model offers the best trade-off between
sensitivity and specificity.
Predictive accuracy: discrimination
the true positive rate (or
sensitivity, or hit rate): TPR =
TP/(TP+FN)
the false positive rate (or fall-
out): FPR = FP/(FP+TN)
Predictive accuracy: aggregated results
• Finally, we compare the observed total number of windows open with
simulations. The hybrid model again offers the most reliable predictions.
02
46
81
01
21
4
Time
Op
en
win
do
ws
03/2005 05/2005 07/2005 09/2005 11/2005 01/2006
Observed
Mean simulated
02
46
81
01
21
4
Time
Op
en
win
do
ws
03/2005 05/2005 07/2005 09/2005 11/2005 01/2006
Observed Mean simulated
Accounting for individual diversity
Conventional behaviour
• Actions increase with in and out.
Predicitve thermal behaviours
• Similar, but decreased actions for
high out to avoid overheating.
Non-thermal behaviours
• almost independent of thermal
stimuli.
Pre-process: random behaviour assignment
)B 119.4E 000272.0787.0exp(1
)B 119.4E 000272.0787.0exp()B,E(
lowin
lowinlowin
raiseP
)B 380.2E 548000.0708.6-exp(1
)B 380.2E 548000.0708.6-exp()B,E(
lowin
lowinlowin
lowerP
• Probability of lowering:
• Probability of raising:
)B 1.085E 767000.07.501exp(1
)B 1.085E 767000.07.501exp()B,E(
lowin
lowinlowin
lowerP
)B 607.2E 000205.0571.3exp(1
)B 607.2E 000205.0571.3exp()B,E(
lowin
lowinlowin
raiseP
0 500 1000 1500 2000 2500 3000 3500
0.0
0.2
0.4
0.6
0.8
1.0
Indoor illuminance (lx)
Un
sh
ad
ed
fra
ctio
n
0
0.001
0.0025
0.005
0.01
0.025
0.05
0.1
0.25
0.5
1
0.0
1
0.0
2
0.0
3
0.0
4
0.0
5
0.0
6
0.0
7
0.0
8
0.0
9
0.1
0.1
1
0 500 1000 1500 2000 2500 3000 3500
0.0
0.2
0.4
0.6
0.8
1.0
Indoor illuminance (lx)
Un
sh
ad
ed
fra
ctio
n
0
0.001
0.0025
0.005
0.01
0.025
0.05
0.1
0.25
0.5
1
0.01
0.02 0.03
0.04
0.05 0.06
0.07
0.08
0.09
0.1
0.11
0.12
0.13
0.14 0.17
Blinds: Markov process
Robinson and Haldi, JBPS (in press).
Modelling shaded fraction choice
• In case of action, we determine whether full lowering or raising takes place (based on initial shaded fraction and outdoor illuminance).
• For partial lowering / raising, the lower unshaded fraction is well described by a Weibull distribution (scale l proportional to the initial blind position)
0.0 0.2 0.4 0.6 0.8 1.0
02
46
8
Added shaded fraction
De
nsity
SL init
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
Check occupancy
status
Does blind lowering
occur?
Does blind lowering
occur?
Drawshaded fraction
Does blind raising
occur?
Drawshaded fraction
Does blind raising
occur?
NO NO
NO NO
YESYES
YES YES
Occupancy model(pre-process)
Arrival Intermediate
Daylight model
Indoorilluminance
Climate data:outdoor
illuminanceBlinds status
Shading behaviour algorithm
Predictive accuracy
• Comparison of repeated simulations (Blower,sim, Bupper,sim) with measurements: predicted distribution close to reality, albeit with a slight over-estimation of shading.
• The total unshaded fraction for the whole building is well reproduced.
0
0.2
0.4
0.6
0.8
1
0 100 200 300 400 500minimum work plane illuminance [lux]
sw
itc
h-o
n p
rob
ab
ilit
y a
t a
rriv
al
user considers daylight
user does not consider daylight
Hunt
Measured switch-off probabilities as a function of
absence duration
0
0.25
0.5
0.75
1
sw
itch
-off
pro
bab
ilit
y w
hen
lea
vin
g
no controls
occ. sensor
indirect dimmed lighting
<30 minutes
30-59 minutes
1-2 hours
2-4 hours
4-12 hours
12-24 hours
24+ hours
Pigg
0
0.01
0.02
0.03
0 100 200 300 400 500 600 700
minimum work plane illuminance [lux], Emin
inte
rmed
iate
sw
itch
-on
pro
bab
ilit
y
P(Imin)=0.0027+0.017/{-64.19(log10Emin-2.41} for Emin=0
1 for Emin=0
Reinhart
Within-day switch-on probabilities
On arrival switch-on probabilities
Lights: Lightswitch 2002
Lights: Lightswitch 2002
Switch lights on
Is work place
already occupied?
YES
stochastic process:
switch on probability
(Fig. 3(a))
NO
NO
YES
NOYES
Has the room been
deserted for longer than sensor
delay time?
Are lights switched on?
Does occupant arrive?
Is there an
occupancy sensor?
YES
YES
NO YES
Does occupant leave?
Are lights switched on?NO
YES
stochastic process:
switch off probability
(Fig. 3(c))
NONO
NO
Switch lights off
YES
YES
NO
YES
stochastic process:
intermediate switch on
probability
(Fig. 3(b))
NO
YES
Does occupant arrive?YES
NO
NO
Set blinds.
Has the blind setting
just changed?
Does occupant leave?
Switch lights off Switch lights on Switch lights on
YES
stochastic process:
switch on probability
(Fig. 3(a))
NO
NO
Reinhart (2004), Solar Energy 77(1) 15-28.
61
Personal characteristics
Daily mean outdoor temperature (°C)
Clo
thin
g le
ve
l (c
lo)
0 5 10 15 20 25
0.0
0.2
0.4
0.6
0.8
1.0
0.3 clo
0.4 clo
0.5 clo
0.6 clo
0.7 clo
0.95 clo
Indoor temperature (°C)
Clo
thin
g le
ve
l (c
lo)
20 22 24 26 28
0.0
0.2
0.4
0.6
0.8
1.0
0.3 clo
0.4 clo
0.5 clo
0.6 clo
0.7 clo
0.95 clo
Indoor temperature (°C)
Pro
po
rtio
n o
f co
ld d
rin
ks c
on
su
mp
tio
n
18 20 22 24 26 28 30
0.0
0.2
0.4
0.6
0.8
1.0
Running mean outdoor temperature (°C)
Pro
po
rtio
n o
f h
ot d
rin
ks c
on
su
mp
tio
n
-5 0 5 10 15 20 25
0.0
0.2
0.4
0.6
0.8
1.0
Indoor temperature (°C)
Me
tab
olic a
ctivity le
ve
l (m
et)
20 22 24 26 28
0.0
0.2
0.4
0.6
0.8
1.0
1.0 met
1.2 met
1.6 met
Running mean outdoor temperature (°C)
Me
tab
olic a
ctivity le
ve
l (m
et)
0 5 10 15 20 25
0.0
0.2
0.4
0.6
0.8
1.0
1.0 met
1.2 met
1.6 met
62
Appliances: Principles
USA
France
Bottom-up stochastic model:
1. appliance allocation
2. activity-dependent switch-on
3. duration of use
4. varying demand whilst in use
63
Appliances: Principles
Bottom-up stochastic model:
1. appliance allocation
2. activity-dependent switch-on
3. duration of use
4. varying demand whilst in use
64
Appliances: Principles
Tij(P(t+1) = Ij | P(t) = Ii)
Bottom-up stochastic model:
1. appliance allocation
2. activity-dependent switch-on
3. duration of use
4. varying demand whilst in use
65
Towards smart grid simulation…
Micro-simulation: Appliance → Person → Vehicle → Building → City
66
Proposed MAS platform
1) Create synthetic
population of N agents
4) Long workplace absences
& destination
5) Presence chains (arrivals
/ departures)
6) Presence-dependent
activities 7) Activity location
8) Thermal / gaseous
emissions
11) Environmental
perception
S’(t)
Action(t)
9) Environmental solver
10) Behavioural actions on:
• appliances
• lights
• shading devices
• windows
• HVAC systems
• personal characteristics
S(t) Action(t)
Action(t)
2) Assign parameters to
each agent of population
3) Assign appliances to
household / workplace
-5 0 5 10 15 20 25
18
20
22
24
26
28
30
Running mean outdoor temperature (°C)
Re
po
rte
d c
om
fort
te
mp
era
ture
(°C
)
Regression line
CEN standard
Local regression
N W c C Wc WC Fc FC WFc WFC
20
22
24
26
28
30
Controls in use
Re
po
rte
d c
om
fort
te
mp
era
ture
(°C
)
1 2 3 4 5 6 7 9 11 16 28 19 25 27 30 33 37 39
Occupant
Re
po
rte
d c
om
fort
te
mp
era
ture
(°C
)
18
20
22
24
26
28
30
Adaptive algorithms v Comfort models
• Proposition: a linear model accounting for the effects of actions,
individuals and season: comf,ijk = + ai + bj + gout,rm + eijk
• Compared to the model with out,rm only, R2 increases from 0.226 to
0.574, with a strong decrease of g, the slope of out,rm.
• This also provides a measure of the comfort feedback of actions: ad
= in – S bi pi(in).
New adaptive comfort model
• In addition to comfort temperature, the
probability distribution of thermal
sensation completes the assessment of
indoor environment. This can be described
by an ordinal logistic model.
• Comfort probability is generally of more
direct interest. We model pcold and phot
through logistic functions which determines
pcomf = (1-pcold) (1-phot)
Indoor temperature (°C)
Th
erm
al se
nsa
tio
n
18 20 22 24 26 28
23
45
67
0.0
0.2
0.4
0.6
0.8
1.0
Cool
Slightly cool
Neutral
Slightly warm
Warm
Hot
18 20 22 24 26 28 300
.00
.20
.40
.60
.81
.0
Indoor temperature (°C)
Pro
po
rtio
n c
om
fort
ab
le
Probabilistic Comfort
Linking actions and comfort
Dynamic building
simulation program
Adaptive increments
Adaptive actions
(probabilistic /
stochastic models)
Sensation, comfort &
overheating probability
θi(t) θ´i(t)
(θ´i-Δθ)(t)
Action(t)
For constant charging of overheating
probability (tolerance discharge)
Overheating: capacitor analogy
JBPS 1(1), p43-55, 2008
Energy and Buildings, 40 p1240-1245.
ttOH DHtP ,041075.4exp1
ttOH DHtP ,041075.4exp1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5 10 15 20 25
Time
To
lera
nce
1,0T TPOH 1
72
Proposed MAS platform
1) Create synthetic
population of N agents
4) Long workplace absences
& destination
5) Presence chains (arrivals
/ departures)
6) Presence-dependent
activities 7) Activity location
8) Thermal / gaseous
emissions
g+1
t + 1?,
12) b + 1?,
g + 1?
t+1, b+1
11) Environmental
perception
S’(t)
Action(t)
9) Environmental solver
10) Behavioural actions on:
• appliances
• lights
• shading devices
• windows
• HVAC systems
• personal characteristics
S(t) Action(t)
Action(t)
2) Assign parameters to
each agent of population
3) Assign appliances to
household / workplace
Sta
tic (
new
rep
licat
e) o
r
dyna
mic
(ne
w a
ttrib
utes
)
73
From single results to distributions…
74
Proposed MAS platform
1) Create synthetic
population of N agents
4) Long workplace absences
& destination
5) Presence chains (arrivals
/ departures)
6) Presence-dependent
activities 7) Activity location
8) Thermal / gaseous
emissions
13) destroy synthetic
population of N agents
g+1
t + 1?,
12) b + 1?,
g + 1?
t+1, b+1
11) Environmental
perception
S’(t)
Action(t)
9) Environmental solver
10) Behavioural actions on:
• appliances
• lights
• shading devices
• windows
• HVAC systems
• personal characteristics
S(t) Action(t)
Action(t)
2) Assign parameters to
each agent of population
3) Assign appliances to
household / workplace
Further work…
MASS of activity & subsequent behaviour:
• Presence: long absences; calibration parameters: data
• Activity: location + pollutants: data
• Windows: IAQ stimuli; simplified airflow model; calibration parameters: data (Vienna+)
• Blinds: calibration parameters: data (Vienna+)
• Lighting: calibration parameters: data (residential!)
• Appliances: testing of core hypotheses / simplified activity-based model; calibration: data
• Comfort: coupling behaviour and dynamic human thermo-regulation modelling; more detailed adaptive model: data
• Oh, and did I mentioned we need more data ;-)
Some CitySim Applications...
A single building: Martigny
A district: Neuchâtel
440 buildings
4,700 residents
Raw data acquisition
Cadastral and LIDAR data: Visual survey:
Buildings / inhabitants register data; energy use data:
Data processing
• 2.5D model creation
• Data harmonisation
• Managing conflicts / errors
/ missing data
Input files for model
Climate file Geometry file (.kml) - Parameters files
Citysim model
Import tools
A clumsy first
solution for district
energy modelling:
Some first results
MEU: Integrated System Architecture
We need a DBMS that supports:
• Multi-user remote access
• Full 3D modelling
• Spatio-temporal data
management
• Interconnectivity
- Climate
- LIDAR
- Census
- Visual observations
A City: Zürich (2kW City)
75%
25%
40%
15% 5% 10% 20%
30% 60% 45%
Visual observations: glazing ratio
Q-GIS interface to PostgreSQL database
Kreis 3 – Alt-Wiedikon
Pre-renovation: old buildings dominate
Post-renovation: 8% total heating reduction
89
Conclusions
Urban energy simulation has advanced considerably:
- Core deterministic models
- Basic behavioural models
Decision modelling is the way forward:
-Investments and activities
-Activity-dependent behaviours
We still have a long way to go:
- Development & validation of a core MASS platform
- Coupling of complementary modules (climate, transport, space...)
But isn’t it exciting?
90
Insomnia cure (Earthscan, 2011)
91
Thank you!
