Anderson, B (2014) Small Area Estimation as a tool for thinking about temporal and spatial variation in energy demand. Paper presented at AURIN/NATSEM Microsimulation Workshop, University of Melbourne, Thursday 4th December 2014
Seismic Method Estimate velocity from seismic data.pptx
Small Area Estimation as a tool for thinking about temporal and spatial variation in energy demand
1. Small Area Estimation as a tool for thinking about
temporal and spatial variation in energy demand
Dr Ben Anderson
Sustainable Energy Research Centre
University of Southampton
@dataknut
Aurin Microsimulation Symposium, December 2014
University of Melbourne
2. Small Area Estimates of Electricity Consumption
Contents
What & Why
How?
Results
– Overall consumption
– Consumption inequalities
– Temporal profiles
Conclusions & future Directions
@dataknut
2
3. Small Area Estimates of Electricity Consumption
Contents
What & Why
How?
Results
– Overall consumption
– Consumption inequalities
– Temporal profiles
Conclusions & future Directions
@dataknut
3
?
4. Small Area Estimates of Electricity Consumption
Digression: Geography
@dataknut
4
Southampton (UK)
5. Small Area Estimates of Electricity Consumption
Digression: What’s a small area?
@dataknut
5
In this case…
– English Lower Layer
Super Output Areas
– Census 200/2011
LSOAs
– c. 630 households each
– 148 in Southampton City
6. Small Area Estimates of Electricity Consumption
What & Why
Basically we want something for nothing
– Small area estimates of energy demand
– Without a bespoke energy census
Why?
– Infrastructure planning
– Energy efficiency intervention analysis
– Energy inequality analysis
– Politics!
@dataknut
6
7. Small Area Estimates of Electricity Consumption
What’s the problem?
Domestic electricity demand is
‘peaky’
Carbon problems:
Peak load can demand ‘dirty’
@dataknut
generation
Cost problems:
Peak generation is higher priced
energy
Infrastructure problems:
Local/national network ‘import’
overload on weekday evenings;
Local network ‘export’ overload at
mid-day on weekdays due to under-used
PV generation;
Inefficient use of resources (night-time
trough)
800
700
600
500
400
300
200
100
@dataknut 7
ANZSRAI 2014, Christchurch, New Zealand
UK Housing Energy Fact File
Graph 7a: HES average 24-hour electricity use profile for owner-occupied
homes, England 2010-11
Gas consumption
The amount of gas consumed in the UK varies dramatically between
households. The top 10% of households consume at least four times as
much gas as the bottom 10%.60 Modelling to predict nhouseholds’ e ergy
consumption – based on the property, household income and tenure – has
so far been able to explain less than 40% of this variation.
Gas use varies enormously from
household to household, and the
variation has more to do with
behaviour than how dwellings are
built.
0
00:00 02:00 04:00 06:00 08:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00
Heating
Water heating
Electric showers
Washing/drying
Cooking
Lighting
Cold appliances
ICT
Audiovisual
Other
Unknown
Watts
Filling the
trough
Peak load
8. Small Area Estimates of Electricity Consumption
Data issue:
Small area summaries exist
But they are aggregates
– Or averages
– Or estimates
– And they are annual values
We want a micro-level simulation model to assess the
socio-economic and spatial impact of
@dataknut
Price changes
Incentive changes
‘Efficiency’ interventions
Changes in ‘energy habits’ (appliances & practices)
Socio-demographic change
8
9. Small Area Estimates of Electricity Consumption
What can we do?
A bespoke energy census
– ££££££££££££££££££££
@dataknut
9
10. Small Area Estimates of Electricity Consumption
What can we do?
A bespoke energy census
– ££££££££££££££££££££
A large sample energy survey covering all
LSOAs
– ££££££££££
@dataknut
10
11. Small Area Estimates of Electricity Consumption
What can we do?
A bespoke energy census
– ££££££££££££££££££££
A large sample energy survey covering all LSOAs
– ££££££££££
Small Area Estimation
– Take existing area level data
– Take (ideally) an existing large n survey
– Combine £
@dataknut
11
12. Small Area Estimates of Electricity Consumption
Small Area Estimation
Econometric approaches
Income, income deprivation,
– Well known
– Multi-level Models
– Usually requires census microdata for anything other
@dataknut
than means
Re-weighting (and other) approaches
– Increasingly well known
– 'Spatial microsimulation'
– Does not require census microdata
12
income inequality
smoking prevalence,
obesity,
consumption expenditure,
CO2, water…
Innovation Network:
“Evaluating and improving small area
estimation methods”
http://eprints.ncrm.ac.uk/3210/
13. Small Area Estimates of Electricity Consumption
Contents
What & Why
How?
Results
– Overall consumption
– Consumption inequalities
– Temporal profiles
Conclusions & Future Directions
@dataknut
13
Estimation
14. Small Area Estimates of Electricity Consumption
Data
@dataknut
14
UK Living Costs and Food Survey 2008-2010
– Consumption proxies (reported energy
expenditure)
UK Time Use Survey 2001
– Appliance use proxies (modeled energy
demand)
Census 2001 (2011)
15. Small Area Estimates of Electricity Consumption
Conceptually…
@dataknut
15
LSOA census ‘constraint’ tables
LSOA 2.1
(Region2)
Survey data cases
LSOA 1.1
(Region1)
Iterative Proportional Fitting
Ballas et al (2005)
If Region = 2
Weights
If Region = 1
16. Small Area Estimates of Electricity Consumption
Key First Job:
Choose your constraints
@dataknut
16
Census data
Survey data
How?
– Selection via regression in micro data
– You may have little choice
17. Small Area Estimates of Electricity Consumption
‘Iterative Proportional Fitting’
@dataknut
17
Well known!
Deming and Stephan 1940
– Fienberg 1970; Wong 1992
A way of iteratively adjusting statistical tables
– To give known margins (row/column totals)
– ‘Raking’
In this case
– Create weights for each case so LSOA totals ‘fit’
constraints
– Weighting ‘down’
18. Small Area Estimates of Electricity Consumption
Contents
What & Why
How?
Results
– Overall consumption
– Consumption inequalities
– Temporal profiles
Conclusions & Future Directions
@dataknut
18
19. Small Area Estimates of Electricity Consumption
Key First Job:
The constraints
LSOA census ‘constraint’ tables
– Selected by stepwise regression
@dataknut
19
Expenditure Share of expenditure
Most important Number of persons Employment Status
Accommodation type Number of earners
Age of HRP Age of HRP
Employment Status Composition
Number of rooms
Number of children
Least important Ethnicity (non-white)
R sq 0.136 0.01
20. Small Area Estimates of Electricity Consumption
Internal Validation methods
@dataknut
20
Use of constraints to re-create the Census
tables
Difference = Absolute Error
– Total Absolute Error (TAE) = sum of all
errors
– Standardised AE = TAE/(n persons x n
constraint categories)
Smith et al:
– SAE of less than 20% and ideally less
than 10%
– in 90% of the areas is desirable.
Consumption Mean SAE p90
Ethnicity 2.18% 3.05%
Number of children 0.11% 0.22%
Number of rooms 0.05% 0.10%
Employment status
(HRP) 0.88% 1.22%
Age (HRP) 0.34% 0.75%
Tenure 0.07% 0.14%
Accomodation type 0.21% 0.51%
Number of persons 0.00% 0.00%
21. Small Area Estimates of Electricity Consumption
Preliminary results: Electricity
Mean weekly
household £
Modelled
Census 2001
LC&F Survey 2008-
2010
@dataknut
21
22. Small Area Estimates of Electricity Consumption
Validation: Electricity
Mean weekly
household £
Observed @LSOA
– DECC 2010
Spearman: 0.317
@dataknut
22
23. Small Area Estimates of Electricity Consumption
Preliminary results: Electricity
Total weekly household £
Modelled
Census 2001
LC&F Survey 2008-2010
@dataknut
23
24. Small Area Estimates of Electricity Consumption
Validation: Electricity
Total weekly household £
Observed @LSOA
– DECC 2010
Spearman: 0.509
@dataknut
24
25. Small Area Estimates of Electricity Consumption
What is causing the error?
Housing growth
Model overestimates
@dataknut
25
Model underestimates
26. Small Area Estimates of Electricity Consumption
What is causing the error?
Heating!
– 2011 data
Model overestimates
Combined:
– Heating = 60%
– Growth = 5%
@dataknut
26
Model underestimates
27. Small Area Estimates of Electricity Consumption
Consumption inequality
Area level gini
£ mean spend
R = -0.413
– (p < 0.001)
@dataknut
27
28. Small Area Estimates of Electricity Consumption
Consumption inequality
Area level gini
Index of Multiple
Deprivation 2010
– Income score
R = 0.463
– (p < 0.001)
@dataknut
28
29. Small Area Estimates of Electricity Consumption
My big worry
Data quality
@dataknut
29
Source: SPRG/ARCC-Water Survey, 2011
www.sprg.ac.uk
30. Small Area Estimates of Electricity Consumption
Conclusions
Outliers and errors are informative
Reported consumption data
– Could be dangerous
Census 2011 central heating
– Critical new constraint
@dataknut
30
31. Small Area Estimates of Electricity Consumption
Contents
What & Why
How?
Results
– Overall consumption
– Consumption inequalities
– Temporal profiles
Conclusions & Future Directions
@dataknut
31
32. Small Area Estimates of Electricity Consumption
What do we need?
Model:
– When do people do what at home?
– What energy demand does this generate?
– Scenarios for change
@dataknut
Appliance efficiency
Mode of provision
Changing practices
– What affect might this have for local areas?
32
33. Small Area Estimates of Electricity Consumption
How might this be done?
When do people do what at home?
@dataknut
Time Use Diaries
What energy demand does this generate?
Imputed electricity demand for each household
A microsimulation model of change
Ideally based on experimental/trial evidence
Or presumed appliance efficiency gains
Or ‘what if?’ scenarios of behaviour change
A way of estimating effects for local areas
Spatial microsimulation
33
UK ONS 2001
Time Use Survey
J Widén et al., 2009
doi:10.1016/j.enbui
ld.2009.02.013
Using UK Census
2001
34. Small Area Estimates of Electricity Consumption
When do people do what?
60%
50%
40%
30%
20%
10%
0%
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
@dataknut
60%
50%
40%
30%
20%
10%
Winter (November 2000 - February 2001)
% of respondents reporting a selection of energy-demanding activities
Source: Author’s calculations using UK Time Use Survey 2000/1 [http://discover.ukdataservice.ac.uk/catalogue/?sn=4504], weighted) 34
% respondents
Audio
TV
Reading
Computer
Ironing
Laundry
Cleaning
Dish washing
Cooking
Wash/dress self
Aged 25-64 who are in work
0%
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
% respondents
Audio
TV
Reading
Computer
Ironing
Laundry
Cleaning
Dish washing
Cooking
Wash/dress self
Aged 65+
35. Small Area Estimates of Electricity Consumption
Imputing electricity consumption
90
80
70
60
50
40
30
20
10
@dataknut
6.00%
5.00%
4.00%
3.00%
2.00%
1.00%
35
Imputation at individual level
– For each primary & secondary activity in each 10 minute time
slot
Then aggregated to household level
– Assume 100W for lighting if at home
– Max: Cooking, Dish Washing, Laundry
– Sum: everything else
Problems:
– Wash/dress might just be ‘dress’
– Hot water might be gas heated
– TVs might be watched ‘together’
– Not all food preparation = cooking and might be gas
– People have MANY more lights on!
– Several appliances may be ‘on’ but not recorded (Durand-
Daubin, 2013)
– No heating
=> a very simplistic ‘all electricity non-heat’ model!
J Widén et al., 2009
doi:10.1016/j.enbuild.2009.02.013
Assumes ‘shared’ use
Assumes ‘separate’ use
0.00%
0
0:00
1:00
2:00
3:00
4:00
5:00
6:00
7:00
8:00
9:00
10:00
11:00
12:00
13:00
14:00
15:00
16:00
17:00
18:00
19:00
20:00
21:00
22:00
23:00
% of recorded laundry
Mean watts per half hour 'washing/drying'
June ('work days', n = 76) June ('holidays', n = 74) summer laundry (ONS TU Survey 2005)
36. Small Area Estimates of Electricity Consumption
Results: Mean consumption I
2000
1800
1600
1400
1200
1000
800
600
400
200
@dataknut
36
2000
1800
1600
1400
1200
1000
800
600
400
200
Mean power consumption per half hour in winter (November 2000 - February 2001, all households)
Source: Author’s calculations using UK Time Use Survey 2000/1 [http://discover.ukdataservice.ac.uk/catalogue/?sn=4504], weighted) and Model 1
power assumptions
Age of household response person
0
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
Mean consumption per half houir (Watts)
0
1
2
3+
0
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
Mean consumption per half houir (Watts)
25-64
65+
Number of earners
37. Small Area Estimates of Electricity Consumption
Results: Mean consumption II
2000
1800
1600
1400
1200
1000
800
600
400
200
@dataknut
37
0
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
Mean consumption per half houir (Watts)
None
One
Two or more
2000
1800
1600
1400
1200
1000
800
600
400
200
0
0:00
2:00
4:00
6:00
8:00
10:00
12:00
14:00
16:00
18:00
20:00
22:00
Mean consumption per half houir (Watts)
married/partnered
single parent
single person
other
Number of children present Household composition
Mean power consumption per half hour in winter (November 2000 - February 2001, all households)
Source: Author’s calculations using UK Time Use Survey 2000/1 [http://discover.ukdataservice.ac.uk/catalogue/?sn=4504], weighted) and Model 1
power assumptions
38. Small Area Estimates of Electricity Consumption
But this is what the network sees…
1600000
1400000
1200000
1000000
800000
600000
400000
200000
@dataknut
38
1600000
1400000
Morning
1200000
1000000
spike too
800000
600000
spiky!
400000
200000
Sum of power consumption per half hour in winter (November 2000 - February 2001, all households)
Source: Author’s calculations using UK Time Use Survey 2000/1 [http://discover.ukdataservice.ac.uk/catalogue/?sn=4504], weighted) and Model 1
power assumptions
Number of earners
0
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
Sum of watts per half hour
3+ earners
2 earners
1 earner
0 earners
0
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
Sum of watts per half hour
HRP: 75+
HRP: 65-74
HRP: 55-64
HRP: 45-54
HRP: 35-44
HRP: 25-34
HRP: 16-24
Age of household response person
39. Small Area Estimates of Electricity Consumption
Microsimulation: But what if…?
@dataknut
39
We change the
washing
assumption?
=> an “all
electricity non-wash,
non-heat’
model!
40. Small Area Estimates of Electricity Consumption
Sum of power consumption per half hour in winter by number of earners (November 2000 - February 2001, all households)
@dataknut
1600000
1400000
1200000
800000
600000
400000
200000
0
0:00
1:00
2:00
3:00
4:00
5:00
6:00
7:00
8:00
9:00
10:00
11:00
12:00
13:00
14:00
15:00
16:00
17:00
18:00
19:00
20:00
21:00
22:00
23:00
3+
2
1
0
1000000
Sum of watts per half hour
earners
earners
earner
earners
Now the network sees..
40
Source: Author’s calculations using UK Time Use Survey 2000/1 [http://discover.ukdataservice.ac.uk/catalogue/?sn=4504], weighted) and Model 2
power assumptions
41. Small Area Estimates of Electricity Consumption
But that’s the big picture
@dataknut
41
We need a way to estimate these totals
– At small area level
Solution:
– Spatial microsimulation
IPF re-weighting of survey cases
– Using UK Census 2001
To match time use survey
– At UK Lower Layer Super Output Area level
c. 800-900 households
For Southampton (146 LSOAs)
42. Small Area Estimates of Electricity Consumption
Key First Job:
Choose your constraints
@dataknut
42
Census data
Survey data
How?
– Regression selection methods?
– Whatever is available!
43. Small Area Estimates of Electricity Consumption
Constraints used
@dataknut
43
Age of household response person (HRP)
Ethnicity of HRP
Number of earners
Number of children
Number of persons
Number of cars/vans
Household composition (couples/singles)
Presence of limiting long term illness
Accommodation type
Tenure
“Everything”
! Why?
No clear way to
select or
prioritise?
44. Small Area Estimates of Electricity Consumption
Results (Model 1)
140000
120000
100000
80000
60000
40000
20000
@dataknut
44
140000
120000
100000
80000
60000
40000
20000
Sum of half hourly power consumption (winter 2000/1) by number of earners
Source: Author’s calculations using UK Time Use Survey 2000/1 [http://discover.ukdataservice.ac.uk/catalogue/?sn=4504], weighted), UKL Census
2001 small area tables and Model 1 power assumptions
0
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
Sum of watts per half hour
3+
2
1
0
0
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
Sum of watts per half hour
3+
2
1
0
LSOA E01017139: highest % of households with 0
earners in Southampton
LSOA E01017180: lowest % of households with 0
earners in Southampton
45. Small Area Estimates of Electricity Consumption
Results (Model 1)
@dataknut
45
Sum of half hourly power consumption (winter 2000/1)
Source: Author’s calculations using UK Time Use Survey 2000/1 [http://discover.ukdataservice.ac.uk/catalogue/?sn=4504], weighted), UKL Census
2001 small area tables and Model 1 power assumptions. Map created in R (ggmap)
46. Small Area Estimates of Electricity Consumption
Results (Model 2)
140000
120000
100000
80000
60000
40000
20000
LSOA E01017139: highest % of households with 0
@dataknut
140000
120000
100000
80000
60000
40000
20000
LSOA E01017180: lowest % of households with 0
46
earners in Southampton
earners in Southampton
Sum of half hourly power consumption (winter 2000/1) by number of earners
Source: Author’s calculations using UK Time Use Survey 2000/1 [http://discover.ukdataservice.ac.uk/catalogue/?sn=4504], weighted), UKL Census
2001 small area tables and Model 2 power assumptions
0
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
Sum of watts per half hour
3+
2
1
0
0
0:00
1:30
3:00
4:30
6:00
7:30
9:00
10:30
12:00
13:30
15:00
16:30
18:00
19:30
21:00
22:30
Sum of watts per half hour
3+
2
1
0
47. Small Area Estimates of Electricity Consumption
Contents
What & Why
How?
Results
– Overall consumption
– Consumption inequalities
– Temporal profiles
Conclusions & Future Directions
@dataknut
47
48. Small Area Estimates of Electricity Consumption
Summary & Next Steps
@dataknut
48
It works!
– A temporal electricity demand spatial microsimulation
– But we don’t know how well
The model is over-simple
– But we knew that!
Constraint selection should be evidence based
– ?
And we need to update to 2011!!
– But no UK time use data since 2005
Next steps:
– “Solent Achieving Value through Efficiency” (SAVE) project
Large n RCT tests of demand response interventions
Linked time use & power monitoring
(Some) substation monitoring
=> evidence base for model development!
Validation against
observed substation
loads?
Implement more
complex model
(Widen et al, 2010)
or gather better data
Separate ½ hour
models??
49. Small Area Estimates of Electricity Consumption
Thank you
b.anderson@soton.ac.uk
This work has been supported by the UK Low Carbon Network Fund
(LCNF) Tier 2 Programme "Solent Achieving Value from Efficiency
(SAVE)” project:
– http://www.energy.soton.ac.uk/save-solent-achieving-value-from-efficiency/
STATA code (not the IPF bit):
– https://github.com/dataknut/SAVE
– GPL: V2 - http://choosealicense.com/licenses/gpl-2.0/ applies
@dataknut
@dataknut 49
ANZSRAI 2014, Christchurch, New Zealand
Editor's Notes
Data from SPRG linked water demand survey
2 implications:
Error in the estimates (spurious correlation with constraints)
Error in any policy microsimulation