Small Area Temporal Household Electricity Demand Models for New Zealand: Why, how and how far have we got (update)?

Department of Geography Seminar, April 10th 2019: Dunedin: University of Otago

Transcript

1. Small Area Temporal Household Electricity Demand Models for New Zealand Ben Anderson b.anderson@soton.ac.uk / ben.anderson@otago.ac.nz @dataknut Why, how and how far have we got?
2. @dataknut The menu  The problem · Local demand peaks  The solution · Local demand models  Initial results · Observation based · Time-Use based  Where have we got to?  Where next? 2
3. @dataknut What’s the problem? 3 Total NZ electricity demand per half hour (June) Source: Electricity Authority GW (sum)
4. @dataknut Why is ‘peak’ a problem? • ‘Dirty’ energy (?) Carbon problems: Carbon problems: • Higher priced energy Cost problems: Cost problems: • PV & Wind Renewables mis-match Renewables mis-match • Inefficient use of resources; • ‘Local’ (LV network) overload; Infrastructure problems: Infrastructure problems: 4 Filling the trough Filling the trough Peak load Peak load Depends on hydro levels in Feb – April Khan et al (2018) 10.1016/j.jclepro.2018.02.309
5. @dataknut Why is ‘peak’ a problem? • ‘Dirty’ energy (?) Carbon problems: Carbon problems: • Higher priced energy Cost problems: Cost problems: • PV & Wind Renewables mis-match Renewables mis-match • Inefficient use of resources; • ‘Local’ (LV network) overload; Infrastructure problems: Infrastructure problems: 5 Filling the trough Filling the trough Peak load Peak load Depends on hydro levels in Feb – April Khan et al (2018) 10.1016/j.jclepro.2018.02.309 Net-Zero economy by 2050 ? -> ~$50 billion new investment under BAU
6. @dataknut What makes up peak demand? What makes up peak demand? What might be reduced? Who might respond? And what are the local network consequences? What to do? Storage Storage • Just reducing it per se Demand Reduction Demand Reduction • Shifting it somewhere else in time (or space and time) Demand Response Demand Response 6 Key Questions
7. @dataknut The local problem 7 Areas with more electric heating? Areas with larger households? Areas with more EVs? 1. Targeted interventions 2. Network investment decisions £££
8. @dataknut The menu  The problem · Local demand peaks  The solution · Local demand models  Initial results · Observation based · Time-Use based  Where have we got to?  Where next? 8
9. @dataknut Local demand models: Concept Synthetic Electricity Census Census data Household data (demand) 9 Source: http://datashine.org.uk • NZ examples: • Area Units • ~ 600 households • Meshblock areas • ~ 100 households
10. @dataknut Local demand models: Data Synthetic Electricity Census Census data Household data (demand) 10 Source: http://datashine.org.uk Household attributes (area level) Household attributes (area level) Bespoke kW monitoring? Bespoke kW monitoring? Household attributes Household attributes Trials: kW demand response? Trials: kW demand response? Time Use Survey Data? (imputed kW) Time Use Survey Data? (imputed kW) Smart meter kW? Smart meter kW?
11. @dataknut Conceptually… 11 AU 2 Survey households with ‘constraint’ variables + kW AU 1 Iterative Proportional Fitting Deming and Stephan 1940; Fienberg 1970; Wong 1992 Birkin & Clarke, 1989; Ballas et al, 1999 Ballas et al (2005) R package: ipfp Blocker (2016) Weights Census ‘constraint’ tables
12. @dataknut GREENGrid area unit model (v0.01a) 12 • Sample of ~ 30 monitored households • Hawke’s Bay & Taranaki Using NZ GREENGrid Data Using NZ GREENGrid Data • ~ 600 households per AU • For Hawke’s Bay & Taranaki At NZ Area Unit level At NZ Area Unit level • IPF re-weighting of survey cases (Ballas et al, 2005) Spatial Microsimulation Method Spatial Microsimulation Method
13. @dataknut Data: NZ GREENGrid 13 Get the data: https://dx.doi.org/10.5255/UKDA-SN-853334 • Circuits measured: • Hot water • Lighting • Heat pumps • Kitchen • Bedrooms • etc • Data: • Household survey • Mean power (W) per minute
14. @dataknut Data: NZ GREENGrid – Hot Water 14 Get the data: https://dx.doi.org/10.5255/UKDA-SN-853334 We want to estimate these for each unit area! VERY small n…
15. @dataknut Data: NZ GREENGrid – Heat Pumps 15 Get the data: https://dx.doi.org/10.5255/UKDA-SN-853334 We want to estimate these for each unit area! VERY small n…
16. @dataknut Data: NZ GREENGrid – Lighting 16 Get the data: https://dx.doi.org/10.5255/UKDA-SN-853334 We want to estimate these for each unit area! VERY small n…
17. @dataknut  Area Unit level · Hawke’s Bay · Taranaki  Variables used: · N rooms · N people · Presence children  Potential future variables: · Main heating fuel · Dwelling type · Income band · Age of adults/children Data: NZ Census 17 Matches GREENGrid sample ~ 90,000 households Some are not in GREENGrid data Because they correlate with demand
18. @dataknut Remember how this works… 18 AU 2 Survey households with ‘constraint’ variables + kW AU 1 Iterative Proportional Fitting Deming and Stephan 1940; Fienberg 1970; Wong 1992 Birkin & Clarke, 1989; Ballas et al, 1999 Ballas et al (2005) R package: ipfp Blocker (2016) Weights Census ‘constraint’ tables
19. @dataknut Remember how this works… 19 AU 2 Survey households with ‘constraint’ variables + kW AU 1 Iterative Proportional Fitting Deming and Stephan 1940; Fienberg 1970; Wong 1992 Birkin & Clarke, 1989; Ballas et al, 1999 Ballas et al (2005) R package: ipfp Blocker (2016) Weights Census ‘constraint’ tables GREEN Grid sample
20. @dataknut The consequence… 20 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], weighted), NZ Census 2013 small area tables [http://nzdotstat.stats.govt.nz/wbos/Index.aspx] We’re replicating a lot of households Each dot = 1 unit area so weird stuff can happen…
21. @dataknut But even so… 21 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], weighted), NZ Census 2013 small area tables [http://nzdotstat.stats.govt.nz/wbos/Index.aspx] Simulated household counts work OK Each dot = 1 unit area
22. @dataknut But even so… 22 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], weighted), NZ Census 2013 small area tables [http://nzdotstat.stats.govt.nz/wbos/Index.aspx] Simulated household counts in categories used as constraints work OK Each dot = 1 unit area
23. @dataknut And quite surprisingly… 23 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], weighted), NZ Census 2013 small area tables [http://nzdotstat.stats.govt.nz/wbos/Index.aspx] Simulated household counts in categories NOT used as constraints work quite well (sometimes) Each dot = 1 unit area
24. @dataknut The menu  The problem · Local demand peaks  The solution · Local demand models  Initial results · Observation based · Time-Use based  Where have we got to?  Where next? 24
25. @dataknut Example 1: Hot Water (Monday 25th May 2015) 25 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], unweighted] Small n… As measured (mean) Why this day? • Dry summer • No wind • Unusually cold • Gas pipeline broke • Avoid burning coal
26. @dataknut Example 1: Hot Water (Monday 25th May 2015) 26 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], unweighted) Real world heterogeneity As measured Households
27. @dataknut Example 1: Hot Water (Monday 25th May 2015) 27 As modelled Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], weighted), NZ Census 2013 small area tables [http://archive.stats.govt.nz/Census/2013-census/data-tables/meshblock-dataset.aspx] Each line = 1 area unit Shiftable demand? Areas with biggest potential to shift?
28. @dataknut Example 2: Hot water (seasonal) 28 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], unweighted) Real world heterogeneity As measured Small n… Why? • Avoid burning coal in dry years • Reduce local lines peaks (congestion)
29. @dataknut Example 2: Hot water (seasonal) 29 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], weighted), NZ Census 2013 small area tables [http://archive.stats.govt.nz/Census/2013-census/data-tables/meshblock-dataset.aspx] As modelled Each line = 1 area unit Shiftable demand? Met by solar PV? But is this ‘weird stuff’?
30. @dataknut Example 3: Lighting (seasonal) 30 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], unweighted) Real world heterogeneity As measured VERY small n… Why? • Avoid burning coal in dry years • Reduce local lines peaks (congestion)
31. @dataknut Example 3: Lighting (seasonal) 31 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334], weighted), NZ Census 2013 small area tables [http://archive.stats.govt.nz/Census/2013-census/data-tables/meshblock-dataset.aspx] Where might LEDs reduce demand? As modelled Each line = 1 area unit Highest lighting Lowest lighting
32. @dataknut Efficiency Model: Lighting 32 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334]  ‘Sheep-dip’ model – Take our 22 households – Replace non-LEDs with LEDs  Problems – Based on ‘most of your bulbs’ survey item – We seem to have more low-energy already · So our savings will be underestimates – We have VERY small numbers: · Only 2 Taranaki households in the lighting data!
33. @dataknut Efficiency Model: Lighting 33 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334] • Before spatialisation • As you’d expect: demand reduction is large Observed LED model But: We need 100* this
34. @dataknut Efficiency Model: Lighting 34 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334] • Before spatialisation • As you’d expect: demand reduction is large Observed LED model But: We need 100* this
35. @dataknut Efficiency Model: Lighting 35 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334] As you’d expect: demand reduction is large Observed LED model Before spatialisation
36. @dataknut Efficiency Model: Lighting 36 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334] After spatialisation: Power savings vary from place to place LED model Each line = 1 area unit Big savings here Smaller savings here
37. @dataknut Efficiency Model: Lighting 37 Source: Author’s calculations using NZ GREENGrid data [https://dx.doi.org/10.5255/UKDA-SN-853334] After spatialisation: Power savings vary from place to place LED model Each line = 1 area unit Big savings here Smaller savings here
38. @dataknut The menu  The problem · Local demand peaks  The solution · Local demand models  Initial results · Observation based · Time-Use based  Where have we got to?  Where next? 38
39. @dataknut  We have shown: – The method works… – But the GREENGrid data is insufficient – The results are probably garbage  We need to: – Gather better kW data – Impute demand from Time Use Surveys Where have we got to? 39 N * 100 Representative
40. @dataknut  We have shown: – The method works… – But the GREENGrid data is insufficient – The results are probably garbage  We need to: – Gather better kW data – Impute demand from Time Use Surveys Where have we got to? 40 N * 100 Representative
41. @dataknut Questions?  ben.anderson@otago.ac.nz  @dataknut  www.energy.soton.ac.uk/tag/spatialec – 2 year EU Global Fellowship @Otago CfS – 2017-2019  https://git.soton.ac.uk/ba1e12/spatialec  “This research was supported by a Marie Sklodowska-Curie Individual Global Fellowship within the H2020 European Framework Programme (2014 -2020) under grant agreement no. 700386.” 41 pixabay.com