Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Small Area Temporal Household Electricity Deman...

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

Ben Anderson

April 10, 2019
Tweet

More Decks by Ben Anderson

Other Decks in Research

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