You are on page 1of 58

ELECTRICITY PEAK DEMAND

FORECASTS
OVERVIEW OF OUR PEAK DEMAND FORECAST METHODLOGY

Transpower New Zealand Limited


September 2016
Table of Contents

1 Introduction ............................................................................................................... 4
2 Our approach............................................................................................................ 5
3 The Ensemble Models .............................................................................................. 9
4 Transpower’s peak demand ensemble forecast ...................................................... 15
5 Regional detail ........................................................................................................ 19
Appendix A: Industrial loads .............................................................................................. 22
Appendix B: New industrial loads ...................................................................................... 23
Appendix C: Embedded generators................................................................................... 25
Appendix D: Grid Exit Points by region .............................................................................. 26
Appendix E: Expected national and island peak demand forecast (MW) ........................... 31
Appendix F: Prudent national and island peak demand forecast (MW) .............................. 33
Appendix G: Expected regional peak demand forecast (MW) ............................................ 34
Appendix H: Prudent regional peak demand forecast (MW) .............................................. 37
Appendix I: Grid Exit Points by region ............................................................................... 40
Appendix J: Development of our forecasting methodology ................................................ 45
Appendix K: Functional specification of our modelling approach ....................................... 54
1 Introduction
This document describes the methodology that we use to derive peak electricity demand
forecasts, and summarises our most recent forecasts.

Peak demand forecasts are an important input into our planning process and are a key
driver of many of our new investment decisions. As peak demand grows it gets closer to
the capacity of the grid to deliver electricity. This can result in transmission constraints
occurring, which may result in the uneconomic dispatch of generation, or periods of
reduced security, where there are higher risks (and likely costs) associated with a power
outage. Therefore, it is important that these issues are identified with enough warning that
they can be investigated and, if appropriate, allow investment to occur to ensure efficient
delivery.

Forecasting peak demand (as with any forecasting) is challenging. The future is uncertain.
This has been particularly the case in recent years with demand growth slowing
significantly since 2007. With this in mind it is important to reflect this in the derivation of
our forecast. Our forecasting approach produces a distribution of possible future demand
levels. It is from this range of plausible future demand that we calculate an expected and
prudent (upper) estimate for our planning purposes.

Reflecting this uncertainty we use a range of models in constructing our forecasts. This
recognises that there is not one “perfect” model for producing a forecast of future demand.
Rather, a range of models are likely to provide different information that together can help
produce a better forecast overall.

The remainder of this document sets out our methodology in detail and summarises our
most recent results. Further technical information is contained in the Appendices and our
approach is illustrated in an Excel workbook published on our website1. .

We welcome feedback on our approach. If you wish to discuss or comment on our


approach please write to:

Mark Walkington
DemandForecasting@transpower.co.nz

1
The Excel illustrative model was first published in 2015 and it will shortly be updated to
reflect all of our latest model updates. It can be found at

http://www.transpower.co.nz/about-us/what-we-do/planning-future/planning-inputs

4
2 Our approach
We forecast national, island and region levels of peak demand (MW) using a “top-down”
approach which focuses on peaks at national, island and regional levels. Focusing directly
on national, island and regional peaks helps ensure our forecasts look reasonable at an
aggregate level with reference to historical peak levels of demand.

To derive GXP level forecasts we also utilise a “bottom-up” approach by forecasting


demand at each Grid Exit Point (GXP) independently.

A consolidation process is used to ensure we can marry our top-down and bottom-up
forecast.

Figure 1 provides an overview of our demand forecasting process.

Figure 1: Peak demand forecasting process

5
2.1 Ensemble approach
At the heart of our top-down approach is an ensemble of forecast models. We use four
different models:

1. a long-term endogenous trend model


2. a short-term endogenous trend model
3. an econometric model
4. a model based on the forecasts of the Ministry of Business, Innovation, and
Employment (MBIE) that is used in our island and national forecasts

Each model provides a different forecast of demand and plausible range for future
demand. This approach acknowledges that there is not one “perfect” model for forecasting
future levels of growth.

A variety of data sources are used by each of the models, including historical peak
demand, National Institute of Water and Atmospheric Research (NIWA) temperature data,
NZ GDP data, and New Zealand Institute of Economic Research (NZIER) GDP forecasts.

The four models used are detailed in Section 3.

2.2 Modelled demand


We use our ensemble of models to forecast “modelled demand”. Modelled demand is the
demand that is most likely to be driven by the econometric drivers (e.g. GDP) in line with
the historical relationship between these variables.2

We define “modelled demand” as grid offtake demand plus the contribution of embedded
generation to demand minus industrial demand. In this way the demand that is modelled
using our ensemble of models is that supplied by the grid and by embedded generation to
consumers, other than large industrial consumers. This is illustrated in the following
diagram.

2
Historical demand data is sourced directly from Energy Market Services. Additional information relating
to historical embedded generation is sourced from the Electricity Authority’s Electricity Market
Information (EMI) website at http://www.emi.ea.govt.nz/

6
Figure 2 Modelled Demand

Future industrial demand and embedded generation are forecast separately. These
forecasts are informed by the stated plans of the industrial participants and/or by making
“business as usual” assumptions. The forecast industrial demand is added, and forecast
embedded generation subtracted from, forecast modelled demand to derive a forecast of
total grid offtake.

As presented in Appendix E-H, all our peak demand forecasts are expressed in terms of
grid offtake.

2.3 Industrial demand forecasts


Industrial demand is defined as the demand from the directly connected grid customers
that are listed in Appendix A, and includes some of New Zealand’s largest industrial sites.

As indicated in Error! Reference source not found., industrial demand is deducted from
total demand to give modelled demand. We do this to recognise that demand growth for a
large industrial consumer may be significantly different to that of other sectors of the
economy so it makes sense to consider large industrial consumers individually.

We also forecast expected new industrial loads as we are made aware of them, such as
from our stakeholder discussions. The new industrial loads that we are aware of are listed
in Appendix B.

In New Zealand such large industrial loads have been concentrated in mining and metal
production, wood production and dairy production sectors. Investors in such plant always
have a range of variables and economic conditions that affect their business decisions and
they need to remain flexible in their decision making for as long as possible. As a result we
are typically informed of such plans with relatively little notice restricting the time for grid
planning. Even where new developments are well heralded the actual timing of investment

7
typically remains unconfirmed for as long as possible. This is true of a number of the
irrigation schemes currently proposed for Canterbury and South Canterbury. The inability
to foresee the scale and timing of these industrial loads remains one of the challenges in
grid planning.

2.4 Embedded generation forecasts


The embedded generation data has been sourced from the Electricity Authority’s
Electricity Market Information (EMI) website. The generators modelled are listed in
Appendix C. We model the larger and more significant existing embedded generators.

We forecast future levels of generation from the existing embedded generators based on
their historical contributions to national, island and regional peaks. In our expected
forecasts we model the average generation while in our prudent forecast we expect only
the minimum generation seen in recent years. For some generators there is considerable
variation year to year and this adds to the modelled variance seen between our expected
and prudent forecasts.

We do not attempt to forecast the growth of new embedded generation through this
approach. This is considered independently of this process when developing generation
scenarios.

2.5 Independent application to seasons and regions


Our approach is applied to forecast peak demand in different seasons and regions
independently to recognise that the drivers of demand growth could be different between
winter, summer and shoulder seasons and between regions3. For example, irrigation
growth in the South Island has had, and will continue to have, a significant impact on
summer demand growth as opposed to winter demand growth.

The regions we model are: Northland, Auckland, Waikato, Bay of Plenty, Hawkes Bay,
Taranaki, Central Districts and Wellington in the North Island and West Coast, Nelson-
Marlborough, Canterbury, South Canterbury and Otago-Southland in the South Island. Full
definitions of these regions can be found in Appendix D.

2.6 Trough forecasts


For grid planning purposes trough forecasts are also generated for each island and region
using growth rates based on our peak forecasts. These represent forecasts of the lowest
level of demand that we can expect on the grid. Trough demand forecasts are used to test
the ability of the network to export surplus generation from a particular region or to
manage voltage.

3
Seasons are defined as: Winter 10 May - 19 October; Shoulder 15 March - 9 May and 20 October - 30
November; Summer 1 December through to 14 March (i.e. of the calendar year following e.g. summer
2016 runs from 1 December 2016 to 14 March 2017)

8
2.7 GXP forecasts
As part of our overall process we also derive, independently from our top down forecasts,
a forecast for each of our GXPs. We use a variety of information to derive these forecasts
including:

 local knowledge from various sources including the relevant lines’ company, and
local industry
 lines’ company intentions of (permanent or temporary) feeder changes
 confirmed transfer of assets from Transpower to lines companies and other grid
developments
 forecasts of embedded generation and major industrial loads, and
 detailed history of the load.
We consult with lines companies in finalising these forecasts.

We also use this information, along with historical information about the contribution each
GXP makes to island and regional peak demand relative to its peak demand (commonly
referred to as “diversity”), to derive forecasts for each GXP’s contribution to the island and
regional peak forecasts derived using our top down approach.

3 The Ensemble Models


To forecast “modelled demand” we use an ensemble of four models:

1. Long-term endogenous model – regression of peak demand on year from 1997 to


the latest year
2. Short-term endogenous trend – regression of peak demand with AR(2) (auto-
regressive) error term.
3. Econometric model – regression of peak demand on GDP from 1997 to the current
year
4. MBIE-based model – based on MBIE EDGS Scenario energy forecasts.
This section includes a high level description of each individual model and the ensemble.
Additional information is provided in Appendix J and our approach is illustrated in an Excel
workbook published on our website4.

3.1 Model re-specification 2016


Prior to the development of the forecast presented here we received feedback from the
Ministry of Business Innovation and Employment (MBIE) and the New Zealand Institute of
Economic Research (NZIER). This feedback was sought after public consultation on the
draft Electricity Demand and Generation Scenarios (EDGS) suggested that MBIE should
take a role in validating Transpower’s peak demand forecast.

4
http://www.transpower.co.nz/about-us/what-we-do/planning-future/planning-inputs

9
MBIE asked NZIER to undertake a high-level review of Transpower’s forecasts and
methodology. NZIER’s memo5 stressed the value of simplicity and transparency in our
forecast methodology and suggested some changes to our approach. We have considered
their recommendations and have, in consultation with MBIE and NZIER, implemented a
number of changes aimed at increasing confidence in our forecasts. These changes have
not required fundamental changes in the model specifications and data requirements and
we have aimed to not overly complicate our approach.

The modelling changes are detailed in Appendix N and on MBIE’s EDGS website4.

In addition to some re-specification of the models we have

1. Now included peak demand from all data years; re-introducing years 2001 and
2003 These year’s were considered “dry years” where hydro storage shortages
triggered the need for associated conservation campaigns and demand restraint.
While it was a common practice to exclude these years we now find their inclusion
does not affect the model’s goodness of fit to the full time series of historical
demand peaks.
2. Removed the modelling of a break-point in the endogenous short-term model but
instead introduced an AR(2) (auto-regressive) error term. This model specification
effectively places more weight on more recent year’s data and it is seen as a more
consistent approach to reflect shorter term dynamics such as the period of low
growth in recent years.
3. All models other than the MBIE-based model are now differenced once in order to
obtain more stationary inputs.
4. We have tested log transformations for all models and found that it provides little
benefit in fit and results in unrealistically high growth-rates in some regions.
Therefore we have left the response and predictor models untransformed in all
models.
5. Each model can now also include a temperature variable when modelling the
winter season. A statistical information criteria (AIC) is used to determine if the
inclusion of temperature is significant enough to warrant an additional variable in
the model. If not, the temperature variable is discarded.

3.1.1 Temperature correction

As mentioned above a temperature variable is included in the regression models when it is


found to be statistically significant. While it is certainly the case that significant additional
electricity demand is driven by cooler temperatures in winter it remains a non-trivial task to
separate the temperature effect from other “noise” in the historical time series. We
experimented with how best to summarise the temperature measurements relevant to the

5
http://www.mbie.govt.nz/info-services/sectors-industries/energy/energy-data-
modelling/modelling/electricity-demand-and-generation-scenarios/edgs-2016

10
peak demands and found that the best results in terms of model fit, were gained using a
variable defined as the average of the midpoint between maximum and minimum
temperatures on the peak demand days. This variable helps capture the effect of periods
that are consistently cold where little warming occurs during the day.

3.2 Long-term endogenous trend model


The long-term endogenous trend model is based on straight-line (unweighted, linear, least-
squares) regression of annual peak data from 1997 to the end of the 2015. Annual
demand data is differenced and a temperature variable is also included in the regression
when it is found to be significant.

Figure 3 shows the forecast produced for national, winter peak demand. The expected
peak forecast continues the long term growth seen in peak demand between 1997 and
2015. The variation allowed for between the high and low bounds reflects the uncertainty
inherent in the forecast. In Figures 3-6 the high and low bounds represent the 5th and
95th percentiles of the forecast. Demand growth is expected to average 1.2% per annum
to 2030.

Figure 3: Winter peak forecast - Long-term endogenous trend model

3.3 Short-term endogenous trend model


During 2013, in response to continuing low levels of growth, we consulted a number of
interested stakeholders about their ongoing view of demand growth. Overall there was an
expectation that demand growth would be modest over the medium-term.

11
In response to this feedback we added a short-term trend model to our ensemble. It was
based on the assumption that shorter-term trends may be a better indicator of future levels
of demand growth than long-term trends. This model included a “knee-bend” fit allowing for
a distinct change of slope some time between 1997 and 2014.

We have now removed the modelling of such a break-point in the short-term endogenous
model and instead have introduced an AR(2) error term. The inclusion of this term has the
effect of placing more weight on recent years data and it is seen as a more consistent
approach to reflect shorter term dynamics.

The forecast produced by this model is shown in Figure 4 with average growth to 2030 of
1.0% per annum with the lower bound indicating that flat demand remains a possibility.

Figure 4, Winter peak forecast - Short-term endogenous trend model

3.4 Econometric forecast


The econometric model assumes the historical relationship between econometric drivers
(such as GDP) and peak demand remains relatively constant and forecasts demand
growth using forecasts of these drivers.

12
Figure 5: Winter peak forecast - Econometric model

The forecast produced by this model also sees expected growth at 1% per annum
somewhat slower than projected economic growth. This relative decoupling between
electricity demand and GDP has been seen to be an ongoing feature not just in New
Zealand and reflects developments in energy efficiency and re-balancing of the economy
with growth in less energy intensive sectors of the economy.

3.5 MBIE-based forecast


The MBIE-based forecast model is derived from the draft Electricity Demand and
Generation Scenarios (EDGS)6 electricity energy forecasts. We use the Mixed
Renewables scenario to provide a mean growth forecast and the High and Low Demand
sensitivity cases to provide a range around this. Average growth of 1.3% per annum was
expected.

This forecast model regresses peak demand on energy demand and projects this forward
using MBIE’s energy demand projections. Temperature is also made available to the
model as an explanatory variable for winter peak demand if it is statistically significant.

Figure 6 illustrates the results of the MBIE-based forecast. It is worth noting that both the
Econometric forecast and the MBIE models draw on similar data sources for their model
inputs of population and GDP expectations, so adding the MBIE model explicitly includes
MBIE’s view, but does not greatly differ from our econometric forecast.

6
http://www.med.govt.nz/sectors-industries/energy/energy-modelling/modelling/electricity-demand-and-
generation-scenarios At the time we undertook our modelling the EDGS had not been finalised.

13
Figure 6: Winter peak forecast – MBIE-based model

MBIE’s forecast is provided with a split between the North and South island. We use it in
our forecasts of peak national demand and at an aggregated island level. However, due
to the aggregated nature of this forecast we have not applied it at a regional level.

3.6 Uncertainty associated with each forecast


Each of the models in our ensemble is formulated to allow estimation of the level of
uncertainty expected in the future. Uncertainty is allowed for within:
 the estimated parameters of the regression
 the random residual associated with the regression

 forecasts of the variables, such as future forecasts of GDP and population


 with regard to the MBIE-based model, variations between the high and low demand
forecasts, and the peak to energy demand relationship

We quantify this uncertainty for each model to derive a distribution of forecast values for
each model.

3.7 Energy Efficiency


Implicitly we assume that future rates of energy efficiency improvements will be similar to
past efficiency improvements. In part, this is due to the difficulties associated with
measuring historical efficiency rates and forecasting future rates. A range of technology
and factors have historically contributed to improvements in energy efficiency, and this will
continue in the future. A recent significant example is the substitution to compact
fluorescent and now LED lighting. As these effects are embedded in the historical data we
use we have implicitly incorporated these improvements in energy efficiency into our
forecasts.

14
3.8 Other factors
We are mindful that there are other factors that may heavily affect grid offtake demand in
the future that are not explicitly modelled in the forecasts we present. Examples of such
influences include the uptake of energy efficient appliances, demand response
programmes, changes to customer tariff structures (including revisions to the Transmission
Pricing Methodology) and the uptake of distributed photovoltaics, battery storage and
electric vehicles.

We consider that these are best considered through scenario analysis given the high level
of uncertainty associated with the rate of uptake of some of these technologies.

In future, we intend to present forecasts that account for some of these factors. Our intent
at this stage is to develop forecasts using a two-stage process. The first stage would
forecast grid demand in a business as usual manner in a similar way to that described
above. In effect, we would assume that there is no rapid uptake of new distributed
technologies. The second stage would take the first stage forecast and estimate grid
demand with various levels of these factors, for example the uptake of new photovoltaic
distributed generation and battery storage.

This is similar to the approach we used in Transmission Tomorrow where we set out our
vision of New Zealand’s electricity industry over the next 5 – 40 years. In that document
four “what-if” scenarios are described which were chosen to test the extremes of demand
changes and technological impact. This diagram from Transmission Tomorrow7 describes
some of the potential impacts of this technology on demand.

4 Transpower’s peak demand ensemble forecast


4.1 The ensemble

7
https://www.transpower.co.nz/resources/transmission-tomorrow

15
The distributions of the outputs from the four forecast models are combined in equal
proportions to form the distribution of the ensemble. Iterative techniques are used to
determine percentiles (e.g. P90) for the ensemble based on the cumulative distribution
function derived for each model.

An advantage of our approach is that we are able to quantify a plausible range of future
peak demand levels based on the uncertainty associated with our models.8 We consider
that it important to recognise the future demand growth is highly uncertain and that a wide
range of future demand paths are possible. This also provides a basis from which to
compare different assumptions or scenarios for demand growth.

Figure 7 illustrates a plausible range of future peak modelled demand growth9. Our
approach indicates that average growth per annum could vary between 1.5% and -0.1% in
the period to 2030.

For practical purposes we derive two main forecasts from this range - an expected and
prudent forecast – that are defined as follows:

 Expected forecast – our expected forecast represents what we expect to see


happen, or the 50th percentile of the possible future level of peak demand growth.
Our current forecast expects demand to grow by 1.1% per annum.

 Prudent forecast – transmission upgrades may take 5-7 years to complete. To


ensure we identify issues (in all but the most extreme cases) with enough time to
complete upgrades we produce a prudent forecast. Our prudent forecast
represents the 90th percentile of possible future peak demand growth until 2021
and then is assumed to grow at the same rate as our expected demand forecast.
Our prudent forecast grows by 1.2% per annum.

8
Scenario analysis could also be used to determine a plausible range of future peak demand. The
disadvantage of a scenario approach is that the range is limited by what one considers plausible within
the scenario, and therefore may not capture the full range of uncertainty inherent in the forecast. We
consider both our approach and a scenario based approach have their uses. For example, the effect of
solar photovoltaics and electric vehicles we consider is best confronted using a scenario approach.
9
Assuming no major demand step change, such as the retirement of the Tiwai Aluminium Smelter

16
Figure 7: New Zealand Modelled Peak Demand - Ensemble Forecast

4.2 The ensemble models compared


The table below compares how the four models in the ensemble project future peak
modelled demand. The Econometric and Short-term model predict somewhat lower
expected growth than the other two models.
Table 1 New Zealand Peak Modelled Demand – forecast per annum growth rates 2016-2030

MBIE-
Long-term Short-term Econometric Ensemble
based
Expected growth 1.2% 1.0% 1.0% 1.3% 1.2%

Prudent growth 1.5% 1.3% 1.3% 1.4% 1.4%

High bound growth 2.0% 1.7% 1.9% 1.4% 1.7%

Low bound growth 0.2% 0.1% -0.2% 1.2% 0.1%

The Ensemble forecast has expected growth to 2030 of 1.2% but allows for growth ranging
from a low of just 0.1% per annum up to a high growth rate of 1.7% per annum.

4.3 Peak Grid Offtake forecasts


As discussed in sections 3.3 and 3.4 independent forecasts are made of future industrial
demand and of generation from the existing embedded generators. We add the industrial
demand forecast onto the ensemble forecasts of peak modelled demand and then subtract
the forecast of future embedded generation. This gives a forecast of peak grid offtake.

17
This forecast of national peak grid offtake has expected average growth of 1.1% per
annum to 2030. There is little expectation of growth in industrial demand and thus the
forecast growth rate in grid offtake is lower than that forecast for modelled demand.

Table 2 New Zealand Peak Grid Offtake – forecast per annum growth rates 2016-2030

National
Peak Demand
Forecast
Expected growth 1.1%

Prudent growth 1.3%

High bound growth 1.6%

Low bound growth 0.2%

Tables of our national and island peak grid offtake forecasts are provided in Appendix E
and Appendix F.

4.4 Comparison with previous forecasts

Figure 8 compares this national grid demand forecast with previous expected peak
forecasts made in the Statement of Opportunities in 201010 and Transpower’s Annual
Planning Reports (APR) 2012-14, Transmission Planning Report 2015 (TPR 2015) and
this most recent update TP 2016. The figure shows that our latest prudent forecast is

10
The SoO (Statement of Opportunities) was published by the Electricity Commission as part of its duties
in facilitating industry development.

18
again lower than all the previous forecasts which reflects the ongoing low demand growth,
and also some impact from the various modelling changes

Figure 8: Historical comparison of expected peak forecasts

5 Regional detail
The figures on the next two pages illustrate the regional growth rates – both expected and
prudent.

In the South Island, South Canterbury has the greatest growth expectations with strong
dairy and irrigation demand fuelling growth in this already summer peaking area.
Canterbury and Nelson-Marlborough also have above average growth expectations while
Otago-Southland has lower overall growth as its demand is dominated by the Tiwai Point
Aluminium Smelter at Bluff. The forecasts presented here make no assumption about any
shutdown of the smelter.

In the North Island the highest growth is expected in the Waikato and Auckland followed
by Northland. This reflects the higher levels of population and GDP growth expected in
these regions. Growth is expected to be modest in Central Districts and Hawkes Bay.

Note that we show the growth rates foreseen in our prudent and expected forecasts, and
also highlight the change in these rates (figures in brackets) from that seen in our last
published forecast from our Transmission Planning Report 2015.

19
Tables of our regional forecasts are provided in Appendices G and H.

Figure 9: North Island Regional Growth Forecasts, 2016-2030

20
Figure 10: South Island Regional Growth Forecasts, 2016-2030

21
Appendix A: Industrial loads
The following industrial loads are modelled specifically and not included in our definition of modelled demand as explained in Section 2.2.

Peak Modelled
Industrial load Region Type GXP Demand Demand 2015
2015 (MW) (MW)
Refining NZ Northland Non Connected Customer (Bream Bay) 46.0
New Zealand Steel Limited Auckland Direct Connect Glenbrook 128
Origin Energy Resources (Kupe) Limited Taranaki Direct Connect Hawera (Kupe) 9.2
Norske-Skog Tasman Bay of Plenty Direct Connect Kawerau 60.9
Carter Holt Harvey Waikato Direct Connect Kinleith 78.5
Pacific Steel Auckland Non Connected Customer (Mangere 110kV) 46.8
Methanex New Zealand Limited Taranaki Direct Connect Motunui 9.1
Winstone Pulp International Limited Central Districts Direct Connect Tangiwai 36.2
New Zealand Aluminium Smelters Limited Otago Southland Direct Connect Tiwai 586.1
Pan Pacific Forest Products Limited Hawkes Bay Direct Connect Whirinaki 11kV 76.7
Daiken New Zealand Canterbury Direct Connect Ashley 10.6
Dongwha Patinna Otago Southland Direct Connect Brydone 8.8
Fonterra Lichfield Waikato Non Connected Customer (Lichfield) 4.5
Fonterra Todd Cogeneration Joint Venture Taranaki Direct Connect Hawera 110kV 4.9
Kiwirail Hamilton Waikato Direct Connect Hamilton 6.5
Kiwirail Tangiwai Central Districts Direct Connect Tangiwai 8.7
Kiwirail Taumarunui Central Districts Direct Connect Taumarunui 7.3
Kiwirail Bunnythorpe Central Districts Direct Connect Bunnythorpe 6.1

22
Appendix B: New industrial loads
The table below lists known and likely new industrial loads that are captured in our
forecasting. The majority of these loads are now associated with the expansion of the
dairy industry and include loads at new dairy plants and at new irrigation schemes.
These are chiefly in the southern regions of Canterbury and South Canterbury. Step
increases at Auckland GXPs largely reflect plans for new housing developments as
specified in the Unitary Plan.

New Load - Earliest


GXP Region MW peak
category year

Industrial PEN Auckland 2017 5


Housing TAK Auckland 2018 3
Commercial PEN-110 Auckland 2017 10
Commercial PEN-110 Auckland 2018 10
Industrial PEN Auckland 2017 10
Industrial MNG Auckland 2016 -25
2016 to
Housing BOB Auckland 6
2028
2029 to
Housing BOB Auckland 4
2042
# Waikato 2016
# Waikato 2016
Dairy MGM Central_Districts 2016 6
Dairy WGN Central_Districts 2016 2
Dairy WGN Central_Districts 2017 1
Irrigation HOR Canterbury 2017 1
Irrigation HOR Canterbury 2017 3.3
Irrigation HOR Canterbury 2017 2.8
Irrigation HOR Canterbury 2017 -2
Irrigation HOR Canterbury 2018 -4
Irrigation HOR Canterbury 2019 -3
Irrigation HOR Canterbury 2020 -3
Dairy HOR Canterbury 2017 2.5
Irrigation HOR Canterbury 2016 1
Irrigation HOR Canterbury 2017 -2
Irrigation HOR Canterbury 2018 -4
Irrigation HOR Canterbury 2019 -4
Irrigation HOR Canterbury 2020 -3
Industrial ISL Canterbury 2017 4
Commercial ISL Canterbury 2016 2
Commercial ISL Canterbury 2017 2
Dairy ISL Canterbury 2018 3
Dairy ISL Canterbury 2020 3

23
Dairy ISL Canterbury 2022 3
Irrigation KBY Canterbury 2017 2.8
Dairy KBY Canterbury 2018 4
Dairy BPD South_Canterbury 2017 1.5
Dairy BPD South_Canterbury 2018 2.5
Dairy BPD South_Canterbury 2019 1.5
Irrigation BPD South_Canterbury 2016 0.6
Irrigation BPD South_Canterbury 2017 1
Irrigation BPD South_Canterbury 2017 3
Irrigation SAW South_Canterbury 2020 12
Irrigation SAW South_Canterbury 2021 17
# STU South_Canterbury 2018
# TMK South_Canterbury 2018
Commercial TKA South_Canterbury 2016 0.5
Industrial TIM South_Canterbury 2016 7
Commercial TIM South_Canterbury 2016 1
Commercial TIM South_Canterbury 2017 1
Irrigation TWZ South_Canterbury 2016 4
Farming TWZ South_Canterbury 2016 1
Irrigation WTK South_Canterbury 2016 2
Irrigation BPT South_Canterbury 2016 2
Irrigation BPT South_Canterbury 2016 2
Commercial CLH West_Coast 2018 0.2
Commercial CLH West_Coast 2018 5
Dairy DOB West_Coast 2017 0.4
Dairy DOB West_Coast 2017 0.6
Dairy HKK West_Coast 2017 0.3
Dairy HKK West_Coast 2018 0.6
Dairy HKK West_Coast 2019 0.1
Dairy HKK West_Coast 2020 0.3
Commercial OTI West_Coast 2020 0.8
Commercial OTI West_Coast 2021 0.4
Commercial OTI West_Coast 2020 0.8
Commercial OTI West_Coast 2021 0.4
Mining RFN West_Coast 2016 -4.5
Mining RFN West_Coast 2017 -0.5
Industrial WPT West_Coast 2016 -8

# - some detail removed for confidentiality reasons

24
Appendix C: Embedded generators
Below are the embedded generators explicitly modelled in our forecasts.

Embedded Capacity
Connects at node Type Region
Generator (MW)

Ngawha Kaikohe geothermal 25 Northland


Glenbrook Kilns Glenbrook - NZ Steel co-gen 65 Auckland
Te Rapa Te Kowhai co-gen 30 Waikato
Kaimai Tauranga 33kV hydro 36 Bay of Plenty
Aniwhenua Matahina hydro 25 Bay of Plenty
Mangahewa Huirangi thermal 4 Taranaki
Rotokawa Wairakei geothermal 35 Central Districts
Tararua Wind BPE Bunnythorpe 33kV wind 36.3 Central Districts
Tararua Wind LTN Linton wind 31.7 Central Districts
Te Uku Te Kowhai wind 64 Central Districts
Te Huka Wairakei geothermal 23 Central Districts
Te Rere Hau Linton wind 48.5 Central Districts
Mill Creek Wilton wind 60 Wellington
Highbank Ashburton 66 hydro 24 Canterbury
White Hill North Makarewa wind 58 Otago-Southland
Waipori Halfway Bush -1 hydro 31 Otago-Southland
Paerau Naseby hydro 12 Otago-Southland
Mahinerangi Halfway Bush -1 wind 36 Otago-Southland
Cobb Stoke 66kV hydro 32 Nelson-Marlborough

25
Appendix D: Grid Exit Points by region
APR GXP label
Island Region APR GXP name
(unique)
N Auckland Albany 110 (Wairau Rd) ALB-110
N Auckland Albany 33kV ALB-33
N Auckland Bombay 110kV BOB-110
N Auckland Bombay 33kV BOB-33
N Auckland Glenbrook - Counties GLN-33-2-COUP
N Auckland Glenbrook - NZ Steel GLN-33-2-NZST
Glenbrook - NZ Steel load
N Auckland GLN-33-1
only
N Auckland Henderson HEN
N Auckland Hepburn Rd HEP
N Auckland Hobson HOB
N Auckland Liverpool Street LST
N Auckland Mangere 110kV MNG-110
N Auckland Mangere 33kV MNG-33
N Auckland Meremere MER
N Auckland Mt Roskill 110kV - KING ROS-KING
N Auckland Mt Roskill 22kV ROS-22
N Auckland Otahuhu OTA
N Auckland Pakuranga PAK
N Auckland Penrose 110kV PEN-110
N Auckland Penrose 22kV PEN-22
N Auckland Penrose 25kV PEN-25
N Auckland Penrose 33kV PEN-33
N Auckland Silverdale SVL
N Auckland Takanini TAK
N Auckland Wairau Road WRD
N Auckland Wiri WIR
N Bay of Plenty Edgecumbe EDG
N Bay of Plenty Kaitimako KMO
N Bay of Plenty Kawerau Horizon KAW-11-T1-T2
N Bay of Plenty Kawerau T11 and T14 KAW-11-T11-T14
N Bay of Plenty Kawerau T6 - T9 KAW-11-T6-T7-T8-T9
N Bay of Plenty Matahina MAT
N Bay of Plenty Mt Maunganui 33kV MTM-33
N Bay of Plenty Owhata OWH
N Bay of Plenty Rotorua 11kV ROT-11
N Bay of Plenty Rotorua 33kV ROT-33
N Bay of Plenty Tarukenga 11kV TRK
N Bay of Plenty Tauranga 11kV TGA-11
N Bay of Plenty Tauranga 33kV TGA-33

26
N Bay of Plenty Te Kaha TKH
N Bay of Plenty Te Matai TMI
N Bay of Plenty Waiotahi WAI
N Central Districts Bunnythorpe 33kV BPE-33
N Central Districts Bunnythorpe NZR BPE-55
N Central Districts Dannevirke DVK
N Central Districts Linton LTN
N Central Districts Mangahao MHO
N Central Districts Mangamaire MGM
N Central Districts Marton MTN
N Central Districts Mataroa MTR
N Central Districts National Park NPK
N Central Districts Ohakune OKN
N Central Districts Ongarue ONG
N Central Districts Tangiwai 11kV TNG-11
N Central Districts Tangiwai NZR TNG-55
N Central Districts Tokaanu TKU
N Central Districts Waipawa WPW
N Central Districts Wairakei WRK
N Central Districts Woodville WDV
N Hawkes_Bay Fernhill FHL
N Hawkes_Bay Redclyffe RDF
N Hawkes_Bay Tuai 110kV TUI-110
N Hawkes_Bay Whakatu WTU
N Hawkes_Bay Whirinaki 11 kV Bus A and B WHI
N Northland Bream Bay BRB
N Northland Kaikohe 110kV KOE-110
N Northland Kaitaia KTA
N Northland Kensington KEN
N Northland Maungatapere MPE
N Northland Maungaturoto MTO
N Northland Wellsford WEL
N Taranaki Brunswick BRK
N Taranaki Carrington St 33kV CST-33
N Taranaki Hawera HWA-33-1
N Taranaki Hawera (Kupe) HWA-33-2
N Taranaki Hawera 110 kV -1 HWA-110-1
N Taranaki Hawera 110 kV -2 HWA-110-2
N Taranaki Huirangi HUI
N Taranaki Motunui MNI
N Taranaki Moturoa MRA
N Taranaki Opunake OPK

27
N Taranaki Stratford 33kV SFD-33
N Taranaki Taumarunui NZR TMN
N Taranaki Wanganui WGN
N Taranaki Waverley WVY
N Waikato Cambridge CBG
N Waikato Hamilton 11kV HAM-11
N Waikato Hamilton 33kV HAM-33
N Waikato Hamilton NZR HAM-55
N Waikato Hangatiki HTI
N Waikato Hinuera HIN
N Waikato Huntly HLY
N Waikato Kinleith 11kV T1 through T3 KIN-11-T1-T2-T3
N Waikato Kinleith 11kV T5 KIN-11-T5
N Waikato Kinleith 33kV KIN-33
N Waikato Kopu KPU
N Waikato Lichfield LFD
N Waikato Piako PAO
N Waikato Putaruru PUT
N Waikato Te Awamutu TMU
N Waikato Te Kowhai TWH
N Waikato Waihou WHU
N Waikato Waikino WKO
N Wellington Central Park 11kV CPK-11
N Wellington Central Park 33kV CPK-33
N Wellington Gracefield GFD
N Wellington Greytown GYT
N Wellington Haywards 11kV HAY-11
N Wellington Haywards 33kV HAY-33
N Wellington Kaiwharawhara KWA
N Wellington Masterton MST
N Wellington Melling 11kV MLG-11
N Wellington Melling 33kV MLG-33
N Wellington Paraparaumu PRM
N Wellington Pauatahanui PNI
N Wellington Takapu Rd TKR
N Wellington Upper Hutt UHT
N Wellington Wilton WIL
S Canterbury Addington 11kV ADD-11
S Canterbury Addington 66kV ADD-66
S Canterbury Ashburton 33 ASB-33
S Canterbury Ashburton 66 ASB-66
S Canterbury Ashey 11kV Daikon ASY-11-MPAS

28
S Canterbury Ashley 11kV Main Power ASY-11-MPOW
S Canterbury Bromley 11kV BRY-11
S Canterbury Bromley 66kV BRY-66
S Canterbury Coleridge COL
S Canterbury Culverden CUL-33
S Canterbury Culverden CUL-66
S Canterbury Hororata 33kV HOR-33
S Canterbury Hororata 66kv HOR-66
S Canterbury Islington 33kV ISL-33
S Canterbury Islington 66kV ISL-66
S Canterbury Kaiapoi KAI
S Canterbury Kimberley KBY
S Canterbury Southbrook 33kV SBK-33
S Canterbury Southbrook 66kV SBK-66
S Canterbury Waipara 33kV WPR-33
S Canterbury Waipara 66kV WPR-66
S Nelson-Marlborough Blenheim BLN
S Nelson-Marlborough Stoke 33kV STK-33
S Nelson-Marlborough Stoke 66kV STK-66
S Otago-Southland Balclutha BAL
S Otago-Southland Brydone - Rayonier BDE-11-RAYN
S Otago-Southland Brydone - Solid Energy BDE-11-SOLE
S Otago-Southland Clyde CYD
S Otago-Southland Cromwell CML
S Otago-Southland Edendale EDN
S Otago-Southland Frankton FKN
S Otago-Southland Gore GOR
S Otago-Southland Halfway Bush - Palmerston HWB-110
S Otago-Southland Halfway Bush -1 HWB-33-1
S Otago-Southland Halfway Bush -2 HWB-33-2
S Otago-Southland Invercargill INV
S Otago-Southland Naseby NSY
S Otago-Southland North Makarewa NMA
S Otago-Southland South Dunedin SDN
S Otago-Southland Tiwai TWI
S South Canterbury Albury ABY
S South Canterbury Bells Pond BPD
S South Canterbury Blackpoint BPT
S South Canterbury Oamaru OAM
S South Canterbury Studholme STU
S South Canterbury Tekapo A TKA
S South Canterbury Temuka 33kV TMK-33

29
S South Canterbury Timaru TIM
S South Canterbury Twizel TWZ
S South Canterbury Waitaki WTK
S West Coast Arthur's Pass APS
S West Coast Atarau ATU
S West Coast Castle Hill CLH
S West Coast Dobson DOB
S West Coast Greymouth GYM
S West Coast Hokitika HKK
S West Coast Kikiwa KIK
S West Coast Kumara KUM
S West Coast Murchison MCH
S West Coast Orowaiti 110kV 1 and 2 ORO
S West Coast Otira OTI
S West Coast Reefton 110kV 1 and 2 RFN
S West Coast Westport WPT

30
Appendix E: Expected national and island peak demand
forecast (MW)

Winter Shoulder Summer


North South New North South New North South New
Island Island Zealand Island Island Zealand Island Island Zealand
1997 3728 1815 5514 3362 1710 5071 2964 1593 4516
1998 3529 1819 5339 3331 1752 5082 2937 1607 4501
1999 3682 1880 5545 3477 1817 5243 3058 1696 4679
2000 3661 1896 5558 3430 1814 5212 3071 1671 4739
2001 3858 1954 5737 3437 1827 5230 3149 1728 4864
2002 3818 1985 5773 3580 1863 5349 3179 1812 4867
2003 3807 1926 5663 3370 1810 5130 3272 1793 5050
2004 4059 2008 6012 3620 1972 5592 3286 1843 5129
2005 4005 2054 5980 3689 2006 5648 3328 1899 5176
2006 4241 2098 6334 3718 1955 5576 3286 1882 5123
2007 4293 2156 6414 3770 2044 5764 3358 1927 5238
2008 4138 2058 6161 3803 2101 5869 3334 1788 5028
2009 4336 1999 6335 3827 1974 5680 3275 1973 5240
2010 4197 2071 6261 3764 1976 5718 3230 1985 5120
2011 4498 2105 6584 3707 1981 5752 3193 1874 5040
2012 4095 2041 6099 3840 1989 5812 3210 1908 4991
2013 4205 2028 6181 3696 1978 5607 3226 1949 5158
2014 4073 1985 6055 3512 1925 5393 3188 1968 5139
2015 4108 2056 6088 3688 2010 5663 3255 1941 5092

2016 4276 2054 6275 3676 2026 5685 3306 2004 5258
2017 4313 2080 6388 3716 2065 5771 3354 2046 5314
2018 4362 2104 6464 3757 2120 5860 3386 2100 5403
2019 4414 2123 6544 3787 2136 5911 3418 2120 5452
2020 4463 2143 6616 3817 2153 5962 3447 2142 5503
2021 4508 2161 6691 3849 2173 6015 3475 2166 5554
2022 4554 2178 6759 3881 2193 6066 3503 2191 5605
2023 4602 2196 6831 3914 2211 6120 3533 2215 5656
2024 4649 2213 6901 3945 2229 6167 3560 2237 5705
2025 4697 2230 6972 3973 2247 6218 3588 2260 5753
2026 4743 2247 7045 4003 2264 6264 3617 2283 5803
2027 4791 2265 7114 4036 2282 6315 3645 2307 5853
2028 4840 2281 7185 4066 2300 6365 3673 2330 5903
2029 4889 2298 7256 4093 2316 6410 3701 2353 5952
2030 4940 2315 7329 4125 2334 6460 3730 2376 6002

31
2031 4988 2332 7402 4157 2351 6510 3759 2400 6056
2032 5039 2349 7473 4187 2370 6558 3788 2425 6105
2033 5089 2366 7546 4218 2388 6605 3816 2448 6157
2034 5140 2384 7624 4251 2406 6659 3846 2472 6210
2035 5190 2401 7696 4280 2424 6706 3876 2497 6261
2036 5242 2419 7769 4309 2443 6755 3905 2522 6315
2037 5294 2437 7846 4339 2462 6804 3934 2546 6369
2038 5346 2455 7921 4370 2480 6854 3963 2571 6421
2039 5398 2471 7996 4402 2499 6904 3993 2596 6472
2040 5452 2490 8074 4434 2518 6959 4023 2621 6525
2041 5506 2508 8144 4464 2538 7006 4052 2648 6579
2042 5558 2526 8217 4494 2556 7056 4082 2672 6635
2043 5611 2545 8296 4527 2575 7106 4112 2697 6690
2044 5665 2563 8372 4559 2594 7151 4142 2723 6744
2045 5717 2581 8449 4589 2614 7208 4171 2748 6799
2046 5772 2600 8530 4618 2633 7259 4201 2774 6856
2047 5826 2618 8607 4649 2650 7309 4232 2800 6910
2048 5880 2636 8682 4682 2673 7360 4262 2825 6964
2049 5936 2655 8759 4710 2691 7414 4290 2852 7018
2050 5991 2673 8837 4741 2711 7466 4322 2878 7076

32
Appendix F: Prudent national and island peak demand
forecast (MW)

Winter Shoulder Summer


North South New North South New North South New
Island Island Zealand Island Island Zealand Island Island Zealand
2016 4650 2189 6616 4132 2176 6156 3615 2125 5554
2017 4762 2230 6778 4239 2251 6307 3715 2200 5652
2018 4871 2275 6896 4342 2338 6458 3799 2284 5778
2019 4970 2309 7005 4430 2382 6555 3876 2329 5864
2020 5061 2344 7106 4511 2423 6651 3943 2374 5948
2021 5145 2376 7203 4584 2463 6740 4008 2416 6025
2022 5229 2408 7295 4657 2504 6826 4066 2459 6100
2023 5280 2426 7375 4694 2525 6884 4099 2487 6157
2024 5334 2444 7447 4728 2547 6934 4130 2512 6208
2025 5386 2461 7522 4762 2567 6989 4160 2539 6261
2026 5439 2479 7598 4798 2588 7045 4192 2565 6314
2027 5494 2496 7676 4832 2609 7097 4223 2592 6368
2028 5547 2514 7751 4864 2628 7152 4254 2619 6420
2029 5600 2530 7830 4896 2648 7200 4286 2646 6473
2030 5655 2548 7908 4932 2670 7256 4318 2672 6529
2031 5710 2566 7986 4969 2691 7310 4350 2700 6581
2032 5766 2583 8063 5000 2711 7361 4379 2725 6634
2033 5821 2600 8145 5033 2731 7415 4413 2752 6691
2034 5878 2618 8221 5069 2752 7472 4446 2779 6747
2035 5937 2636 8296 5105 2775 7524 4479 2807 6804
2036 5995 2654 8379 5140 2797 7579 4509 2835 6860
2037 6051 2673 8458 5178 2817 7635 4544 2864 6914
2038 6109 2691 8536 5212 2839 7689 4576 2892 6970
2039 6167 2709 8615 5247 2860 7746 4609 2919 7030
2040 6225 2727 8700 5285 2882 7804 4642 2948 7085
2041 6287 2745 8782 5318 2904 7859 4675 2977 7145
2042 6345 2763 8863 5352 2927 7913 4707 3005 7200
2043 6404 2782 8946 5386 2949 7966 4741 3034 7258
2044 6464 2801 9029 5424 2970 8021 4775 3061 7321
2045 6525 2819 9111 5455 2993 8078 4807 3091 7379
2046 6585 2837 9191 5490 3015 8131 4840 3121 7437
2047 6646 2856 9277 5524 3038 8187 4874 3149 7496
2048 6707 2875 9361 5559 3061 8242 4905 3178 7556
2049 6767 2894 9444 5595 3084 8299 4940 3207 7615
2050 6831 2914 9531 5630 3107 8356 4973 3237 7677

33
Appendix G: Expected regional peak demand forecast (MW)
Bay of Central Hawkes Nelson Otago South
Region Auckland Northland Taranaki Waikato Wellington Canterbury West Coast
Plenty Districts Bay Marlborough Southland Canterbury

1997 1539.9 381.1 302.2 267.4 193 156.3 451.9 556 614.2 159.1 914.5 111.2 41.6
1998 1454.8 374 278.1 272.2 188.8 158.9 463.1 545 610.3 155.1 929.2 110.3 43.1
1999 1466.1 390.1 280.2 278.4 185 166.6 447.2 556.1 640 164.2 939.3 109.9 44
2000 1453.9 402.4 258.1 284.6 189.7 175.3 447 571.8 627.8 163.7 960 113.4 43.9
2001 1586.8 393.1 274.4 292.5 191.4 179 466.6 567.2 675.6 173.2 981.1 116.4 41.9
2002 1545.4 400.7 264.1 287.5 198.6 176.3 475.5 571.2 711.2 177.4 986.7 126 40.8
2003 1599.2 401.2 282.4 278.8 203.5 169.6 461.6 546.7 646.9 178.6 969.9 131.5 43.4
2004 1651.5 420.4 264.1 290.8 210.2 174.8 495.1 616 699.8 186.8 999.1 132.1 39.8
2005 1695.2 428.5 235.2 290.8 218.7 168.6 492.9 583.3 694.4 191.8 1017.8 140.3 46.9
2006 1832.1 437.3 300.6 283.4 224.2 180.1 503.2 628.6 738.3 196.2 1018.9 146.3 46.9
2007 1824.7 420.4 266.5 295.9 227.1 183.7 517.5 642.9 736 210.6 1064.3 161.7 53.8
2008 1805.3 395.1 258.1 279.7 223.4 183 499.9 626.7 761.6 204.7 1020 167.4 55.7
2009 1860.7 401.4 276.3 305.3 217.1 187.7 556.6 665 721.3 207.3 947.1 168.8 62.6
2010 1823.7 416.4 253.5 302.1 216.9 192.2 551 678.7 723.9 209.8 986.4 166.9 62.3
2011 2026.9 422.8 285.2 295.2 226.6 201 551.8 728.8 764.2 216.8 1028.9 158.5 59.2
2012 1820.9 400.1 245.2 295.4 223.9 201.4 545.2 653.8 698.8 212.9 995.4 169.3 61.1
2013 1835.2 361.4 234.4 299.8 232.2 202.5 489.2 667.1 732.8 209.7 966.4 173.6 52.5
2014 1799.9 374.6 237.8 301.3 222.5 208.1 535.6 634.1 712.2 213.8 955.2 182.1 52.3
2015 1814.3 372.8 253.7 308.0 227.4 210.3 554.6 613.3 738.2 193.9 992.1 185.3 50.8

2016 1853.3 393.8 262.3 307.1 236.0 214.8 590.6 654.8 764.6 197.6 1014.1 201.7 36.2

34
2017 1881.3 405.2 262.0 311.4 237.3 216.9 600.0 656.7 799.4 204.0 1020.1 218.6 37.2
2018 1907.5 413.1 263.2 313.8 239.5 219.2 608.3 663.1 810.0 208.4 1025.8 254.8 40.3
2019 1930.4 420.9 263.9 316.0 242.0 221.5 617.2 669.6 821.2 212.5 1030.8 262.4 40.6
2020 1951.4 429.3 264.5 317.8 244.2 223.8 626.1 676.2 832.9 216.7 1037.1 267.5 41.8
2021 1972.7 436.4 264.9 319.5 246.8 226.2 634.6 682.4 843.2 220.6 1042.8 272.4 42.2
2022 1993.8 443.8 265.4 321.6 249.0 228.8 643.1 688.4 852.9 224.4 1048.6 277.6 42.5
2023 2013.7 451.9 265.5 323.9 251.5 231.3 651.8 695.4 864.3 228.4 1055.3 282.8 42.7
2024 2034.3 459.3 266.1 325.5 254.0 233.9 660.8 701.1 875.5 232.3 1061.1 288.1 43.0
2025 2055.6 466.5 266.5 327.4 256.5 236.5 669.8 707.2 891.7 236.1 1066.4 293.2 43.3
2026 2077.6 473.9 266.9 329.3 259.1 239.1 678.6 713.4 907.6 240.0 1072.7 298.5 43.6
2027 2099.6 481.0 267.8 331.3 261.7 241.7 687.3 719.0 923.1 243.9 1078.0 303.7 43.9
2028 2120.7 488.1 268.0 332.9 264.3 244.3 696.2 725.2 938.7 247.9 1083.7 309.0 44.3
2029 2143.1 495.1 268.9 334.6 267.0 246.9 704.9 730.4 954.2 251.8 1088.6 314.3 44.6
2030 2164.1 502.8 269.2 336.5 269.6 249.5 713.8 736.8 969.5 255.8 1094.0 319.7 44.9
2031 2187.0 509.7 269.7 338.3 272.2 252.2 722.6 742.9 985.3 259.8 1099.4 325.1 45.3
2032 2208.1 517.0 270.6 339.9 275.0 254.7 731.4 748.4 1000.7 263.7 1104.4 330.7 45.6
2033 2230.8 523.5 270.9 342.0 277.7 257.2 740.3 754.7 1015.8 267.7 1109.2 336.0 46.0
2034 2254.3 530.7 271.8 343.9 280.4 259.9 749.1 760.6 1031.4 271.7 1114.9 341.7 46.4
2035 2277.1 538.2 272.4 345.5 283.0 262.5 757.9 766.8 1046.5 275.7 1120.7 347.1 46.7
2036 2299.6 545.7 273.4 347.5 286.0 265.0 766.3 773.0 1061.7 279.8 1126.2 352.6 47.1
2037 2323.3 552.8 274.1 349.3 288.8 267.6 774.8 778.4 1077.1 283.9 1131.4 358.3 47.4
2038 2345.2 560.0 274.5 350.9 291.6 270.3 783.5 784.8 1093.0 287.9 1137.0 363.8 47.7
2039 2370.1 567.4 275.4 352.8 294.4 272.8 792.1 791.3 1107.4 291.8 1142.4 369.6 48.2
2040 2392.0 574.9 276.1 354.9 297.2 275.3 800.8 797.5 1123.0 296.1 1148.2 375.3 48.5
2041 2415.2 583.0 276.8 356.7 300.1 277.8 809.6 804.1 1139.1 300.2 1153.5 380.9 48.9
2042 2437.9 590.2 277.3 358.4 302.8 280.5 818.0 809.1 1154.4 304.4 1158.6 386.7 49.3
2043 2461.2 597.3 278.3 360.2 305.5 283.0 826.7 814.9 1169.7 308.8 1164.3 392.6 49.7
2044 2485.2 604.3 279.1 362.0 308.5 285.4 835.5 821.3 1185.8 313.1 1169.6 398.3 50.2
35
2045 2508.3 611.2 280.2 363.7 311.2 287.9 844.4 827.2 1201.4 317.3 1175.0 404.1 50.6
2046 2531.1 618.6 281.1 365.7 314.1 290.5 852.8 833.2 1216.9 321.7 1179.8 409.9 51.1
2047 2553.5 626.3 282.1 367.5 316.9 292.8 861.4 838.8 1233.5 325.9 1184.7 416.1 51.5
2048 2578.8 633.6 283.0 369.3 319.7 295.3 870.2 844.8 1249.1 330.2 1189.8 422.0 52.0
2049 2601.8 641.3 284.3 371.4 322.6 297.9 878.6 850.3 1265.5 334.5 1196.0 428.1 52.5
2050 2626.1 648.8 285.4 373.2 325.4 300.6 887.6 856.3 1282.1 338.7 1202.1 434.1 52.8

36
Appendix H: Prudent regional peak demand forecast (MW)

Bay of Central Hawkes Nelson Otago South


Region Auckland Northland Taranaki Waikato Wellington Canterbury West Coast
Plenty Districts Bay Marlborough Southland Canterbury

1997 1539.9 381.1 302.2 267.4 193 156.3 451.9 556 614.2 159.1 914.5 111.2 41.6
1998 1454.8 374 278.1 272.2 188.8 158.9 463.1 545 610.3 155.1 929.2 110.3 43.1
1999 1466.1 390.1 280.2 278.4 185 166.6 447.2 556.1 640 164.2 939.3 109.9 44
2000 1453.9 402.4 258.1 284.6 189.7 175.3 447 571.8 627.8 163.7 960 113.4 43.9
2001 1586.8 393.1 274.4 292.5 191.4 179 466.6 567.2 675.6 173.2 981.1 116.4 41.9
2002 1545.4 400.7 264.1 287.5 198.6 176.3 475.5 571.2 711.2 177.4 986.7 126 40.8
2003 1599.2 401.2 282.4 278.8 203.5 169.6 461.6 546.7 646.9 178.6 969.9 131.5 43.4
2004 1651.5 420.4 264.1 290.8 210.2 174.8 495.1 616 699.8 186.8 999.1 132.1 39.8
2005 1695.2 428.5 235.2 290.8 218.7 168.6 492.9 583.3 694.4 191.8 1017.8 140.3 46.9
2006 1832.1 437.3 300.6 283.4 224.2 180.1 503.2 628.6 738.3 196.2 1018.9 146.3 46.9
2007 1824.7 420.4 266.5 295.9 227.1 183.7 517.5 642.9 736 210.6 1064.3 161.7 53.8
2008 1805.3 395.1 258.1 279.7 223.4 183 499.9 626.7 761.6 204.7 1020 167.4 55.7
2009 1860.7 401.4 276.3 305.3 217.1 187.7 556.6 665 721.3 207.3 947.1 168.8 62.6
2010 1823.7 416.4 253.5 302.1 216.9 192.2 551 678.7 723.9 209.8 986.4 166.9 62.3
2011 2026.9 422.8 285.2 295.2 226.6 201 551.8 728.8 764.2 216.8 1028.9 158.5 59.2
2012 1820.9 400.1 245.2 295.4 223.9 201.4 545.2 653.8 698.8 212.9 995.4 169.3 61.1
2013 1835.2 361.4 234.4 299.8 232.2 202.5 489.2 667.1 732.8 209.7 966.4 173.6 52.5
2014 1799.9 374.6 237.8 301.3 222.5 208.1 535.6 634.1 712.2 213.8 955.2 182.1 52.3
2015 1814.3 372.8 253.7 308.0 227.4 210.3 554.6 613.3 738.2 193.9 992.1 185.3 50.8

2016 1961.1 451.8 292.7 324.6 249.3 226.4 619.1 696.8 819.2 216.3 1066.8 218.2 40.5

37
2017 2016.5 465.6 298.8 330.9 252.5 230.1 632.3 708.9 852.2 223.0 1081.4 240.1 42.8
2018 2059.5 482.3 306.3 335.9 256.4 233.7 644.1 722.0 869.1 228.9 1095.0 280.0 47.1
2019 2097.2 497.2 312.0 340.3 260.0 237.3 655.5 734.3 885.5 234.3 1106.2 291.1 48.6
2020 2129.7 511.3 317.3 344.1 263.6 240.8 666.9 746.4 902.4 239.6 1117.8 299.1 50.7
2021 2162.8 523.6 321.6 347.6 267.2 244.3 678.3 757.5 917.1 244.7 1129.1 306.4 51.9
2022 2194.5 535.4 326.0 351.3 270.5 247.8 689.5 768.0 930.9 249.6 1140.2 314.0 53.0
2023 2216.4 543.4 326.1 353.8 273.1 250.5 698.6 775.7 943.2 253.8 1147.8 320.0 53.2
2024 2238.9 551.0 326.8 355.6 275.8 253.3 708.1 782.0 954.2 257.9 1154.3 326.2 53.6
2025 2262.2 557.6 327.2 357.6 278.4 256.0 717.6 788.6 964.6 262.0 1160.3 332.1 54.0
2026 2286.3 565.2 327.8 359.7 281.2 258.8 726.9 795.4 981.0 266.2 1167.4 338.3 54.3
2027 2310.4 572.6 328.7 361.9 284.0 261.5 736.1 801.5 997.8 270.3 1173.4 344.4 54.7
2028 2333.6 579.9 329.0 363.6 286.8 264.4 745.5 808.3 1014.7 274.5 1179.8 350.5 55.1
2029 2358.0 587.1 330.1 365.5 289.7 267.0 754.6 814.1 1031.4 278.7 1185.3 356.8 55.5
2030 2381.0 594.2 330.4 367.6 292.4 269.9 763.9 821.1 1048.0 282.9 1191.4 363.0 55.9
2031 2406.2 601.7 330.9 369.5 295.2 272.7 773.3 827.8 1065.1 287.2 1197.5 369.3 56.4
2032 2429.2 608.9 332.0 371.4 298.2 275.4 782.5 833.8 1081.8 291.3 1203.2 375.8 56.8
2033 2454.1 616.8 332.3 373.6 301.0 278.1 791.9 840.7 1098.1 295.6 1208.6 382.0 57.2
2034 2479.8 624.3 333.4 375.7 303.8 280.9 801.2 847.2 1115.0 299.8 1215.0 388.7 57.6
2035 2504.8 632.7 334.2 377.5 306.7 283.6 810.4 854.0 1131.3 304.1 1221.6 394.9 58.0
2036 2529.4 641.2 335.3 379.6 309.8 286.4 819.3 860.8 1147.8 308.4 1227.7 401.4 58.5
2037 2555.4 649.4 336.1 381.7 312.8 289.1 828.3 866.7 1164.5 312.8 1233.6 407.9 58.9
2038 2579.4 657.6 336.5 383.4 315.8 291.9 837.4 873.7 1181.6 317.1 1240.0 414.4 59.3
2039 2606.6 666.0 337.7 385.5 318.9 294.6 846.4 880.9 1197.3 321.2 1246.0 421.1 59.8
2040 2630.6 674.6 338.5 387.8 321.8 297.3 855.7 887.6 1214.1 325.7 1252.6 427.8 60.2
2041 2656.0 683.9 339.2 389.7 324.9 299.9 864.9 894.9 1231.5 330.2 1258.5 434.3 60.6
2042 2680.9 692.2 339.9 391.6 327.8 302.8 873.8 900.4 1248.1 334.6 1264.3 441.1 61.2
2043 2706.4 700.2 341.0 393.6 330.7 305.4 883.0 906.7 1264.6 339.2 1270.7 447.9 61.7
2044 2732.7 708.3 341.9 395.6 333.8 308.0 892.2 913.8 1282.1 343.8 1276.7 454.5 62.2
38
2045 2758.0 716.2 343.2 397.5 336.7 310.7 901.6 920.2 1298.9 348.3 1282.8 461.3 62.7
2046 2782.9 724.7 344.2 399.7 339.8 313.4 910.5 926.7 1315.7 352.9 1288.3 468.1 63.3
2047 2807.5 733.4 345.4 401.6 342.8 315.9 919.5 932.9 1333.6 357.4 1293.8 475.3 63.8
2048 2835.2 741.8 346.6 403.6 345.7 318.5 928.8 939.5 1350.5 362.0 1299.5 482.1 64.4
2049 2860.4 750.6 348.0 405.9 348.8 321.3 937.7 945.6 1368.3 366.6 1306.1 489.2 65.0
2050 2886.9 759.2 349.3 407.8 351.8 324.1 947.1 952.1 1386.2 371.0 1312.2 496.2 65.4

39
Appendix G: Grid Exit Points by region
APR GXP label
Island Region APR GXP name
(unique)
N Auckland Albany 110 (Wairau Rd) ALB-110
N Auckland Albany 33kV ALB-33
N Auckland Bombay 110kV BOB-110
N Auckland Bombay 33kV BOB-33
N Auckland Glenbrook - Counties GLN-33-2-COUP
N Auckland Glenbrook - NZ Steel GLN-33-2-NZST
Glenbrook - NZ Steel load
N Auckland GLN-33-1
only
N Auckland Henderson HEN
N Auckland Hepburn Rd HEP
N Auckland Hobson HOB
N Auckland Liverpool Street LST
N Auckland Mangere 110kV MNG-110
N Auckland Mangere 33kV MNG-33
N Auckland Meremere MER
N Auckland Mt Roskill 110kV - KING ROS-KING
N Auckland Mt Roskill 22kV ROS-22
N Auckland Otahuhu OTA
N Auckland Pakuranga PAK
N Auckland Penrose 110kV PEN-110
N Auckland Penrose 22kV PEN-22
N Auckland Penrose 25kV PEN-25
N Auckland Penrose 33kV PEN-33
N Auckland Silverdale SVL
N Auckland Takanini TAK
N Auckland Wairau Road WRD
N Auckland Wiri WIR
N Bay of Plenty Edgecumbe EDG
N Bay of Plenty Kaitimako KMO
N Bay of Plenty Kawerau Horizon KAW-11-T1-T2
N Bay of Plenty Kawerau T11 and T14 KAW-11-T11-T14
N Bay of Plenty Kawerau T6 - T9 KAW-11-T6-T7-T8-T9
N Bay of Plenty Matahina MAT
N Bay of Plenty Mt Maunganui 33kV MTM-33
N Bay of Plenty Owhata OWH
N Bay of Plenty Rotorua 11kV ROT-11
N Bay of Plenty Rotorua 33kV ROT-33
N Bay of Plenty Tarukenga 11kV TRK
N Bay of Plenty Tauranga 11kV TGA-11

40
N Bay of Plenty Tauranga 33kV TGA-33
N Bay of Plenty Te Kaha TKH
N Bay of Plenty Te Matai TMI
N Bay of Plenty Waiotahi WAI
N Central Districts Bunnythorpe 33kV BPE-33
N Central Districts Bunnythorpe NZR BPE-55
N Central Districts Dannevirke DVK
N Central Districts Linton LTN
N Central Districts Mangahao MHO
N Central Districts Mangamaire MGM
N Central Districts Marton MTN
N Central Districts Mataroa MTR
N Central Districts National Park NPK
N Central Districts Ohakune OKN
N Central Districts Ongarue ONG
N Central Districts Tangiwai 11kV TNG-11
N Central Districts Tangiwai NZR TNG-55
N Central Districts Tokaanu TKU
N Central Districts Waipawa WPW
N Central Districts Wairakei WRK
N Central Districts Woodville WDV
N Hawkes_Bay Fernhill FHL
N Hawkes_Bay Redclyffe RDF
N Hawkes_Bay Tuai 110kV TUI-110
N Hawkes_Bay Whakatu WTU
N Hawkes_Bay Whirinaki 11 kV Bus A and B WHI
N Northland Bream Bay BRB
N Northland Kaikohe 110kV KOE-110
N Northland Kaitaia KTA
N Northland Kensington KEN
N Northland Maungatapere MPE
N Northland Maungaturoto MTO
N Northland Wellsford WEL
N Taranaki Brunswick BRK
N Taranaki Carrington St 33kV CST-33
N Taranaki Hawera HWA-33-1
N Taranaki Hawera (Kupe) HWA-33-2
N Taranaki Hawera 110 kV -1 HWA-110-1
N Taranaki Hawera 110 kV -2 HWA-110-2
N Taranaki Huirangi HUI
N Taranaki Motunui MNI
N Taranaki Moturoa MRA

41
N Taranaki Opunake OPK
N Taranaki Stratford 33kV SFD-33
N Taranaki Taumarunui NZR TMN
N Taranaki Wanganui WGN
N Taranaki Waverley WVY
N Waikato Cambridge CBG
N Waikato Hamilton 11kV HAM-11
N Waikato Hamilton 33kV HAM-33
N Waikato Hamilton NZR HAM-55
N Waikato Hangatiki HTI
N Waikato Hinuera HIN
N Waikato Huntly HLY
N Waikato Kinleith 11kV T1 through T3 KIN-11-T1-T2-T3
N Waikato Kinleith 11kV T5 KIN-11-T5
N Waikato Kinleith 33kV KIN-33
N Waikato Kopu KPU
N Waikato Lichfield LFD
N Waikato Piako PAO
N Waikato Putaruru PUT
N Waikato Te Awamutu TMU
N Waikato Te Kowhai TWH
N Waikato Waihou WHU
N Waikato Waikino WKO
N Wellington Central Park 11kV CPK-11
N Wellington Central Park 33kV CPK-33
N Wellington Gracefield GFD
N Wellington Greytown GYT
N Wellington Haywards 11kV HAY-11
N Wellington Haywards 33kV HAY-33
N Wellington Kaiwharawhara KWA
N Wellington Masterton MST
N Wellington Melling 11kV MLG-11
N Wellington Melling 33kV MLG-33
N Wellington Paraparaumu PRM
N Wellington Pauatahanui PNI
N Wellington Takapu Rd TKR
N Wellington Upper Hutt UHT
N Wellington Wilton WIL
S Canterbury Addington 11kV ADD-11
S Canterbury Addington 66kV ADD-66
S Canterbury Ashburton 33 ASB-33
S Canterbury Ashburton 66 ASB-66

42
S Canterbury Ashey 11kV Daikon ASY-11-MPAS
S Canterbury Ashley 11kV Main Power ASY-11-MPOW
S Canterbury Bromley 11kV BRY-11
S Canterbury Bromley 66kV BRY-66
S Canterbury Coleridge COL
S Canterbury Culverden CUL-33
S Canterbury Culverden CUL-66
S Canterbury Hororata 33kV HOR-33
S Canterbury Hororata 66kv HOR-66
S Canterbury Islington 33kV ISL-33
S Canterbury Islington 66kV ISL-66
S Canterbury Kaiapoi KAI
S Canterbury Kimberley KBY
S Canterbury Southbrook 33kV SBK-33
S Canterbury Southbrook 66kV SBK-66
S Canterbury Waipara 33kV WPR-33
S Canterbury Waipara 66kV WPR-66
Nelson-
S Blenheim BLN
Marlborough
Nelson-
S Stoke 33kV STK-33
Marlborough
Nelson-
S Stoke 66kV STK-66
Marlborough
S Otago-Southland Balclutha BAL
S Otago-Southland Brydone - Rayonier BDE-11-RAYN
S Otago-Southland Brydone - Solid Energy BDE-11-SOLE
S Otago-Southland Clyde CYD
S Otago-Southland Cromwell CML
S Otago-Southland Edendale EDN
S Otago-Southland Frankton FKN
S Otago-Southland Gore GOR
S Otago-Southland Halfway Bush - Palmerston HWB-110
S Otago-Southland Halfway Bush -1 HWB-33-1
S Otago-Southland Halfway Bush -2 HWB-33-2
S Otago-Southland Invercargill INV
S Otago-Southland Naseby NSY
S Otago-Southland North Makarewa NMA
S Otago-Southland South Dunedin SDN
S Otago-Southland Tiwai TWI
S South Canterbury Albury ABY
S South Canterbury Bells Pond BPD
S South Canterbury Blackpoint BPT
S South Canterbury Oamaru OAM

43
S South Canterbury Studholme STU
S South Canterbury Tekapo A TKA
S South Canterbury Temuka 33kV TMK-33
S South Canterbury Timaru TIM
S South Canterbury Twizel TWZ
S South Canterbury Waitaki WTK
S West Coast Arthur's Pass APS
S West Coast Atarau ATU
S West Coast Castle Hill CLH
S West Coast Dobson DOB
S West Coast Greymouth GYM
S West Coast Hokitika HKK
S West Coast Kikiwa KIK
S West Coast Kumara KUM
S West Coast Murchison MCH
S West Coast Orowaiti 110kV 1 and 2 ORO
S West Coast Otira OTI
S West Coast Reefton 110kV 1 and 2 RFN
S West Coast Westport WPT

44
Appendix J: Development of our forecasting
methodology
J1 2011 Review
In 2011 we reviewed our forecasting methodology to provide national and regional
peak demand forecasts.

Our aim was to produce forecasts that:

 recognised that the future is uncertain and determine a credible range of


future demand growth (i.e. covers the range of possible outcomes)
 are prepared and presented in a form that is suitable for grid planning
activities
 are reasonably stable between different forecasting years
 accurately reflect seasonality, recognising that there may be different drivers
to winter and summer peak demand.

The forecast method we developed took advantage of an ‘ensemble approach’ – i.e.


it uses a combination of several prediction models, rather than relying on a single
“best” model to produce forecasts. Four models were used in our ensemble:

 the Econometric model, which used forecasts of population and GDP


 the Endogenous model, a linear regression approach
 the Ad-hoc model, based on expert judgement
 the MED-derived model, which was adapted from projections in the Ministry
of Economic Development’s Energy Outlook

The models are described in the published document “Transpower Long-term


demand forecast, September 2011”.11

J2 2013 review
As shown in the Figure H1 below national peak demand grew strongly from 1997
through to 2006. However since 2007 growth in peak demand at a national level has
has been subdued.

11
https://www.transpower.co.nz/sites/default/files/plain-page/attachments/transpower-
demand-forecast-sept-2011.pdf

45
Figure J1: National peak demand

The high point achieved in 2011 occurred during the August “polar blast” weather
event where record low temperatures were recorded throughout the country.

In 2013, following continued flat growth we undertook a further review of our


approach which included gathering opinions on future levels of growth from several
industry stakeholders, including generator/retailers, lines companies, industrial
consumers, interest groups, and Government departments.

The main reasons cited for the low growth included:

 Reduction in energy intensive industry in New Zealand including the impact of


the global financial crisis and in particular the influence of our high exchange
rate
 Continuing energy efficiency gains with particular reference to the uptake of
compact fluorescent bulbs. There has been about a 25% uptake of compact
florescent lights between 2005 and 2012
 Impact of the Transmission Pricing Methodology (TPM) that provides some
incentive for lines companies and in particular industrial companies to
manage their peak demand
 Growth in embedded generation. Since 2007 in excess of 225MW of
embedded generation (equivalent to 3.5% of New Zealand’s winter peak
demand) has been built that has reduced demand supplied by transmission.
 Christchurch earthquake. Christchurch’s electricity demand reduced by 10%
in 2011 (but peak demand was still high due to the extreme cold snap) and in
2012 was still 25MW below the levels observed from 2006 to 2010.
 Recent mild winters.

46
We also asked people about their view regarding future growth. Overall there was an
expectation that demand will grow at a lower rate – 0.8% to 1.2% p.a. in the medium
term. In addition there was:

 General support for forecasting demand with models based on population (in
particular) and GDP growth. There was some acknowledgement that
demand’s link with GDP may be changing as we move to a less energy
intensive economy.

 There was recognition that energy efficiency gains are on-going. It has
happened in the past with the installation of more efficient appliances (e.g.
water heaters etc.) and it will continue to happen in the future. The effect of
the move to more efficient lighting was pointed out by a number of people as
a key near term influence on demand.

As a result of the review, and in response to some of the comments we received, we


have made some changes to our 2011 methodology. These included:

 The addition of a model to the ensemble that is based on fitting a short-term


trend. This model is an attempt to account for the view that more recent
trends in demand growth are more indicative of future trends. In particular
this model recognises that the recent low level of growth could continue.
 The removal of the “ad-hoc” model, which while it has some merit, was based
to some extent on subjective opinion.
 Alteration of the Econometric (or exogenous) model to focus on producing
peak demand forecasts whereas previously it produced an energy forecast
and then derived a peak demand forecast from that. More direct forecasting
of the peak was seen as a better forecasting strategy.
 More customised approach to forecasting industrial demand, rather than
treating this demand in a similar way to other loads.
 Adding a temperature variable to reflect the influence temperature has on
peak demand in winter.

J3 2016 review
J3.1 Background
During public consultation on the draft Electricity Demand and Generation Scenarios
(EDGS)12, stakeholders suggested that MBIE should take a role in validating
Transpower’s peak demand forecast.

MBIE asked NZIER to undertake a high-level review of Transpower’s forecasts and


methodology. NZIER’s memo stressed the value of simplicity and transparency in
our forecast methodology and suggested some changes to our approach. We have
12
http://www.mbie.govt.nz/info-services/sectors-industries/energy/energy-data-
modelling/modelling/electricity-demand-and-generation-scenarios/draft-edgs-2015

47
considered their recommendations and have, in consultation with MBIE and NZIER,
implemented a number of changes aimed at increasing confidence in our forecasts.
These changes have not required fundamental changes in the model specifications
and data requirements and we have aimed to not overly complicate our approach.

J3.2 Action points


We agreed to consider the following potential changes to our peak demand forecast
model.

Model changes:

1. Implement a scheme of using individual model performance estimates to


weight the ensemble results.
2. Investigate adding autoregressive time-series terms to individual model
regressions to deal with auto-correlation in model errors

Additional testing:

3. Implement in- and out- of-sample testing of individual forecasts and the
ensemble.
4. Undertake testing of break-points in the short-term endogenous model.
5. Undertake testing of dropping data for dry years.
6. Undertake testing of using levels vs growth rates.
7. Undertake testing of log transformations and temperature correction for
individual models.

Data consistency:

8. Communicate with MBIE to make sure that specific industrial customer


forecasts are consistent with EDGS assumptions.

The following sections summarise actions taken in response to the items listed
above.

J3.3 Ensemble weighting


Status: Investigated but not Implemented.

We have investigated weighting the ensemble by the quality of fit and are not
satisfied that it provides an improvement over evenly-weighting the models.

The method of merging the ensemble provides some quality weighting through the
variance in the model fit. At this stage we consider this an appropriate trade-off
between accuracy and transparency.

J3.4 Autoregressive time series terms


Status: Implemented

48
All models other than the short-term endogenous model are now regression models
with AR(1) errors,

𝑦𝑡 = β0 + 𝛽𝑋𝑡 + 𝜖𝑡

Where :

 yt is the response variable at time t


 Xt are the predictor variables at time t
 β0 is a constant term
 β are the regression coefficients associated with the predictors in X

And εt is the tth element of an autoregressive series of errors, such that:

𝜖𝑡 = 𝜃𝜖𝑡−1 + 𝜔𝑡

Where

 θ is the auto-regression coefficient in the time-series of errors


 ω is a white-noise term

We have found through our investigation that the addition of an AR(2) term into our
endogenous model provides a better option for a short-term endogenous model.
Therefore, for the short-term trend forecast, we have specified the autoregressive
error series as:

𝜖𝑡 = 𝜃1 𝜖𝑡−1 + 𝜃2 𝜖𝑡−2 + 𝜔𝑡

In particular, this approach removes the need to identify a “break-point” in the


historical data. This model specification appears to better capture shorter-term
dynamics and in particular gives a level forecast that is more consistent with the level
of the most recent data. Hence, we have respecified the short-term endogenous
model to use an additional lag in the error rather than a break-point.

The inclusion of the AR terms was found to significantly reduced the autocorrelation
seen in the modelled residuals, as seen for example in the graph below.

49
Figure J2: Comparison of autocorrelation in residuals from ensemble model (Auckland
winter peak)

We have also found that the level of the first forecast year is more consistent with the
level of the last data. In general, this has tended to translate the forecast down over
the whole forecast horizon. In some regions, there has also been an impact on the
growth-rate, but the strongest effect has been on the starting point of the forecast.

The graph below illustrates this adjustment in forecasting the National winter peak
forecast.

50
J3.5 Out-of-sample testing
We tested the model by withholding 1 to 5 observations and forecasting the last
remaining years of data using a model fitted to the earlier subset.

Results

We considered several sets of individual model specifications and calculated an out-


of-sample error for each. Several observations arose from this were that

1. Differencing tended to reduce out-of-sample error and the difference between


in and out-of-sample error
2. Logging the response variable tended to increase out-of-sample error as
there is little evidence of exponential growth in the last several years of
observed demand

We found that there was not one unique model specification that had the lowest out-
of-sample mean absolute percentage error across all regions. However, this test did
allow us to pick a specification that performed well across most regions.

The graphs below show winter peak forecasts to 2020 fitted to data truncated at 2010
to 2015 to show the effect of additional years of data on the long-term trend. The
shaded regions indicate years where at least one forecast has had data held back.

J3.6 Dry year’s demand


Status: Implemented

51
All data years are now included in all models.

The 2001 and 2003 data-points stand out in some regions but not others. For
simplicity, we have decided to include this data in the model rather than conduct a
test for each region that decides whether or not they are genuine outliers for each
region.

J3.7 Testing break-points


Status: Not applicable

We have re-specified the endogenous short-term model as the same as the long-
term endogenous model but with an additional AR(2) error term. The break-point has
been removed. The break-point provided some forecast value if the observed peaks
flattened but produced unrealistically positive/negative growth in several regions
where the last few years have dropped significantly.

A second autoregressive term tends to put more weight on the recent years of low
growth as appropriate to the data. Therefore we consider it a more consistent and
transparent approach to a short-term model.

J3.8 Modelling Choice: levels or growth rates?


Status: Implemented.

All models other than the MBIE-based model are now differenced once in order to
obtain more stationary model inputs.

J3.9 Log transformations


Status: Implemented

We have tested log transformations for all models and found that it provides little
benefit in fit and results in unrealistically high forecast growth-rates in some regions.
Hence we have left both the response and predictor variables untransformed in all
models.

J3.10 Temperature correction


Status: implemented.

Each model can now use the temperature variable for the winter forecast. The AIC
for a fitted model with and without temperature variable is calculated. If the AIC
comparison suggests that temperature is significant enough to warrant an additional
variable in the model, the model with temperature is used. Otherwise. the
temperature variable is discarded.

Results:

Temperature is found to be more significant in North Island and in cities.

52
J3.11 Industrial data
Status: We have and will continue to discuss this with MBIE to ensure some
alignment in assumptions.

53
Appendix K: Functional specification of our modelling
approach
We use four models to derive our forecasts:

a) long-term endogenous model – regression of historic peaks on year from


1997 to the latest year.
b) short-term endogenous model – regression of historic peaks on year from
1997 to the latest year.We include two autoregressive terms in the error
model which makes the model more responsive to near term dynamics.
c) Econometric – regression of historic peaks on GDP from 1997 to the latest
year. Temperature is also included where this is found to be of benefit.
d) MBIE derived – based on Ministry of Business, Innovation and Employment’s
energy forecast

We generate separate forecasts for 13 regions, as well as the North and South
Islands and national total. For each of these we produce summer, winter, and
shoulder season peak demand forecasts. We also produce region and island trough
forecasts for both day and night time.

Our approach is illustrated in an Excel workbook published on our website13.

K1 Inputs
K1.1 Historical Metered Demand
Historical net grid offtake data is sourced from Electricity Market Services (EMS)’
data feed to Transpower. At locations with both injection back onto the grid and
offtake from the grid, a net offtake figure is calculated. Generation station and local
substation loads are not counted as demand.

All top-down forecasting is done in terms of the modelled demand which is an


estimate of the end-user demand minus large industrial demand. It is defined as:

𝑀𝑜𝑑𝑒𝑙𝑙𝑒𝑑 𝑑𝑒𝑚𝑎𝑛𝑑
= 𝑔𝑟𝑜𝑠𝑠 𝑑𝑒𝑚𝑎𝑛𝑑 𝑚𝑖𝑛𝑢𝑠 𝑘𝑛𝑜𝑤𝑛 𝑚𝑎𝑗𝑜𝑟, 𝑚𝑜𝑠𝑡𝑙𝑦 𝑑𝑖𝑟𝑒𝑐𝑡 𝑐𝑜𝑛𝑛𝑒𝑐𝑡, 𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑖𝑎𝑙 𝑑𝑒𝑚𝑎𝑛𝑑

where

𝐺𝑟𝑜𝑠𝑠 𝑑𝑒𝑚𝑎𝑛𝑑 = 𝑜𝑓𝑓𝑡𝑎𝑘𝑒 + 𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑙𝑒 𝑒𝑚𝑏𝑒𝑑𝑑𝑒𝑑 𝑔𝑒𝑛𝑒𝑟𝑎𝑡𝑖𝑜𝑛 (𝐸𝐺)

The EG series come from the Electricity Authority’s Electricity Market Information
(EMI) website. Industrial demand is derived from EMS data.

When forecasting we estimate future EG and industrial demand externally from our
ensemble models based on the stated plans of the industrial participants or by

13
http://www.transpower.co.nz/about-us/what-we-do/planning-future/planning-inputs

54
making “business as usual” assumptions. We then subtract and add them
respectively to the modelled demand forecast to produce our forecast in terms of grid
offtake.14

K1.2 Temperature
As discussed in appendix N, we use a temperature adjustment when it is significant.
The variable used is the average of the midpoint between the maximum and
minimum temperatures on the days that include the 12 half-hours with maximum
demand. Days with more than one of the top half-hours will be duplicated. The
temperature data is sourced from the National Institute of Water and Atmospheric
Research’s (NIWA’s) National Climate Database15 using a representative weather
station for each region. For island and national forecasts we use population-
weighted averages. The future distribution of the temperature variable is estimated to
be normal with a variance given by the historical values.

K1.3 Population
We no longer include population as a specific explanatory variable having re-
specified the exogenous model to include just GDP and temperature where it is
found significant. We found little fit benefit in including the additional term and several
regions returned illogical coefficients if allowed to fit to both GDP and population. We
consider consistency of having the same variable in all models greater than the
benefit of the extra term.

K1.4 Gross Domestic Product (GDP)


Historical GDP figures are sourced from Statistics New Zealand16.

The New Zealand Institute of Economic Research (NZIER) provides us national and
regional expected GDP projections. We generate high and low GDP national series
by assuming growth is as expected +/- 0.5% per annum and then generate high and
low regional series keeping the regions in proportion to the expected series. We
assume the medium series is the mean of a distribution and the high-low range is
four standard deviations.

We derive GDP per capita by dividing by population. We assume that GDP per
capita is log-normally distributed and independent of population.

K2 Individual Forecasts
All models other than the short-term endogenous model are now regression models
with AR(1) errors, i.e.:

𝑦𝑡 = β0 + 𝛽𝑋𝑡 + 𝜖𝑡

14
See Error! Reference source not found. for an illustration.
15
See, http://cliflo.niwa.co.nz/
16
http://www.stats.govt.nz/infoshare/

55
Where :

 yt is the response variable at time t


 Xt are the predictor variables at time t
 β0 is a constant term
 β are the regression coefficients associated with the predictors in X

And εt is the tth element of an autoregressive series of errors, such that:

𝜖𝑡 = 𝜃𝜖𝑡−1 + 𝜔𝑡

Where

 θ is the auto-regression coefficient in the time-series of errors


 ω is a white-noise term

We have found through our investigation that the addition of an AR(2) term into our
endogenous model provides a better option for a short-term endogenous model.
Therefore, for the short-term trend forecast, we have specified the autoregressice
error series as:

𝜖𝑡 = 𝜃1 𝜖𝑡−1 + 𝜃2 𝜖𝑡−2 + 𝜔𝑡

Each individual model is more fully specified in section 0.

K2.1 Long-term trend forecast

This forecast is based on regression of peak demand on year from 1997 to the end of
winter 2015. All data years are now included. Temperature is included in the
regression for winter forecasts where the result is statistically significant but excluded
otherwise.

Note that the year and temperature variables are centred (e.g. the year 2000 is
represented as a 0, and the average of the temperature variable is represented as a
0) in the regression.

We do not attempt to forecast future temperatures. Instead we assume the mean


and variance of this variable will be the same in the future. In this way we include the
variance in temperature in our calculations of forecast uncertainty.

Uncertainty is driven by the error in the model fit and the variability in historical
temperature.

K2.2 Short-term trend forecast

In the Short-term trend model we have found that including two autoregressive terms
in the error model reduced the growth rate in many regions. Overall we feel that this

56
model preforms better than our original short term endogenous model with its
breakpoint – see our comments further below. Hence, we have respecified the short-
term endogenous model to use an additional lag in the error rather than a break-
point.

𝑦𝑡 = β0 + 𝛽𝑋𝑡 + 𝜖𝑡

with,

𝜖𝑡 = 𝜃1 𝜖𝑡−1 + 𝜃2 𝜖𝑡−2 + 𝜔𝑡

Uncertainty is driven by the error in the regression fit and the variability in the
historical temperature. The calculation of the uncertainty in the forecast for each year
is shown in our Excel illustration of the model.

K2.3 Econometric forecast

The econometric forecast regresses differenced demand on GDP over 1997-2015


and then projects this forward using GDP projections. Temperature is also included
as an additional explanatory variable in the regression for winter forecasts where the
result is statistically significant but excluded otherwise.

Uncertainty comes from uncertainty in the regression fit and the predictor variables,
and variability in historical temperature.

K2.4 MBIE-based forecast

The MBIE-based forecast is drawn from the Draft Electricity Demand and Generation
Scenarios17 electricity energy forecasts. We use the Base Case (Mixed Renewables),
scenario to provide an expected growth forecast and the High and Low Demand
scenarios to provide a range around this.

This model regresses peak demand on energy demand and projects this forward
using MBIE’s energy demand projections. Temperature is also included as an
additional explanatory variable in the regression for winter forecasts where the result
is statistically significant but excluded otherwise.

The uncertainty in the forecast comes uncertainty in the regression fit, variation in
historical temperature, and from the variation between the high and low demand
scenario forecasts. We assume that the three energy scenario forecasts are
approximately the 10th, 50th and 90th percentiles of a normal distribution.

17
http://www.med.govt.nz/sectors-industries/energy/energy-modelling/modelling/electricity-
demand-and-generation-scenarios

57
K3 Model Specifications
K3.1 Endogenous long-term
Response variable: peak demand

Predictor variable(s): Year, temperature18

Differenced: Once

Log transformations: peak demand

Error Model: AR(1)

K3.2 Endogenous short-term


Response variable: peak demand

Predictor variable(s): Year, temperature1

Differenced: Once

Log transformations: peak demand

Error Model: AR(2)

K3.3 Exogenous long-term


Response variable: peak demand

Predictor variable(s): GDP, temperature1,19

Differenced: Once

Log transformations: None

Error Model: AR(1)

K3.4 MBIE long-term


Response variable: peak demand

Predictor variable(s): MBIE energy forecast/data, temperature1

Differenced: None

18
The model selects whether or not to include temperature by comparing AIC of the fit with
and without temperature included.
19
We have re-specified the exogenous model as GDP only. We found little fit benefit in
including the additional term and several regions returned illogical coefficients if allowed to fit
to both. We consider consistency of having the same variable in all models greater than the
benefit of the extra term.

58
Log transformations: None

Error Model: AR(1)

K4 Earthquake correction
We no longer find the requirement to make a specific adjustment for the regional
impact of the Canterbury earthquake of 2011.

K5 Ensemble of forecasts
The ensemble of forecasts is far from a standard distribution so its mean and
variance cannot be calculated analytically. They must be found from percentiles of
the individual distributions, which are calculated by using the Student’s t Distribution.

K6 Mixtures of distributions
The mixture of forecasts is calculated as follows. We seek the value y0 such that y0 is
the pth percentile of the mixed distribution:

𝑓LT−endogenous + 𝑓ST−endogenous + 𝑓Econometric + 𝑓𝑀𝐵𝐼𝐸


𝑓𝑚𝑖𝑥 =
4

where fx represents the percentile of the xth model for a given peak demand value.

By way of illustration, we implement the process in the Excel illustration of the model
using the Excel Solver Add-in. The solver loops over all years searching for an
ensemble forecast, y0, that minimises:

𝐴𝐵𝑆(𝑃 − 0.25 ∑ 𝑓𝑓𝑐 (𝑦0 ))


𝑓𝑐 𝑖𝑛 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡𝑠

where P is percentile of the ensemble forecast sought, forecasts is the set of


individual forecasts, and ffc(y0) is the cumulative distribution function value of forecast
fc at y0 given the mean and variation calculated for the individual forecast and
assuming Student t distributions.

K7 Prudent forecast
The prudent forecast is calculated as the 90th percentile forecast for the first 7 years
and then grows at the same rate as the expected forecast thereafter.

The prudent forecast is defined so as to identify issues with enough time to complete
upgrades without exposing New Zealand consumers to excessive risks of unserved
energy.

59

You might also like