Download Summary of Transpower`s peak forecast process

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Data assimilation wikipedia , lookup

Transcript
Summary of Transpower’s peak forecast process
Transpower, June 2013
Overview
Our annual peak forecasting process can be divided into two inter-locking processes:


top-down process – used to derive national and regional forecasts
bottom-up process – used to derive grid exit point forecasts as well as inform
our top-down process.
We focus on finding a plausible range for future demand growth rather than on
specific scenarios. Future demand is uncertain, particularly at the limits of the
planning horizon and we feel it is important that our approach recognises this fact.
Figure 1 shows the plausible range associated with our latest forecast for peak New
Zealand demand.
New Zealand Peak Demand Forecast
12000
10000
Would take major shift
8000
MW 6000
Plausible range
4000
Would take major shift
2000
Historical peaks
2013 provisional
New expected
New prudent
Low bound
High bound
2030
2029
2028
2027
2026
2025
2024
2023
2022
2021
2020
2019
2018
2017
2016
2015
2014
2013
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
1999
1998
1997
0
Figure 1, New Zealand Peak Demand Forecast
Top-down forecast
In 2011 we introduced an ensemble approach where a number of different models
are used to derive forecasts. Each attempts to predict demand, excluding that of
major industrial users but including demand serviced by embedded generation, in a
slightly different way. Major industrials and embedded generation are tracked
separately on a case-by-case basis so as to calculate “grid offtake” demand.
Our ensemble approach uses four different approaches, each with uncertainty, and
mixes these to generate overall national and regional forecasts. See Figure 2 for an
overview.
From the spread of the peak forecasts we take the 50th percentile as the mean peak
forecast and a higher value as the prudent peak forecast. The prudent peak is set at
the 90th percentile for the first seven years and grows at the mean rate thereafter.
Ensemble of regional demand
forecasts
Populationand GDPbased
forecast
Trend
forecast
MET service
temperature
data
Trend 20082012
forecast
MBIE-based
forecast
Mix
Modelled
demand
NZIER
GDP
forecast
MBIE
energy
forecast
Historical demand
Embedded
generation &
industrials
Stats NZ
population
forecast
“Top-Down”
regional
peak forecast
Forecast
embedded
generation &
industrials
Consolidate
TP and lines’
companies’
local
knowledge
Lines’
Company
Consultations
Final GXP
peak
forecast
“Bottom-Up”
GXP
peak forecast
Figure 2 Peak demand forecasting process
Trend forecast
The trend forecast is based on straight-line regression of peak demand from 1997 to
2011 with savings campaign years (2001, 2003, 2008) excluded. The historical peak
demand data is corrected to account for differences due to temperature effects
before the model is fitted. Uncertainty comes from uncertainty in the regression
parameters and uncertainty associated with temperature.
Econometric forecast
The econometric forecast regresses the logarithm of peak demand on the logarithms
of population and GDP over 1974-2011 and then projects this forward using Statistics
New Zealand population projections and New Zealand Institute of Economic
Research GDP projections. As above, the historical peak demand data is also
corrected to account for differences due to temperature effects before the model is
fitted. Uncertainty comes from population and GDP variations, as well as that
associated with regression and temperature.
Ministry of Business Innovation and Employment (MBIE) forecast
The MBIE forecast uses the 2011 Energy Outlook electricity energy forecasts.
Uncertainty comes from the high and low versions of the MBIE forecast.
Trend 2008-12 forecast
The first three forecasts all predict similar growth rates but recent growth has been
much lower. To account for the possibility that recent rates are the beginning of a
new long-term trend, we have added a fourth member to the ensemble which is
based on the recent trend since 2007, but otherwise identical to the standard trend
model.
The ensemble
The four spreads of forecast values are mixed in equal proportions to form the
ensemble.
Bottom-up forecast
To forecast each of the 200+ individual grid exit point (GXP) peak grid offtakes, we
use:




local knowledge from various sources including the relevant lines’ company,
and local industry
lines’ company intentions of (permanent or temporary) feeder changes
forecasts of embedded generation and major industrial loads, and
detailed history of the load.
The individual GXP forecasts are derived with reference to the top down forecasts
and passed to the relevant customer for comment before being finalised. The GXP
forecasts are also used as a basis for calculating the contribution each GXP makes
to regional and island peaks.
Comparison with previous forecasts
Figure 3 compares this national forecast with previous prudent forecasts. The effect
of the lower growth rate from the short-term trend is visible in the difference between
last year’s mean curve (orange dashes) and this year’s (red dashes).
Figure 3, Historical comparison of prudent peak forecasts
Regional Detail
The figures below list the regional growth rates and how they have changed from last
year.
In the North Island, all regions are down except the Bay of Plenty. This is actually an
artifact of redistricting – we have moved Kinleith from Waikato to Bay of Plenty in our
planning region definitions.
In the South Island, all regions show reduced growth, except South Canterbury
where dairy and irrigation load growth continues. The West Coast has reduced the
most as it has lost some Holcim cement load and Solid Energy developments off a
low base.
Figure 4, North Island Regional Forecasts, 2013 forecast vs. 2012 forecast
Figure 5, South Island Regional Growth, 2013 forecast vs. 2012 forecast