Transcript Itron presentation on VELCO load forecast approach

VELCO Long-Term Demand Forecast
Kick-off Meeting
June 7, 2010
Eric Fox
© 2007, Itron Inc.
Agenda
 Discuss proposed framework for developing the longterm VELCO system and zonal demand forecasts
 Review ISO-NE forecasting approach
 Discuss issues related to Energy Efficiency and
Forecasting (EE&F) Forecast Guidelines
> Economic and weather data
> Incorporating the impact of state efficiency activity
> Incorporating the impact of interruptible load and demand
response programs
 Project schedule
© 2007, Itron Inc.
2
VELCO System and Zonal Demand Forecasts
 Develop twenty-year demand forecasts that captures:
> population trends, economic conditions, price
> peak day weather conditions
> end-use saturation and efficiency trends
• Standards, impact of federal tax credit programs, price induced
efficiency gains
• State and utility efficiency programs
• Interruptible load and demand control programs
 Team effort –
> program efficiency savings integration
> implementing forecast within forecast committee
guidelines
© 2007, Itron Inc.
3
VELCO Daily Peak Demand (MW)
© 2007, Itron Inc.
4
VELCO Monthly Peak (MW)
© 2007, Itron Inc.
5
Approaches for Forecasting Demand
 Generalized econometric model
> Approach used by New England ISO
• Demand = f(Energy, trends, peak day weather)
– Energy = g(real income, price, monthly weather)
 Hourly build-up approach
> Approach used last year
• Forecast class and end-use sales (SAE specification)
• Combine end-use sales with end-use load profiles
• Aggregate to system peak
 SAE peak model
> Proposed approach
• Forecast class and end-use sales (SAE specification)
• Demand = f(End-use coincident load, peak-day weather)
© 2007, Itron Inc.
6
Step 1: Estimate SAE Energy Models
 Build monthly revenue class sales models
> Construct SAE models for the residential and small
commercial customer classes base on actual sales data
> Estimate generalized econometric models for the
large/commercial and industrial classes
• Supplement with specific customer estimates where available
(such as IBM)
> Potentially estimate state level and utility service area
models for GMP, Central Vermont, and BED
© 2007, Itron Inc.
7
Statistically Adjusted End-Use (SAE) Framework
AC Saturation
Central
Room AC
AC Efficiency
Thermal Efficiency
Home Size
Income
Household Size
Price
Heating Saturation
Resistance
Heat Pump
XCool
Water Heat
Appliances
Lighting Densities
Plug Loads
Heating Efficiency
Thermal Efficiency
Home Size
Income
Household Size
Price
Heating
Degree Days
Cooling
Degree Days
Saturation Levels
Appliance Efficiency
Income
Household Size
Price
Billing
Days
XHeat
XOther
Sales m  a  bc  XCool m  b h  XHeat m  bo  XOther m  em
© 2007, Itron Inc.
Step 2: Develop End-Use Saturation and Efficiency Trends
 Use AEO 2010 New England Census Region forecast as
a starting points
 Adjust end-use saturation and structural data to reflect
Vermont
> KEMA appliance saturation survey
> BED survey work
> Efficiency Vermont market analysis
 Modify historical and forecasted efficiency trends to
reflect the impact of state and utility specific efficiency
programs
© 2007, Itron Inc.
9
Efficiency Program Impacts
Cooling Efficiency Program
25.0
Marginal
Efficiency
With DSM
Efficiency Path
20.0
SEER
15.0
No DSM
Efficiency Path
10.0
5.0
1995
© 2007, Itron Inc.
2000
2005
2010
2015
2020
2025
2030
Adjusted End-Use Indices (kWh per Cust)
2,500
EFurn
SecHt
CAC
RAC
EWHeat
ECook
Ref1
Ref2
Frz
Dish
CWash
EDry
TV
Light
Misc
2,000
1,500
1,000
500
0
2000
© 2007, Itron Inc.
2003
2006
2009
2012
2015
2018
2021
2024
2027
2030
Statistically Adjusted End-use Modeling (cont.)
Estimate monthly average use regression models:
AvgUse t  b 0  b1  XHeat
© 2007, Itron Inc.
t
  b 2  XCool t   b3  XOther t    t
12
XCool
© 2007, Itron Inc.
13
XHeat
© 2007, Itron Inc.
14
XOther
© 2007, Itron Inc.
15
Residential Average Use Forecast
© 2007, Itron Inc.
16
End-Use Energy Forecast
© 2007, Itron Inc.
17
Last Year’s Approach
Combine end-use energy
with end-use shapes
Residential
Base Use
Cooling
© 2007, Itron Inc.
Peak-Day System Hourly Load Profile (MW)
Aggregate Class Load
Forecasts to System Load Forecast
And
Find Annual System Peak
System
Residential
Commercial
Lighting
Industrial
© 2007, Itron Inc.
Step 3: Estimate SAE Peak Demand Model
 Derive end-use coincident peak load estimates from the
SAE sales models
• weight class estimates to reflect zonal area customer mix
 Construct peak-day weather variables
• 50% and 90% probability weather
 Combine end-use energy stock estimates and peak-day
weather into monthly SAE peak-day variables
 Estimate system and zonal peak demand models
 Develop seasonal peak demand forecasts for 50% and
95% probability weather
 Adjust for interruptible load and demand response
program impacts
© 2007, Itron Inc.
20
Simulation Results from Sales Models
Sales m  a  b c  XCool m  c c  Trend m  CDD m
 b h  XHeat m  c h  Trend m  HDD m
 b o  XOther m  c o  Trend m  e m
Cooling
Residential
Small C&I
Large C&I
Municipal
Heating
Other
© 2007, Itron Inc.
Simulation Results from Sales Models
 Sum of End-Use Energy
Total Monthly Energy (GWh)
> Normal heating for Res, SGS, LGS, …
> Normal cooling for Res, SGS, LGS, …
> Other loads for Res, SGS, LGS, …
© 2007, Itron Inc.
Total Monthly Energy – Normal Weather -- All Classes
Total Monthly Energy
Heating Variable Construction
 Annual Heating Transforms
 Monthly Heating Transforms
© 2007, Itron Inc.
Sum monthly heating values from the
sales model.
Interact heat index values with peak day
temperatures and prior day
temperatures. Use splines if needed.
Cooling Variable Construction
 Annual Cooling Transforms
Sum monthly heating values from the
sales model.
 Monthly Cooling Transforms
© 2007, Itron Inc.
Interact cool index values with peak day
temperatures and prior day
temperatures. Use splines if needed.
Residential Monthly Usage Profiles
Water Heating loads are lower in
summer due to warmer inlet
water temperatures
Lighting Loads are larger in
winter due to increased hours of
darkness.
Heating and Cooling
Refrigerator and Freezer loads
are larger in summer due to
warmer ambient conditions
inside the home.
© 2007, Itron Inc.
Residential Hourly Usage Profiles
Water Heating loads are lower in
summer due to warmer inlet
water temperatures
Refrigerator and Freezer loads
are larger in summer due to
warmer ambient conditions
inside the home.
© 2007, Itron Inc.
Lighting Loads are larger in
winter due to increased hours of
darkness.
Base Use Variable Construction
 Annual Other Transforms
Sum monthly energy values from the
sales model.
 Monthly Other Transforms
CPm,use  Re s _ Base y 
EnergySAE y,use
 EnergySAE
 PeakFrac use
y,u
u
Interact other annual usage with
peak monthly peak fractions by
class and end use.
© 2007, Itron Inc.
Example of Transformations – Res Lighting
EnergySAE
y , use
 EnergySAE
y,u
u
PeakFrac m ,use
Re s _ Base y
CPm,use  Re s _ Base y 
EnergySAE y,use
 EnergySAE y,u
 PeakFrac m,use
u
343 MW
Res Light CP
42 MW
© 2007, Itron Inc.
248 MW
31 MW
Estimate Peak Model
Regression Statistics
Variable
Coefficient
StdErr
T-Stat
BaseVar
1.012
0.095
10.627
CoolVar
151.852
5.1
29.774
CoolVar_May
-51.767
9.89
-5.235
CoolVar_Oct
38.606
17.159
2.25
HeatVar
7.719
3.116
2.477
MA_OtherLoad
1.558
0.175
8.901
Sep01
-1125.709
316.035
Apr02
-1518.308
Oct02
Iterations
1
Adjusted Observations
114
Deg. of Freedom for Error
103
R-Squared
0.91
Adjusted R-Squared
0.902
AIC
11.576
-3.562
BIC
11.84
314.963
-4.821
Std. Error of Regression
311.69
-1364.632
315.374
-4.327
Mean Abs. Dev. (MAD)
236.96
Apr05
-1201.245
314.945
-3.814
Jun06
-995.097
316.178
-3.147
Mean Abs. % Err. (MAPE)
Durbin-Watson Statistic
© 2007, Itron Inc.
3.76%
1.427
ISO New England Energy Requirement Forecast
 Uses a generalized econometric modeling framework
 Forecasts total system energy by state/region
> Annual model. Log/log specification. Forecast drivers
include:
•
•
•
•
Prior year energy
Real personal income
Real price
HDD and CDD
> Historical sales adjusted for past utility program efficiency
savings
> Exogenous adjustment for future efficiency savings
• Federal efficiency standards after 2013 (residential lighting)
• Passive efficiency savings as bid into the market
© 2007, Itron Inc.
30
ISO New England Peak Demand Forecast
 Forecasts system peak by state/region
> Daily demand model by month. Linear specification.
Forecast drivers include:
• Energy requirement forecast
• Peak-day weighted THI
• Trend interactive with peak-day THI
> Historical peaks adjusted for load interruptions
> Exogenous adjustment for future demand impacts
• Passive efficiency savings as bid into the capacity market
© 2007, Itron Inc.
31
ISO Forecast Methodology
 Relatively simple model specifications
> Annual energy vs. monthly sales
> Aggregate system level vs. revenue class
> Peak demand is primarily driven by the energy forecast
 Easier to model data series that have been adjusted for
prior efficiency savings
> No explicit end-use information incorporated in the model
 But significantly less information than that embedded in
the SAE framework
© 2007, Itron Inc.
32
EE&F Forecast Guideline Discussion
 Economic Data
> Forecast Vintage
> State vs. Regional Definition
 Weather Data
> Weather station
> Weather variables




© 2007, Itron Inc.
Modeling Approach
End-Use Efficiency and Saturation Trends
Incorporating the Impact Energy Efficiency Program
Other Issues
33
Proposed Project Schedule
 June
> Complete forecast database
 July
> Develop end-use efficiency and saturation data
 August
> Estimate preliminary system peak forecast
> Present preliminary results
 September
> Develop zonal demand forecasts
> Deliver preliminary forecast report
 October
> Deliver final forecasts and report
> Present final forecast
© 2007, Itron Inc.
34