INCORPORATING DSM AND NEW TECHNOLOGIES INTO THE FORECAST
CAPTURING EE PROGRAM SAVINGS » Customer usage has been trending down (use per customer) for the last ten years • New appliance and construction efficiency standards • Increasing real electricity rates • Increasing multi-family home market share (smaller square footage) • Economic downturn (higher vacancy rates) • State and utility sponsored EE programs » Models estimated with historical sales data already have significant efficiency embedded in the model » May even be worse with an SAE model as end-use intensity inputs are calibrated to saturation survey information and shipments data that reflect the appliance stock. The appliance stock in part is impacted by EE program activity
DSM ISSUE
How much DSM is already included in the forecast?
Load
Regression model assume that historical relationship continue in the future. Measured Load History Before DSM
Using Measured Load implies that past DSM pattern continue in the future
DSM Begins
Historical Period
Forecast Period
ADJUSTING FOR EE SAVINGS METHODS » “Add-Back” Approach • Add historical EE savings back to actual sales data • Forecast the new “reconstituted” sale data – No DSM Forecast • Subtract out all cumulative EE savings (past and forecast) • Used by New England ISO
» “ Incremental” Approach • Assume all past EE program savings are embedded in the baseline forecast • Subtract only future cumulative EE program savings • Used by majority of electric and gas utilities » Integrate through SAE model end-use intensity projections
ADD BACK METHOD (1)
Load + DSMhistory = f(x) Forecast = f(x) – DSMhistory – DSMfuture
Measured Load with DSM added back
Forecast No DSM, but includes National Trends
Load
Forecast with Past and Future DSM
Measured Load History Before DSM
Forecast With Past DSM, but without Future DSM
DSM Begins
Historical Period
Forecast Period
VERMONT RECONSTITUTED SALES Add-Back Past Efficiency savings
Reconstituted Sales Actual Sales
GENERALIZES ECONOMETRIC MODEL Reconstituted sales (Add Back) No EE Model Variable CONST HDD LagHDD CDD LagCDD Real Personal Income AR(1)
Coefficient (1,815,390) 74.430 (3.902) 202.853 54.616 79.910 0.869
No EE Forecast
StdErr 355,147 7.710 7.662 26.049 26.208 13.851 0.059
T-Stat -5.11 9.65 -0.51 7.79 2.08 5.77 14.62
EE Savings
With EE Forecast
ISSUES WITH THE ADD-BACK METHOD » Can be difficult to construct a reasonable reconstituted data series. • • • •
Need a high level of confidence in historical EE savings data series May not have EE historical savings that goes back far enough How do you adjust for EE degradation (measure persistency)? How do you translate annualized historical EE savings estimates to monthly rate class sales adjustments?
» Can be difficult to develop a reasonable forecast model • How would income or GDP relate to energy if we never had any EE programs? - Difficult to find a right-hand drivers to explain adjusted sales growth • Tend to be strong auto-regressive models • You can’t validate model performance against actual sales data
INCREMENTAL APPROACH » Develop baseline forecast with actual sales data and adjust for only future EE program savings. • Need a cumulative incremental monthly EE program savings projection (starting with the first forecast month)
» Generally starting with annualized program savings forecast • Annualized estimates assume that all measures are installed in the first month of the year • Meaningful for developing EE programs, not so meaningful for forecasting load impacts.
» The challenge is to turn annualized savings estimates into meaningful monthly sales impact series. “DSM accounting” is really hard. • Need to address the “double-counting” issue • Need to capture seasonal impacts (e.g., lighting programs have a larger impact in the winter months. cooling programs only impact summer months)
HOW MUCH FUTURE EE IS IN THE BASELINE MODEL? » Add the cumulative historical savings as a model variable • If nothing is captured: DSM coefficient = -1.0 • If half is captured: DSM coefficient = -.5 • If everything is captured: DSM coefficient = 0.0 Vermont residential average use model: 2008 to 2013 Variable XHeat XCool XOther DSM_perCust
Coefficient 1.552 0.989 0.986 -0.224
StdErr 0.100 0.114 0.021 0.118
T-Stat 15.485 8.659 46.437 -1.908
P-Value 0.00% 0.00% 0.00% 6.17%
Indicates 80% of EE program savings is captured by the baseline model
SAE APPROACH DSM Programs Impact End-Use Saturation, Efficiency, or Both
XVary ,m EnergyInte nsity ( EI ) y ,m Utilizatio n y ,m
EI y ,m
Sat Type y Type Eff y Type MoMult Type EI 09 m Type Sat09 Eff 09Type
• Incentives to remove second refrigerators reduces saturation • Lighting program improves efficiency • Promotional heat pump program increases saturation and efficiency
Utilizatio n y ,m
Price y ,m P rice 09
0.15
Income y ,m Income 09
0.10
HHSizey ,m HHSize 09
0.25
BDays y ,m 31
SAE MODEL APPROACH 1. Develop baseline end-use sales forecasts from SAE model 2. Subtract out end-use EE program savings from baseline enduse sales forecasts • adjust future impacts to reflect savings captured by baseline forecast (apply 0.20 to future EE savings forecast)
3. Calculate new end-use energy intensity forecasts that incorporate the EE program impacts 4. Execute estimated SAE model with EE program adjusted enduse intensity forecasts
RESIDENTIAL REFRIDGERATION FORECAST 205,000 200,000
Baseline
195,000
Adjusted
185,000 180,000 175,000 170,000 165,000 2013
2016
2019
2022
2025
2028
2031
30,000
2034
25,000
Program Savings
20,000
MWh
MWh
190,000
15,000
10,000 5,000 2015
2018
2021
2024
2027
2030
2033
REFRIDGERATION END-USE INTENSITY 500 480
Baseline
460
Adjusted
kWh / Hse Hld
440 420 400 380
360
Average annual change: 2015 to 2020
340
Baseline: - 1.0% Adjusted: -1.2%
320 300 2013
2016
2019
2022
2025
2028
2031
2034
EE ADJUSTING END-USE INTENSITIES Baseline
600
380 360
550 500
kWh per Hse Hld
kWh per Hse Hld
340
Adjusted
450
400
300 280 260
Space Heating
240
Water Heating
350
320
220 300
200 2013
2016
2019
2022
2025
2028
2031
2034
2013
2019
2022
2025
2028
2031
2034
920
2,200
900
2,100
880
2,000
kWh per Hse Hld
kWh per Hse Hld
2016
1,900 1,800
Miscellaneous
1,700
860 840 820 800
780
Kitchen and Laundry
760
1,600
740 1,500 2013
2016
2019
2022
2025
2028
2031
2034
720 2013
2016
2019
2022
2025
2028
2031
2034
EFFICIENCY PROGRAMS IMPACTS Residential Baseline
Commercial
Adjusted
Baseline
8,000
Adjusted
12.0
7,000 10.0 8.0
5,000
4,000 3,000
kWh/sqft
AAGR 2014-34: -0.4% 2014-34: -0.8%
AAGR 2014-34: -0.2% 2014-34: -0.5%
6.0 4.0
Estimate that Vermont EE programs will reduce average use growth by half
2034
2033
2032
2031
2030
2029
2028
2027
2026
2025
2024
2023
2022
2021
2020
2019
2018
2017
2016
2015
2034
2033
2032
2031
2030
2029
2028
2027
2026
2025
2024
2023
2022
2021
2020
2019
2018
0.0
2017
0
2016
2.0
2015
1,000
2014
2,000
2014
kWh/household
6,000
TECHNOLOGIES THAT DRIVE THE FORECAST LED Program
Net Metering Program
Cold Climate Heat Pump Program
Electric Vehicles
LED LIGHTING PROGRAM » In Vermont, lighting accounts for 15% of household electric use • CFLs now account for roughly half of residential lighting
» Efficiency Vermont is initiating a new lighting program promoting LEDs as the new “go-to” lighting technology. • LEDs are roughly 1/3 more efficient than CFLs (as measure on a lumen per watt basis)
» Lighting program impacts are captured through the SAE model • A new lighting intensity forecast is derived using a stock accounting model
» Program is expected to reduce residential sales by 54,000 MWh by 2020 (nearly as much as solar net metering).
VERMONT STATE FORECASTING FRAMEWORK Monthly sales and customer models are estimated for: • Residential • Commercial • Industrial • Street Lighting
LED lighting program modeled at the technology level detail
Economic Drivers Structural Changes Weather Conditions Electric Prices
VELCO System Hourly Load Data Normal PeakProducing Weather
End-Use and Customer Class Energy Forecast
System Peak Forecast Model
End-Use Coincident Peak Factors
Other technologies could not be modeled through an econometric model. Needed to take into account impact on system load shape
Net metering impact Cold-weather heat pump impact Electric vehicle impact
System Peak And Energy Forecast
LIGHT BULB ECONOMICS Home Depot Bulb Comparison
Bulb Cost
Life Hours
Incandescent
$1.25
1,000
0.7
CFL
$2.00
6,000
LED
$5.00
25,000
Bulb Type
Annl Life (Yrs)
kWh/yr
First Yr Cost
43
62.78
$19.19
$71.71
4.1
14
20.44
$6.09
$22.88
17.1
9
12.41
$7.48
$17.41
* 4 hours per day; electricity price of $0.20 kWh
Watts
Five Yr Cost
BASELINE LIGHTING TECHNOLOGY Saturation Forecast
LED TECHNOLOGY SHARE
Lighting Program
Baseline Projection
Utility lighting program results in faster market penetration of LEDs
LIGHTING STOCK ACCOUNTING MODEL » Annual bulb estimates by technology -
Incandescent CFL LED
» Factors • Average bulb life expectancy • Average bulb purchase and operating cost • Lighting standards • Customer growth • Fixture per household growth
LIGHTING INTENSITY CALCULATIONS Customers × Fixtures per HH (EIA) = Total Fixtures
Technology Shares (Adjusted for higher LED penetration) stock accounting model
HH Bulbs by technology type
Annual burn hours by technology type
Average wattage by technology type
Lighting sales by technology type (Lighting intensity = lighting sales / HH)
LIGHTING INTENSITY COMPARISON
Baseline Projection
Lighting Program
The lighting program pushes lighting savings forward
PHOTOVOLTAIC (PV)
SOLAR TRENDS » The U.S. installed 6,201 MW of solar PV in 2014, up 30% over 2013. » The Solar Energy Industries Association (SEIA) is forecasting 8,100 MW of solar PV installations in 2015, with the most rapid growth in the residential market. » Federal ITC, currently a 30% tax credit, is set to expire for individual tax payers at the end of 2016 and will be reduced to 10% for corporate tax payers. • The race to beat the expiring credit. Systems must be installed by December 31, 2016. » Continued decline in installation costs • Residential national average $3.84/watt • Non-residential national average $2.27/watt
THE DIFFERENCES BETWEEN STATES » Massachusetts installed over 300 MW of solar in 2014 » Total capacity of over 750 MW » Approx. 111 watts per capita
» New Hampshire installed just 3 MW of solar in 2014 » Total capacity of around 8 MW » Approx. 6 watts per capita » Does the sun not shine in New Hampshire??? STATE REGULATION AND INCENTIVES MATTER
VERMONT NET METERING PROGRAM » Vermont currently has in place a net metering cap of 15% of peak demand. The GMP forecast cap is reached by the end of 2017. » Net metering allows customers to sell excess generation back to the utility. Excess generation is credited to the customers’ future bills. » Excess generation is purchased by the utility at the retail rate. There is also a “solar adder” credit that is applied to total system generation. The solar adder is calculated as the difference between 20 cents and the retail energy rate.
ROOF TOP PV SATURATION Driven by Simple Payback » System costs (typical 5 kW system) » Federal tax credit » State incentive » Retail electricity costs » Utility buyback rate » Generated power • Own use • Excess (sold back to the utility) » Federal tax credit expiration results in an out-of-pocket cost increase of $4,300, resulting in a 2-year increase in payback period – reduces market penetration by 50%.
IMPACT ON RESIDENTIAL ROOFTOP ADOPTION
TOTAL SOLAR CAPACITY FORECAST
Most of the capacity is associated with Group Net Metering Systems » Stand alone solar generation (up to 500 kW) that is “sold” to individual customers (treated as net metering)
SOLAR HOURLY LOAD PROFILE
» Solar shape based on a 1 MW solar system for a typical meteorological year (TMY). » Weighted combined shape based on the assumption that 33% of systems are roof-mounted, 57% fixed-tilt, and 10% 2-axis tracker.
SOLAR GENERATION FORECAST STEPS (MWh) Solar Shape
Capacity Forecast
Month Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
MonthlyLdFct 0.08 0.11 0.14 0.19 0.20 0.21 0.20 0.20 0.16 0.12 0.08 0.06
Generation Forecast
BASELINE SYSTEM LOAD FORECAST Combine energy, peak, and profile forecasts
1. 2. 3.
Energy forecast derived from monthly customer class sales forecast Baseline peak based on heating, cooling, and other load requirements from sales forecast models and peak-day weather System profile generated based on calendar days, months, normal daily degree-days, and hours-of-light
FORECAST SOLAR HOURLY LOAD » Combine solar generation forecast (adjusted for line losses) with solar load profile
IMPACT ON SYSTEM PEAK DEMAND Add Solar Load Shape to Baseline Shape
Shifts peak from 2:00 to 5:00 Savings = peak at 2:00 – peak at 5:00
Given the peak-day profile is relatively flat, the solar demand impact is relatively small. One MW of solar capacity reduces peak demand by 310 kW
No capacity savings in the winter months
NET METERING SYSTEM PEAK IMPACT 45.0 40.0 35.0
25.0 20.0 15.0
10.0 5.0 -
2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034
MW
30.0
By 2023, net metering reduces peak demand by 3.7 percent.
COLD CLIMATE HEAT PUMPS
ELECTRIC HEAT PUMPS » Several kinds of heat pumps: • Central air source • Ductless air source • Ground source » Heat pumps are common in moderate winter climates where temperatures generally do not drop below freezing. » New cold climate heat pumps can operate at temperatures as low as 0° F. » Mini-split unit consists of two components: • External condenser; similar to a central AC condenser. • Internal evaporator, from which the unit emits or extracts heat to or from the building. » $3,000-$4,500 per unit, before state and utility incentives.
HEAT PUMP SATURATION BY CENSUS DIVISION
HEAT PUMP EFFICIENCY (Coefficient of Performance)
VERMONT COLD CLIMATE HEAT PUMP PROGRAM » Vermont plans to aggressively promote cold-climate air-source heat pumps • The target market are homes currently heating with propane and oil
» Unit projections were developed by Efficiency Vermont and GMP » Unlike other efficiency programs, this programs increases electric load
SATURATION PROJECTION » Roughly one percent of Vermont homes have heat pumps in 2014 » Increases to 5% in 10-years and 25% in 20-years
HEATING AND COOLING INTENSITY FORECASTS Saturation projection translates into strong heating and cooling use in the second-half of the forecast horizon (2024 – 2034) Heating
Cooling
2014 - 2024 2024 - 2034
45
Heating -1.20% 2.10%
Cooling 1.20% 5.00%
ESTIMATING HEAT PUMP DEMAND IMPACTS » Our initial forecast attempt was through the SAE Peak Model
Peak m a bc CoolVarm bh HeatVarm bo BaseVarm em CoolVarm CoolLoadm PeakDay _ wtTHIm HeatVarm HeatLoad m PeakDay _ HDDm •
Increase in cooling and heating load due to heat pump growth is captured in the CoolLoadm and HeatLoadm variables derived from the SAE residential sales model
» The winter peak impact, however, was too low. The peak regression model was not sensitive enough to changes in HDD – the elasticity on HeatVar is too small. Little electric heating in the historical load data
LOAD BUILD-UP APPROACH 1. Estimate Heat Pump Hourly Load » Forecast heat pump heating and cooling energy • Derived from residential SAE sales model
» Estimate heating and cooling hourly profiles • Based on daily normal weather, day of the week, season, and holidays
» Combine energy and profiles to generate a heat pump hourly load forecast
LOAD BUILD-UP APPROACH 2. Add-up loads » Aggregate system, solar, and heat pump load profiles and find the monthly or seasonal peaks
HEAT PUMP DEMAND IMPACT By 2024, heat pumps add 14 MW to summer peak and 8 MW to the winter peak
And 29,500 MWh in sales
ELECTRIC VEHICLES (EV)
ELECTRIC VEHICAL SALES TREND
» Total EV sales for Q1 2015 were only 3.0% higher than Q1 2014. » Nissan Leaf and Chevy Volt Q1 2015 sales were both down more than 20% from Q1 2014, likely a result of lower gasoline prices. » Tesla sales continue to grow; sales not likely tied to gasoline prices. They are just cool cars to own.
ELECTRIC VEHICLE FORECAST » In 2014, there were fewer than 1,000 registered electric vehicles in Vermont. » Started with existing number of electric vehicles in Vermont and assumed a 20% long-term annual vehicle growth through 2024 and constant growth (2,500 new vehicles after that, based on a recent Navigant study and other studies)
represents 3.0% of all vehicles by 2024 (includes plug-in hybrids)
SYSTEM IMPACT Energy: 35,166 MWh by 2024
Assume average annual vehicle use of 2,084 KWh.
Reflects current mix of EVs and PHEVs (plug-in hybrid electric) in Vermont (data provided by VEIC).
Peak Demand: 6.9 MW by 2024
Assumes most of the charging is shifted to night via EV charging tariff (based on recent EV assessment study)
SUMMER DEMAND IMPACTS New Technologies Year 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034
Base Net Heat Forecast Metering Pumps 1,026 -1 2 1,028 -3 3 1,031 -8 3 1,032 -13 4 1,032 -18 5 1,031 -22 6 1,030 -27 7 1,030 -31 8 1,032 -36 9 1,034 -38 12 1,036 -38 14 1,038 -39 17 1,041 -39 20 1,043 -39 24 1,047 -40 29 1,048 -40 34 1,049 -41 42 1,050 -41 51 1,051 -42 61 1,052 -42 73 1,053 -43 86
Electric Vehicles 0 1 1 2 2 3 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Total 1,028 1,028 1,027 1,025 1,021 1,018 1,014 1,011 1,010 1,013 1,019 1,024 1,031 1,038 1,047 1,055 1,063 1,073 1,086 1,099 1,113
» Base forecast includes EE program impacts » Net metering capacity cap is reached at the end of 2022
LESSONS LEARNED » For long-term forecasts, we need a flexible modeling framework. The problem does not fit into a nice econometric model • Allows us to evaluate the impact of new technologies and standards that are not reflected in the historical load data • Allows for evaluating EE program impacts » Difficult to generate a long-term peak demand forecast or isolate the demand impact of specific technologies without accounting for how new technology adoption changes the system load profile • Need to incorporate shapes into the long-term forecasting framework