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