Electric Technical Advisory Committee May 24, 2018Gary Dorris, PHDgdorris@ascendanalytics.com

NorthWestern Resource Plan Update

Key Themes for Resource Planning in 2018

Changing Market Dynamics and Forecasts

Resource Adequacy and Valuation

Valuation Approach for NWE’s Resource Options

Appendix

Agenda

2

Key Themes for Resource Planning in 2018

Continued deployment of variable renewable resources in the WECC is fundamentally changing market dynamics.

Weather is the new fundamental driver of WECC market dynamics. Resource planning now requires incorporation of uncertainty in weather, load, generation, and prices.

Need to adjust our understanding of resource adequacy in an era of high renewables.

Characterizing sub-hourly market conditions is becoming more critical for valuation of flexible resources.

Changing Market Dynamics and Forecasts

5 ascend analytics

WECC Supply StackWECC Supply Stack

• High correlation between renewable penetration and declining wholesale energy price

• Lower wholesale prices put economic pressure on traditional inflexible baseload generation

• Wind and solar capacity projected to increase ~30% in the next two years in the WECC

• Colstrip coal unit 1 and 2 retirement by 2022 projected to be replaced by renewables (wind)

Renewable Penetration in CAISO

6 ascend analytics

Dramatic price declines in renewables observed, especially where resource is best

• Solar PPA prices have steadily declined since 2008.

• Solar PPA prices are lowest in southwest US, where solar resource is greatest and good transmission to California markets.

• Prices of <$40/MWh are new norm at sites with high insolation

• Wind PPA prices have steadily declined since 2010.

• Wind PPA prices are lowest in interior of US, where wind resources are greatest.

• Prices of <$20/MWh are new norm at sites with high average wind speeds.

• Technology and supply chain improvements have driven down costs:

▪ Solar: mass panel production in Asia

▪ Wind: larger turbines, better materials, etc.

7 ascend analytics

Xcel RFO Results from Q4 2017

• Xcel Colorado issued an all source RFO for energy and capacity in 2017.

• They received 430 bids with prices coming in much lower than expected.

• Median bid prices:▪ Wind = $18.10/MWh▪ Wind + Battery = $21.00/MWh▪ Solar = $29.50/MWh▪ Solar + Battery = $36.00/MWh

• Solar and wind projects are utility scale.▪ Solar average size = 180 MW

/project▪ Wind average size = 415 MW

/project

Generation Technology # of Bids Bid MW

# of

Projects

Project

MW

Median

Bid Price

or

Equivalent

Pricing

Units

Combustion Turbine/IC Engines 30 7,141 13 2,466 $4.80 $/kW-mo

Combustion Turbine with Battery Storage 7 804 3 476 $6.20 $/kW-mo

Gas-Fired Combined Cycles 2 451 2 451 $6.70 $/kW-mo

Stand-alone Battery Storage 28 2,143 21 1,614 $11.30 $/kW-mo

Compressed Air Energy Storage 1 317 1 317 $14.60 $/kW-mo

Wind 96 42,278 42 17,380 $18.10 $/MWh

Wind and Solar 5 2,612 4 2,162 $19.90 $/MWh

Wind with Battery Storage 11 5,700 8 5,097 $21.00 $/MWh

Solar (PV) 152 29,710 75 13,435 $29.50 $/MWh

Wind and Solar and Battery Storage 7 4,048 7 4,048 $30.60 $/MWh

Solar (PV) with Battery Storage 87 16,725 59 10,813 $36.00 $/MWh

IC Engine with Solar 1 5 1 5 $50.00 $/MWh

Waste Heat 2 21 1 11 $55.40 $/MWh

Biomass 1 9 1 9 $387.50 $/MWh

Total 430 111,964 238 58,284

XCEL RFO Bid Results Summary

8 ascend analytics

Changing Market Dynamics and a Function of Solar

• High solar penetration during mid-day

• Low cost of solar causes prices to drop mid-day causing the heat rates to fall proportionally

• As the sun sets, inflexible thermal generation comes online causing prices to shoot up

• Intermittency of solar and restricted times of use cause declining average prices and more volatile prices

Implied heat rates decline over time due to increased renewable growth

Solar leads to declining mid-day heat rates, and increasing afternoon heat rates due to scarcity of flexible generation.

$10 -15 /MWh? - 2023

CAISO Renewable penetration vs. Implied Heat Rates WECC Renewables versus Implied Heat Rates

• As energy flows throughout the WECC, the “duck curve” is not just a California phenomenon.

• Increase in WECC solar affecting implied heat rates at Mid-C.

9 ascend analytics

Implied Heat Rates for Montana (2016-2017, 2025)

• The duck curve is most pronounced in spring months, when load is low and renewables are high

• Higher summer load reduces the duck curve effect

• Weather factors into drop in 2016 implied heat rates during winter months

• These changing implied heat rates provide a foundation for resource planning and projecting future implied heat rates under shifting market dynamics

• Implied heat rates expected to lower mid-day due to increased solar and increase in the early evening hours due to inflexible generation

Imp

lied

Hea

t R

ate

(MM

Btu

/MW

h)

Hours

Implied Heat Rates for Montana (2016-2017, 2025)

10 ascend analytics

$0

$1

$1

$2

$2

$3

$3

$4

$4

$5

Jan

-15

Mar

-16

May

-17

Jul-

18

Sep

-19

No

v-2

0

Jan

-22

Mar

-23

May

-24

Jul-

25

Sep

-26

No

v-2

7

Jan

-29

Mar

-30

May

-31

Jul-

32

Sep

-33

No

v-3

4

Jan

-36

Mar

-37

May

-38

Jul-

39

Sep

-40

Gas Prices ($/mmbtu)

0

5

10

15

20

25

Jan-

15

Jan-

16

Jan-

17

Jan-

18

Jan-

19

Jan-

20

Jan-

21

Jan-

22

Jan-

23

Jan-

24

Jan-

25

Jan-

26

Jan-

27

Jan-

28

Jan-

29

Jan-

30

Jan-

31

Jan-

32

Jan-

33

Jan-

34

Jan-

35

Jan-

36

Jan-

37

Jan-

38

Jan-

39

Jan-

40

Mid-C Implied Heat Rates (MMBtu/MWh)

On Peak HR (2015-2017) Off Peak HR (2015-2017) On Peak HR (Forward Market)

Off Peak HR (Forward Market) On Peak Heat Rate (Estimated) Off Peak Heat Rate (Estimated)

$0

$10

$20

$30

$40

$50

Jan-

15

Feb-

16

Mar

-17

Apr

-18

May

-19

Jun-

20

Jul-2

1

Aug

-22

Sep-

23

Oct

-24

Nov

-25

Dec

-26

Jan-

28

Feb-

29

Mar

-30

Apr

-31

May

-32

Jun-

33

Jul-3

4

Aug

-35

Sep-

36

Oct

-37

Nov

-38

Dec

-39

Mid-C Forward Power Prices ($/MWh)

On Peak Power Price (2015-2017) Off Peak Power Price (2015-2017)On Peak Power Price (Forward Market) Off Peak Power Price (Forward Market)On Peak Power Price (Estimated) Off Peak Power Price (Estimated)

NWPP 2017 Capacity Mix 2030 Projected Capacity Mix

Forward power prices projected to remain steady or slightly decline over time due to increased renewable penetration and retirement of Colstrip units 1&2 by 2022

Gas prices are escalated by HH escalation until 2030 and at 3% each year after 2030

Forward Power Market

• Due to unusually low AECO prices recently, gas prices historical and projected are a -0.5 basis to HH (historical AECO basis)

• IHR declining due to retirement of Colstrip Units 1&2 after 2022

Implied heat rates expected to decline to 5.1 due to large renewable penetration and coal retirements

8.9

9.2

8.2~6 ~5.1

Mid-C Forward Gas & Power Prices

11 ascend analytics

The Energy Imbalance Market (EIM)

• We are assuming NWE will be involved in some type of sub-hourly market providing a different framework for evaluating resources

• (EIM) provides non-CAISO balancing authorities in the WECC the opportunity to participate in CAISO’s real-time energy markets.

• CAISO now studying allowing BAs to participate in Day Ahead Market as well (“EIM +”)

• EIM is one tool to integrate high renewables

• Leverage greater geographical diversity and more resource options for flexibility

• Reduces need for curtailment and storage

• BA’s get an efficient and low-cost option to solve imbalances

• Stated benefits for participants:

• Less need for costly reserves, and better planning and efficient use of the regional high-voltage transmission system.

• Reduced carbon emissions and more efficient use and integration of renewable energy.

• Enhanced reliability by increasing visibility across electricity grids, improving transmission planning, and enhancing management of congestion across a wider expanse of the high-voltage transmission system.

12 ascend analytics

CAISO Resource Adequacy Requirements

• Resource adequacy for California entities• Monthly and annual compliance filings showing Load Serving Entities have enough owned and

contracted resources to meet at least peak demand plus a 15% reserve margin.• California has been very long on capacity, but the margin is shrinking quickly as thermal resources exit.

• Requirements for non-California balancing authorities (BA) to join EIM• The participating BA is required to maintain sufficient reserves (both energy and ancillaries) to meet

their normal requirements under NERC standards.• Must go into the hour with sufficient resources to meet forecasted load.

• Basic operations of the EIM• NWE would submit system information to ISO

• Outages• Base schedules of resources and loads• Bids (buy and/or sell)

• NWE units responsible for responding to ISO instructions after market clearing• CAISO does not require any more RA or flexible reserves than normal for EIM BA participants.

13 ascend analytics

Real-Time versus Day-Ahead Price Dynamics in Montana

• Day-ahead prices exhibit low volatility and are on average higher than the real-time prices.

• Renewable generation variability tends to drive price-spike volatility in the real-time 5-minute price.

• Negative prices occur, however $1,000 price spikes are significantly more common today.

• Provides market opportunity for flexible resources to earn high returns

Decker Node 5-Min Real Time Price (EIM)Mid-C Day Ahead Price

Scroll Bar

14 ascend analytics

Frequency and Impact of Price Spikes per month 2017 – Decker, Montana

• Price spikes are >$100 and <$-32

• Greater frequency of price spikes in the fall and winter months at Decker node in MT due to high wind and hydro penetration

Hour in Day

Freq

uen

cy

Frequency and Impact of Price Spikes per month 2017 – Decker, Montana

15 ascend analytics

Renewable Penetration (%)

DA LMP Volatility

RT LMP Volatility

2014 8.3% 23.5% 113.7%

2015 10.7% 27.0% 215.0%

2016 14.1% 36.5% 246.6%

2017 17.0% 66.2% 261.2%

LMP Volatility vs. Renewable Penetration (CAISO)

• When renewable penetration in a system increases from 10% to 20%, volatility becomes more apparent

• High correlation between renewables and LMP volatility is more striking in the RT market as RT pricing fills the voids (discrepancies in generation between day-ahead and real-time)

• Volatility increases at a faster rate than renewable penetration in the RT market

• Volatility will increase dramatically as renewable penetration increases

Real-Time (RT) LMP Volatility vs. Renewable Penetration Day-Ahead (DA) LMP Volatility vs. Renewable Penetration

16 ascend analytics

• Volatility approximately doubled across the state of California from 2014 to 2016, as renewable generation almost doubles• The central valley saw high volatility in 2014 due to congestion. Overall the volatility increase is not due to congestion across the whole state and most

explained by increasing renewable generation

% Load served by wind and solar

2014 2016

8.3% 14.1%

LMP Volatility increases drastically over time (CAISO)

*Radius of the bubbles indicates the magnitude of price

17 ascend analytics

• Volatility increases 3x – 4x between 2015 to 2017 in the MT/WY/ID region

2015 2017

LMP Volatility increases drastically over time (MT/WY/ID region)

*Radius of the bubbles indicates the magnitude of price

18 ascend analytics

Historic Forecast

Price Volatility in Montana to increase as renewable penetration increases

• Market price volatility is expected to increase as renewable generation increases.

• Greater premium on highly flexible capacity

• By 2021 the growth in price volatility is constant.

• Volatility peaks during the summer months.

19 ascend analytics

Renewables are the new base energy pushing out

inflexible power plants

Renewables increase

market price volatility and lower power

prices

Flexibility is the new source of

value and renewable

integration is the primary challenge

The bottom line – Flexibility is king

Resource Adequacy and Valuation• Deterministic vs. Stochastic modeling• Loss of Load Probability in Resource Planning• Flexibility for Renewable Integration

21 ascend analytics

Need for new tools to incorporate uncertainty:Deterministic vs. stochastic models

• Deterministic models can bias results with their limited pathways into the future.• Deterministic modeling misses critical

scenarios, producing inconsistent values.

• The likelihood of deterministic results actually occurring are not understood.

• Simulated weather captures actual operations of renewables and load, relative to normalized weather utilized in deterministic models

• What’s the impact of unused information?• Inaccurate forecasting

• Assessing risk becomes difficult

22 ascend analytics

Hybrid Battery- 2 hour provides least NPV cost with low Capital investment, fixed costs, and energy costs

Resource Selection using Stochastic Simulation of Meaningful Uncertainty Case Study of a Southern CA resource selection analysis

Planning Capacity: 371 MWEnergy: 180 MW, 4 hourReg: 180 MW, 15min

Planning Capacity: 360 MWEnergy: 360 MW, 4 hourReg: 0

Capacity: 353 MW Planning Capacity: 371 MWEnergy: 180 MW, 3 hourReg: 180 MW, 15 min

$5,245 M

$4,975 M $4,975 M$4,872 M

$5,397 M

*Planning Capacity defined as dependable capacity which for batteries is 4 hour storage minimum

Planning Capacity: 371 MWEnergy: 360 MW, 2 hourReg: 180 MW, 15min

23 ascend analytics

Resource Selection using Stochastic Simulation of Meaningful Uncertainty Case Study of a Southern CA resource selection analysis

Normalized weather, load, and price conditions systematically

over value baseload gen…

…and under value flexible gen. Plus other models cannot quantify the significant value in sub-hourly markets

Only PowerSimm can value risk for decision analysis…

Bottom line: When you cannot simulate weather, load, renewables and prices to capture “meaningful uncertainty”, you may make the wrong resource choice simply due to modeling limitations. PowerSimm is the right tool for high renewables portfolios!

…and properly value renewables under real-life operating conditions

24 ascend analytics

Keys to Future Supply Portfolios

• Need to manage cost uncertainty and not position uncertainty• Recognize new exposure is to price volatility not cost of energy

• Declining prices limit exposure of average open positions

• Increased volatility in prices increase exposure to future cost uncertainty

• Need supply portfolio to address renewable integration• More regulation and 15-minute INC/DEC capability

• Need to incorporate uncertainty in decision analysis• Increased renewable generation causes increased market price volatility

• May have long-term impact of pricing itself out of market

• Need to maintain key structural relationships• WX → Renewables & Load → Transmission → Market Prices

• Consistency with forward market and fundamentals

25 ascend analytics

Optimized automated resource selection with PowerSimm

• Uses algorithms to minimize net present value of system costs, including production cost and capital investment, across 100s of simulated paths for weather, load, renewables, and prices.

• Stochastic mixed-integer linear program that finds the optimal retirement, conversion, and deployment of new resources across the planning horizon.

• Solves expansion problem while maintaining several constraints including:

• Reserve margin

• Renewable portfolio standard

• Ancillary and flexibility requirements

• Transmission limits

• Physical generation limits for new and existing resources

Optimal solution

26 ascend analytics

Solving for Capacity Expansion Plans: Objective Functions

Min Expected Value of Total Cost

E[NPV of Total Costs] = E

𝑡=1

𝑇

𝑖=1

𝑁

𝛽𝑡 ∗ 𝑇𝑜𝑡𝑎𝑙 𝐶𝑜𝑠𝑡𝑠𝑖𝑡 ∗ 1𝑖𝑡

• Fixed Costs follow revenue requirements

Depreciation, Amortization, Current Taxes, Deferred Taxes, Insurance, Property Taxes, On-going capital improvements, return on equity and debt

• Variable Operating Costs comes from hourly dispatch aggregated up to monthly totals including:

Start up costs, min uptime, and min downtime constraints, emissions & variable heat rates

Subject To:

Reserve Margin Constraints Ancillary Service Requirements

σ𝑖=1𝑁 𝐶𝑎𝑝𝑎𝑐𝑖𝑡𝑦𝑖𝑡 ≥ 𝛾𝑃𝑒𝑎𝑘𝐿𝑜𝑎𝑑𝑡 𝑓𝑜𝑟 𝑡 = 1 𝑡𝑜 𝑇 σ𝑖=1

𝑁 𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑖𝑡 ≥ 𝑅𝑒𝑔 𝐿𝑖𝑚𝑖𝑡𝑡 𝑓𝑜𝑟 𝑡 = 1 𝑡𝑜 𝑇

where γ is the required reserve margin σ𝑖=1𝑁 𝑆𝑝𝑖𝑛 𝑅𝑒𝑠𝑖𝑡 ≥ 𝑆𝑝𝑖𝑛 𝐿𝑖𝑚𝑖𝑡𝑡

Energy Constraint

σ𝑖=1𝑁 𝐸𝑛𝑒𝑟𝑔𝑦𝑖𝑡 ≥ 𝐿𝑜𝑎𝑑𝑡 𝑓𝑜𝑟 𝑡 = 1 𝑡𝑜 𝑇 σ𝑖=1

𝑁 𝑁𝑜𝑛𝑆𝑝𝑖𝑛 𝑅𝑒𝑠𝑖𝑡 ≥ 𝑁𝑜𝑛𝑆𝑝𝑖𝑛 𝑅𝑒𝑞

Renewable Constraint

σ𝑖=1𝑁 𝑅𝑒𝑛𝑒𝑤𝑎𝑏𝑙𝑒𝑠𝑖𝑡 ≥ 𝑅𝑃𝑆𝑡 𝑓𝑜𝑟 𝑡 = 1 𝑡𝑜 𝑇 σ𝑖=1

𝑁 𝐹𝑙𝑒𝑥 𝑅𝑎𝑚𝑝𝑖𝑡 ≥ 𝐹𝑙𝑒𝑥 𝑅𝑎𝑚𝑝 𝑅𝑒𝑞

27 ascend analytics

Optimal Capacity Expansion Plan (2015 Plan)

Min NPV of Rev Requirements while satisfying increasing reserve margin constraint

28 ascend analytics

Future Portfolio Design Process Using a Scenario Approach

Assess RPS Need

Build future portfolios to meet need

Candidate resources

Load Forecast RPS assumptions

Evaluate peak capacity need

Add batteries and/or ICEs

Evaluate flexible capacity needs

Meets reliability,

RPS? No

Production cost

NPVYes

Market Prices

29 ascend analytics

What is Resource Adequacy with High Renewables?

• Planners historically viewed resource adequacy as simply making sure to have enough supply to meet peak demand. • This was straightforward when all capacity was dispatchable

• What does resource adequacy mean in a high renewables system?1. We need to understand how much renewables can be relied on under real-world operating

conditions.

2. We need to understand how batteries enhance capacity by both freeing up thermal gen from regulation duty and providing 4-hour duration energy on peak.

3. We need to understand the need for flexible resource adequacy and plan for adding flexible resources to integrate renewables.

4. Thermal begins to play the “back-up” role when the sun doesn’t shine and wind doesn’t blow. Thus we need to be careful to properly characterize forced outages.

30 ascend analytics

Questions Addressed

• How does PowerSimm identify capacity and capacity attribute needs for an individual utility and for the region? How does it differ from how other modeling tools identify needs?

• How does your modeling tool characterize the capabilities of resources to provide capacity and capacity attributes? How does it differ from other tools?

• How does your modeling approach address the opportunity for regional resource sharing?

31 ascend analytics

Addressing Commission Comments from the 2015 IRP

Comment Topic Issue Action Plan for 2018

Resource adequacy • NorthWestern has not completed a system optimization study with its entire fleet of resources to determine what capacity its base portfolio provides and, thus, what type of incremental capacity is needed

• The Plan provides little analysis and documentation that supports its selected capacity acquisition rate produces “minimal resource adequacy” by 2028.

• NorthWestern’s resource adequacy should be measured by its retail load’s position relative to the region’s or interconnection’s peak demand, while taking into consideration import limitations.

• Use automated resource selection to optimize deployment of diverse resources including both supply and demand side.

• Include significant analysis and documentation of capacity position through time.

• Resource adequacy analysis will include broad consideration of regional markets and dynamics.

Scope of resource alternatives

• The Commission shares many of these concerns and finds that NorthWestern’s 2015 Plan falls short of an “evaluation of the full range of cost effective supply and demand-side management options”

• NorthWestern will perform extensive evaluation of EE and DR programs along side supply-side options. See slide 45

*Slides refer to April 26, 2017 ETAC

32 ascend analytics

Capacity Needs Assessment

1) Capacity to Meet Peak Demand• Requirement: Procure adequate supply to serve peak demand given potential for generation

outages and intermittency of renewable supply

• Measurement: Traditionally reserve margin = firm capacity in excess of peak load

• Assessment of need: Best modeled by simulating available capacity and load over a large variety of future conditions for meteorology and generation outages

2) Flexibility Resource Needs• Required by regional balancing authorities to keep system at 60 Hz

• Ascend’s PowerFlex determines the flexible resource requirements as a function of renewable generation▪ Regulation ~ 1 minute ramp

▪ INC ~ 15 minute ramp up

33 ascend analytics

Loss of load occurs when available generation is less than load

Loads vary due to weatherLoad is most likely to exceed generation during hours with high load, high generator outages, or both

0 5,000 10,000 15,000 20,000 25,000 30,000

Pro

bab

ility

MW

Load

distribution

Generation

distributionLoad Generation

LOLP

Load >

Generation

34 ascend analytics

Resource Adequacy: Calculation of Loss of Load Probability

Simulated Weather

Hydro Gen

Renewables

Load

PowerSimm captures the interactions of weather→renewables, hydro, and load and generation

+

Weather drives load and renewable generation

Simulated renewables generation & load

Simulated generation outages

=

Generation outages

Note: Models that don’t simulate full generation outages with expected duration and variance in outage duration. Traditional models use an expected outage rate times available capacity in a deterministic construct. Thus, they can’t calculate probability of outages.

Capability to serve

peak load

35 ascend analytics

Resource Adequacy from 2015 Resource Plan

0

10

20

30

40

50

60

Ho

urs

(Lo

ad >

Cap

acit

y)

MEAN

P5

P95

0

20

40

60

80

100

120

MW

Sh

ort

MW ShortLoss of load probability analysis can tell planners how much more firm capacity will be needed in order to satisfy reliability requirements such as “1 day in 10 years” and/or reserve margin.

Hours Short by YearNorthwestern would have to add 450 MW to achieve a “1 day in 10 years” loss of load standard

Typically, grid connected entities in the WECC can plan to the mean of the simulated LOLx, since robust markets can fill in when the portfolio is short

36 ascend analytics

Future Portfolio Design Process

Assess RPS Need

Build future portfolios to meet need

Candidate resources

Load Forecast RPS assumptions

Evaluate peak capacity need

Add batteries and/or ICEs

Evaluate flexible capacity needs

Meets reliability,

RPS? No

Production cost

NPVYes

Market Prices

37 ascend analytics

Flexibility Resource Requirements

• NWE is now required to meet the RBC standard for Area Control Error (ACE). At every moment, dynamic ACE limits are applied as a function of WECC frequency instead of the constant ACE limit used under CPS2.

• This change in standard ensures that all balancing area responses to ACE move in the right direction for overall grid frequency.

• The balancing area can exceed limits for 30 minutes before it is considered a violation.

WECC Frequency > 60 Hz → upper limit on ACE

WECC Frequency < 60 Hz → lower limit on ACE

38 ascend analytics

Minutely Dispatch and Regulation

• Solar 1 MW ~ 0.2 MW flexible resources

• Wind 1 MW ~ 0.6 MW of flexible resources

• Changes in gradient of net load

• Base load 3 ’s in gradient

• Net load 7 ’s in gradient

2020 Baseline + 200 MW Wind + 100 MW Solar

39 ascend analytics

Renewable Integration Challenge most pronounced in Spring and Fall

Illustration of NWE portfolio, Spring Illustration of NWE portfolio, Fall

Wind drought

Too much wind, have to curtail or sell

Higher loads in the spring/summer lower curtailment risk, but the regulation challenge remains

Lower load and strong renewables make spring and fall prime for overgeneration risk. Sometimes you either get too much renewable gen or none at all.

40 ascend analytics

50% RPS- Hourly Dispatch – Case of New Mexico

• Dispatch for New Mexico under a 50% RPS target in the summer and spring shown

• As renewable penetration increases in the system, volatility increases, creating a greater need for flexible generation to maintain reliability

• Flexible generation can provide fast ramping as the sun sets

Batteries provide fast ramping as sun sets

Despite adding renewables, CT’s must run during the day due to forecasted load growth

Renewables increase probability of overgeneration. Must have transmission available for market sales or risk curtailment

Regulation from batteries and ICE replace traditional CC and CT dispatch during the day

41 ascend analytics

Market Potential for Resource Adequacy

• Fundamental modeling with PowerSimm suggest contracting for capacity and flexible reserves can be attractive but potential physical limits exist:• Transmission constraints

• Regional deficit of flexible capacity • Regulation prices in CA increased in 2017 by 400% over previous four years

• Regional capacity deficit projected ~ 2020

• Renewables provide negligible amount of dependable capacity

• Generic modeling assumptions are no substitute for testing the market with solicitations• Solicitation will provide the most competitive responses for evaluation

• Economic offers need to assessed relative to portfolio needs

Valuation Approach for Northwestern’s Resource Options

43 ascend analytics

Capturing Structural Change of High Renewables

Relative Importance in Resource Valuation for 2025

ComponentContribution to

ValuationWhy?

Weather +++++ Drives renewable generation, load, and market prices.

Load + Relatively stable and known.

Ave Gas Price + Diminished influence on portfolio cost and lower volatility with recent increases in supply.

Ave Power Price + Declining average prices but increasing volatility.

Volatility of power prices

++++ Significantly increasing with high renewable penetration rates.

5 min Price ++++ Extremely volatile presenting opportunity and cost. Negatively correlated with renewable production.

44 ascend analytics

Relative resource value for 2025 and beyond

Resource TypeRelative Energy Value

Flexibility Value

Capacity Value

Avoided Transmission

ValueWhy for 2025?

Combine Cycle +++ + +++++ N/A High start-up cost and min run times of 4+ hours.

Combustion Turbine

+ ++ +++++ N/A High start-up cost.Ramps to full load in 15 min.

ICE ++ +++++ +++++ N/A No start-up costs. Ramps to full load in 5 minutes.

Battery Storage

++ +++++ +++ +++++ Flexible and responds to 5 min prices.Rapidly declining costs till 2025.

Solar ++ + +++ Regional excess of solar generation in 2025. Limited dependable capacity.

Wind +++++ + + N/A Off-setting production to solar. Greater dependable capacity.

Energy Eff +++ + ++ +++++ Declining cost effectiveness for new programs. Relative cost to declining value of energy.

Demand Response

+ +++++ +++++ +++++ High need for flexibility.

45 ascend analytics

Future Portfolio Design Process

Assess RPS Need

Build future portfolios to meet need

Candidate resources

Load Forecast RPS assumptions

Evaluate peak capacity need

Add batteries and/or ICEs

Evaluate flexible capacity needs

Meets reliability,

RPS? No

Production cost

NPVYes

Market Prices

46 ascend analytics

HDR Generation Unit Characteristics and Operating Costs

Source: HDR

Heavy-Frame

CombustionTurbine

AeroderivativeCombustion

Turbine

Simple-Cycle

Combined-Cycle

18-MW Internal

Combustion Engine

Geothermal Wind Solar

Nominal Capacity (MW)

26 43 44 123 18.5 20 100 100

Heat Rate (Btu/kWh)

10,100 9,500 10,020 7,190 8,350 1000 - -

Cold Startup Time (hr)

0.167 0.083 0.167 4 0.083 - - -

Ramp Rate (MW/hr)

1800 3000 720 720 948 240 - -

Min. Down Time (hr)

0.25 0.25 0.25 0.25 0.25 - - -

Fixed O&M ($/kW*yr)

21.6 14.4 13.7 26.7 24 124 22 24

Cold Startup Cost ($)*

0 0 4,748 3,000 0 0 - -

FOM ($/hr)* 125 170 - 165 80 166 - -

Variable O&M($/MWh)

0.6 0.27 0.10 0.60 0.35 1.6 - -

CTs, SC, CC for plants considered in Anaconda, MT

47 ascend analytics

HDR Storage Unit Characteristics and Operating Costs

Source: HDR

CAES Li-Ion Flow BatteryPumped Storage

Capacity (MW) 100 25 25 400

Max Storage(MWh)

600 100 100 3600

Heat Rate (btu/kWh)/ Leakage Rate(%/hr)

4500 (heat) 0.05 (leak) 0.002 (leak) -

Charge/Discharge VOM($/MWh)

18 / 21 18 / 7 18 / 0 18/0.56

ChargeEfficiency (%)

190 85 73 80

Charge Rate (MWh/hr)

100 25 15.4 100

Allowed Ancillary Contribution(MW)

100 25 25 400

48 ascend analytics

Stand-Alone Battery

▪ Provides capacity and more efficient energy to the system by substituting for conventional power plants in providing for ancillary services. ▪ The conventional power plants no longer have to back down from their most efficient

operating point to hold capacity in reserve for flexibility

Constraints for optimization:

Track energy stored in battery: 𝑆𝑂𝐶𝑡 = 𝑆𝑂𝐶𝑡−1 + 𝐶ℎ𝑎𝑟𝑔𝑒𝑡 − 𝐷𝑖𝑠𝑐ℎ𝑎𝑟𝑔𝑒𝑡

Power Limit: 𝐶ℎ𝑎𝑟𝑔𝑒𝑡 ≤ 𝑃𝑜𝑤𝑒𝑟𝑀𝑎𝑥, 𝐷𝑖𝑠𝑐ℎ𝑎𝑟𝑔𝑒𝑡 ≤ 𝑃𝑜𝑤𝑒𝑟𝑀𝑎𝑥

Energy within capacity: 𝑆𝑂𝐶𝑡 ≤ 𝑆𝑡𝑜𝑟𝑎𝑔𝑒𝑀𝑎𝑥

-300

-100

100

300

500

1

39

77

11

5

15

3

19

1

22

9

26

7

30

5

34

3

38

1

41

9

45

7

49

5

53

3

57

1

60

9

64

7

68

5

72

3

76

1

79

9

83

7

RT

LM

P (

$/M

Wh

)

Chart Title

RT Energy Price SOC

Charges at half power when energy is cheap

Li Ion

Capacity (MW) 25

Max Storage(MWh)

100

Heat Rate (btu/kWh)/ Leakage Rate(%/hr)

0.05 (leak)

Charge/Discharge VOM ($/MWh)

18 / 7

Charge Efficiency (%)

85

Charge Rate (MWh/hr)

25

Allowed Ancillary Contribution (MW)

25

-300

-100

100

300

500

1

39

77

11

5

15

3

19

1

22

9

26

7

30

5

34

3

38

1

41

9

45

7

49

5

53

3

57

1

60

9

64

7

68

5

72

3

76

1

79

9

83

7

RT

LM

P (

$/M

Wh

)

Chart Title

RT Energy Price SOC

Battery 1: 40 MW 4-hour Battery 3: 160 MW 1-hour

1-hour battery captures more volatile price spikes, increasing revenues

Discharges when energy is expensive

49 ascend analytics

Internal Combustion Engine (ICE)

▪ Co-optimized operation between energy and ancillary services▪ Advantages of an ICE include it’s fast ramping time (5 minutes or less) and no start up cost

▪ Constraints for optimization:

Ramp rate limit: 𝐺𝑒𝑛𝑡 − 𝐺𝑒𝑛𝑡−1 ≤ 𝑅𝑎𝑚𝑝𝑅𝑎𝑡𝑒

Power within capacity: 𝐺𝑒𝑛𝑡 ≤ 𝑃𝑜𝑤𝑒𝑟𝑀𝑎𝑥

Regulation bounded by operating power: 𝑅𝑒𝑔𝑡 ≤ 𝐺𝑒𝑛𝑡, 𝑅𝑒𝑔𝑡 ≤ 𝑃𝑜𝑤𝑒𝑟𝑀𝑎𝑥 − 𝐺𝑒𝑛𝑡

0

10

20

30

40

50

60

70

80

90

100

MW

100 MW plant does 50 MW of regulation and 50 MW of energy

Minutely Time-Scale

Internal Combustion Engine (18V50SG)

Nominal Capacity (MW)

18.5

Heat Rate (Btu/kWh)

8,350

Cold Startup Time (hr)

0.083

Ramp Rate (MW/hr)

948

Min. Down Time (hr)

0.25

Fixed O&M ($/kW*yr)

24

Cold Startup Cost ($)*

0

FOM ($/hr)* 80

Variable O&M($/MWh)

0.35

50 ascend analytics

Pumped Hydro Storage

▪ Co-optimized operation between energy and ancillary services and provide capacity. Economics driven by the spread between low and high prices through the day.

▪ Constraints for optimization:

Ramp rate limit: 𝐺𝑒𝑛𝑡 − 𝐺𝑒𝑛𝑡−1 ≤ 𝑅𝑎𝑚𝑝𝑅𝑎𝑡𝑒

Track water stored in reservoir: 𝑉𝑜𝑙𝑡 = 𝑉𝑜𝑙𝑡−1 + 𝑃𝑢𝑚𝑝𝑒𝑑𝑡 − 𝐷𝑖𝑠𝑐ℎ𝑎𝑟𝑔𝑒𝑑𝑡Power and energy within system capacity: 𝑉𝑜𝑙𝑡 ≤ 𝑉𝑜𝑙𝑀𝑎𝑥, 𝐺𝑒𝑛𝑡 ≤ 𝑃𝑜𝑤𝑒𝑟𝑀𝑎𝑥

Pump when extra energy is available from midday solar

-100

-80

-60

-40

-20

0

20

40

60

80

100

MW

Discharge in morning and evening ramps

Perform some regulation all day

Minutely Time-Scale

Pumped Storage

Capacity (MW)

400

Max Storage(MWh)

3600

Heat Rate (btu/kWh)/ Leakage Rate (%/hr)

-

Charge/Discharge VOM($/MWh)

18/0.56

ChargeEfficiency (%)

80

Charge Rate (MWh/hr)

100

Allowed Ancillary Contribution(MW)

400

51 ascend analytics

Energy Efficiency and Demand Response

• Energy efficiency and demand response are playing an increasing role in resource portfolios. NorthWestern seeks to evaluate demand-side resources on the same playing field as supply resources.

• Technological progress, especially in LED lighting, provides opportunity for significant cost-effective savings.

• Internet technology (“grid-of-things”) provides a new level of controllability of demand response, particularly smart thermostats and smart water heaters.

• Modeling EE• EE programs are inputs to PowerSimm, modeled basically as negative load.

• In the final analysis, we compare EE program cost with benefits. The value of EE lies in avoiding energy (including losses), and capacity. Other benefits include avoided T&D spend, carbon benefits, RPS benefits, improved energy performance for customers, etc.

• EE is primary an energy resource, making less cost-effective in a future with low energy prices.

• Modeling DR• Demand response programs (AKA “interruptibles”) save energy for short durations by either

customer behavior or automated response through smart devices initiated by utility signals.

• DR programs modeled as having limits on capacity, duration, and # of interruptions.

• DR is a capacity resource, which is more cost-effective in an era of high renewables but scarce firm dispatchable resources.

52 ascend analytics

Future Portfolio Design Process

Assess RPS Need

Build future portfolios to meet need

Candidate resources

Load Forecast RPS assumptions

Evaluate peak capacity need

Add batteries and/or ICEs

Evaluate flexible capacity needs

Meets reliability,

RPS? No

Production cost

NPVYes

Market Prices

53 ascend analytics

In Front of the Fence: Utility Use of Batteries

54 ascend analytics

Bulk Benefits Analysis of Batteries for Balancing Authorities

Battery benefits consist of two key components▪ Avoided fuel cost from batteries providing regulation services

• More efficient operations of thermal generation

• Avoid start-up costs (e.g. save $8,000 on start-up of a CT)

Typical heat rates for combined cycle and coal generators

55 ascend analytics

Battery Valuation in EIM (5-min market)

~$23M /yr in additional energy value participating in RT 5-min market over DA hourly makes this best option

Battery charges when prices are low, and discharges when prices are high

56 ascend analytics

$59.11

($47.59)

$11.52

$103.24

($63.96)

$39.28

$185.68

($96.53)

$89.15

Payback: ~8 yearsPayback: >10 years Payback: ~4.5 years

Long duration batteries not economic as dependable capacity today

Battery Resource Options installed in 2020 – EIM node

57 ascend analytics

Sizing: Bookend expected merchant revenues using different market strategies to determine optimal battery duration

57

• The results for the three battery sizes at the Huntley node are above, along with revenue projections between the bookends.

• The most optimal battery size is 30 minutes.

0

50

100

150

200

250

300

350

30 min 1 hour 4 hour 30 min 1 hour 4 hour 30 min 1 hour 4 hour 30 min 1 hour 4 hour 30 min 1 hour 4 hour

Case 1: RTE Only Case 1: DA Reg Only Case 4: Marketparticipation rules bymonth for RT Energyand DA Regulation

Case 5: Co-optimization for RT

Energy and DAAncillaries

Case 6: FullOptimization

$/K

Wh

Annual Revenue per Installed KWh

DA Energy RT Energy DA Reg RT Reg DA Spin RT Spin

58 ascend analytics

Capturing declining cost-curves of batteries

Ascend survey values: annual lithium-ion battery price

Ascend survey results of battery manufacturers from December 2017

BNEF observed values from June 2017

Shorter duration storage (< 2hr) is cost-effective today with battery module prices at $195/kWh

Longer duration storage (> 4 hr) is projected to be cost-effective by 2025 when battery modules are priced at $100/kWh or less (as low as $50 in some projections1)

Sub-hourly market access is critical for realizing value of batteries to provide energy and ancillary services

1 Enovation Partners Analysis https://enovation-analytics.squarespace.com/insights-blog/2018/storage-cost-trajectories-drivers

Bloomberg New Energy Finance Projection of Battery Module Costs

Appendix

60 ascend analytics

Flexible Resource: Day-Ahead and Real-Time Operating Example

• Flexible generator maximizes profits relative to RT prices

• Real-time price > Strike price →run generation

• Real-time price < Strike price → buy generation from market to meet day-ahead offer when Real-time price

• Day-Ahead and Real-time markets

0

50

100

150

200

250

300

-1 4 9 14 19 24

$/M

Wh

Hour

Energy Prices

Day-Ahead

Real-Time

Generation Strike Price

Shut down period

Run period

61 ascend analytics

Realizing the Value of Flexible Generation

• Traditional generator valuation:• Supplier of energy in either the Day-ahead (DA)

• May use adder for ancillary revenue

• Analysis is done on an hourly basis

• Valuing flexible resources in this manner considerably undervalues the asset

• Correctly valuing flexible generation assets:• In a resource planning setting, in order to correctly value flexible assets methods are needed to co-

optimize between energy and ancillaries on a sub-hourly time-scale

• In a market setting, flexible generation needs to have the ability to capture sharp swings in real-time prices through joint optimization of energy and ancillary services for both DA and RT markets

62 ascend analytics

Flexible generation participating in wholesale power markets

• Four scenarios for flexible generation to participate in wholesale energy and ancillary markets:

1. Sell energy into the DA and RT energy markets

2. Sell regulation into the DA and RT markets

3. Sell regulation and energy into the DA and RT markets

4. Same as “3” plus physical hedge to support DA premium for renewable generation

63 ascend analytics

Flexible Resource: Operations to Energy & Ancillary Prices

•Energy and regulation 5 min real-time prices are volatile

• Flexible resources extract value from price volatility• Shut-down when prices low

• Start-up when prices high

• Objective: Jointly optimize energy and ancillary service value

s.t. operating constraints

• Strategy: Realize full extrinsic value in the real-time market.

Shut down period

Run period

64 ascend analytics

Day-AheadSchedule

Day-AheadSchedule

Energy Prices: Day-ahead (hourly) and Real-time (5-minute)

0

200

400

600

800

1000

1200

0 3 5 8 11 14 16 19 22

En

erg

y P

rice

($/M

Wh)

Hour of Day

DA and RT Energy Prices ($/MWh)January 14, 2015

RT Energy DA Energy

Day-Ahead Generation Offer Price

Additional RevenueFrom RT Price Spikes

SERVE NATIVE LOAD

SERVE NATIVE LOAD

65 ascend analytics

Optimal Battery Dispatch

30-minute battery optimization Case 5• Battery participates in

energy market when price spikes occurs and otherwise in regulation.

• Battery SOC in regulation decreases due to charging and discharging losses, and the operation is constrained to keep SOC away from 0% and 100% while in regulation to aid in battery life.

Energy Price

Regulation Price

Market Transactions

Battery SOC

100%

50%

0%

Regulation “Shadow” Price

Delivering a Bright Future