Best Practices in Risk-Based Resource Planning A Case Study of NorthWestern s Acquisition of Hydros Presenters: Gary Dorris, PhD August 4, 2015
Overview of Ascend Short, Intermediate, & Long-Tem: operating and planning analytics Founded in 2002, continuing steady growth, currently with over 30 employees Offices in Boulder CO, Oakland CA, and Bozeman MT Decision Analysis Timeframe One Model Through Time Operational Strategy (PowerSimm OPS) Intermediate Analytics (PowerSimm Portfolio Manager) Long-term Planning (PowerSimm Planner) Today 10 days 1 month to ~5 years ~6 years to 30+ years Optimal short-term dispatch Short-term load & price forecasting Decision support for trading Portfolio management Energy purchases and sales CFaR, GMaR, EaR Power Planning, IRP Asset valuation Cost versus risk tradeoff resource analysis Renewable integration studies 2 ascend analytics
Case Study in Resource Planning NorthWestern Energy 3 ascend analytics
Ascend Planning $1 billion utility hydro acquisition in 2014 for NorthWestern Energy Ascend provided economic and physical modeling of Hydros In Evergreen s opinion, NorthWestern s efforts are fully consistent with industry best practices 4 ascend analytics
Best Practices in Resource Analysis Portfolio perspective Fuel diversity Reliability and intermittent resources Uncertainty in fuel, emissions, and power Use of market for fuel and power Analyze generation resource options with respect to market dynamics Capture new dynamics Weather Load Wind generation Market conditions Forward markets for fuel and power Spot price dynamics Resource adequacy Reliable service at cost effective rates Market interactions 5 ascend analytics
Need for New Tools to Incorporate Uncertainty Deterministic vs. Stochastic Models Heavy dependence on deterministic results with scenarios The likelihood of result is not understood Model inputs are variable and interdependent Deterministic modeling misses critical scenarios producing an inconsistent value What s the impact of unused information? Inaccurate forecasting Assessing risk becomes difficult One outcome for deterministi c 6 ascend analytics
PowerSimm use in NorthWestern Energy IRP 7 ascend analytics
Key Points of Analysis Supply Cost and Risk = costs and uncertainty in energy supply costs (fixed and variable) Risk Premium = monetized value of risk uncertainty in supply costs Net Position Exposure = balance of NWE resources relative to load obligations Supply portfolios to be assessed: Market purchases Purchase of PPL hydro (494 MW) Combined cycle (239 MW) Optional Analysis Reliability of supply Flexibility function 8 ascend analytics
Acquisition of PPL Hydro Assets Generator MW Net Capacity Factor (%) Storage (A/F) Thompson Falls 94 60 14,970 Mystic Lake 12 50 Kerr* 194 64 200,000 Madison 8 85 27,200 Hebgen - - 386,000 Hauser Lake 19 81 64,253 Holter 48 75 82,000 Black Eagle 21 73 1,820 Rainbow 60 60 1,050 Cochrane 69 50 2,700 Ryan 60 80 5,000 Morony 48 68 2,700 Total 633 65 787,693 *Kerr generation rights end in 2015 9 ascend analytics
Net Position Report MWh/year 2,000,000 1,000,000 - (1,000,000) (2,000,000) (3,000,000) (4,000,000) (5,000,000) (6,000,000) Annual Net Position Current Current + CC Current + Hydro 10 ascend analytics
Main results: Range of likely costs through time Over the 30 year study horizon, the hydro asset reduces the exposure of NWE ratepayers to price spikes and narrows the range of potential costs 1,400 Annual Supply Costs 1,200 1,000 $Millions 800 600 400 200-2015 2017 2019 2021 2023 2025 2027 2029 2031 2033 2035 2037 2039 2041 2043 Current P5 P95 Current + CC P5 P95 Current + Hydro P5 P95 $455M $403M $289M 11 ascend analytics
CO2 price distribution In this analysis, CO2 prices are modeled beginning in year 2021 CO2 prices in $/tonnes are modeled using a triangular distribution with a miniumum of 0, a mode of 21, and a maximum of 42 The mode is consistent with the EIA forecast for CO2 prices As shown in the figure below, these CO2 price assumptions are conservative relative to other plans 12 ascend analytics
Distribution of Costs How do we measure the difference in risk between the portfolios? RP measures the dollar value of risk. Effective cost = Risk premium (RP) + cost Consider: Base = has higher costs and more risks Base+CC = adding CC slightly reduces risk Base+Hydro = reduces costs and substantially reduces risk to market prices of fuel, power, and potentially CO 2 13 ascend analytics
NPV Total Cost of Supply 30 Year Risk Premium adds to the cost advantage of Current + Hydro Cost of Supply = Fuel Cost + Variable O&M + Emission Costs + Market Purchases Market Sales + New Gen Capital Costs 14 ascend analytics
Aggregation & Monetization of Risk 15 ascend analytics
Risk Premium Definition Risk Premium captures the expected value of the upper tail of the cost distribution for each portfolio Similar to established means of valuing a financial option, or an insurance policy c = Expected cost Probability p 1 Risk premium is the probabilityweighted average of costs exceeding the mean Risk premium = i=1 (c i c ) p i p i : Probability of cost c i k p k c 1 Total portfolio costs $M c k 16 ascend analytics
Risk Premium Calculation Risk Premium (RP) monetizes the portfolio risks Risk Reduction Value = RP(Current of $451M) RP(Current + Hydro of $247M) = $204M NPV of risk premium $Millions $500 $450 $400 $350 $300 $250 $200 $150 $100 $50 $- $451 $380 $247 Current Current + CC Current + Hydro 17 ascend analytics
NPV Total Cost of Supply 30 Year Risk Premium adds to the cost advantage of Current + Hydro Cost of Supply = Fuel Cost + Variable O&M + Emission Costs + Market Purchases Market Sales + New Gen Capital Costs 18 ascend analytics
Definition of Common Planning Risk Metrics Cost at Risk (CaR) for Rev Req = CaR = 95 th percent costs mean costs Standard deviation (Std) of Ann Rev Req (ARR) = Std = E[AAA 2 ] (E AAA ) 2 How do you decide where to be on the frontier? Avista IRP 2011 Timing: Avoid roll-up effect over long time horizons CaR of Rev Requirements (standard risk metric) Ave annual over study period Max annual over study period Reflect 1 in 20 events Standard deviation (66 th percentile) year over year change in Rev Requirements Used by Avista, PacifiCorp Reflects 1 in 3 events 19 ascend analytics
Simulations Realizing Meaningful Uncertainty 20 ascend analytics
Integrating Physical and Financial Uncertainty Unified simulation framework reflecting joint financial and physical uncertainty o o Rigorous validation Capture of critical causal effects During delivery simulations Wx Sim Load Sim Renewables Spot Price Sim Calibrated Spot Prices Optimal Dispatch (Thermal, Hydro) Forward & forecast Prices Forecast Price Sim Power, Gas, Coal, Oil, Emissions,,Dalles, Supply & Transmission Portfolio Summarization Portfolio Selection Seasonal Hydro Sim 21 ascend analytics
Validation Criteria Simulated Risk Factor Validation Activities Forecast/ Forward Prices Spread of prices 5th, mean, 95th; rate of reversion, correlations, seasonality, implied heat rate Weather Match historic 5th, mean, 95th temps, patterns and time-series pattern Load Match 5th, mean, 95th hourly and monthly load of customer system Hydro Flows Match 5th, mean, 95th flows, pattern, incorporate current year forecasts Spot Prices Gas Match uncertainty of 5th, mean, 95th and preserve relationship with temperature, load, renewables Spot Prices Power Match uncertainty of 5th, mean, 95th with gas, price spikes, key explantory variables of ERCOT load, gas prices, and renewables 22 ascend analytics
Forward Price Validation History and Simulations: Natural Gas Simulated forward market prices for Henry Hub Large price spikes followed by mean reversion 2 to five year cycle Large volatility in prices followed by quiescent periods Forward Price Validation- Price Paths for Final Evolved Forward Curve Simulation Historic prompt month prices for Henry Hub Large price spikes followed by mean reversion 2 to five year cycle Large volatility in prices followed by relative quiescent periods 23 ascend analytics
Confidence Intervals and Simulation Speed Forward Price Simulations Simulated forward prices are plotted over time for the mean, 5 th, and 95 th percentiles Expect that uncertainty will grow with time Power price simulations spread from the mean F 0 =E(F t )=E(F T )=E(S t ) Confidence Intervals for Power Price simulations 5 th, mean, 95 th Forward Price Validation Tests uncertainty in the distribution of simulated forward prices Uncertainty grows over time Ranges of prices are consistent with market expectations and historic perspectives of forward price uncertainty 24 ascend analytics
Forecast Fuel Price Simulations Simulated forward prices are plotted over time for the mean, 5 th, and 95 th percentiles Seasonal changes in price and uncertainty Coal Gas Uncertainty grows with time Average price increases to account for inflation 25 ascend analytics
Weather Renewables Load Price Simulations Hydro Flows Intermittent Generation Gas Weather Load Electric Price 26 ascend analytics
Weather Overview Weather data taken from the National Climatic Data Center Publicly available at http://ncdc.noaa.gov 10+ years of historical data More than 30 weather stations across Montana 27 ascend analytics
Weather Load Relationship Weather-Load Validation Simulated vs. Historical Maintaining Correlations Incorporating weather into the load model maintains integrity in the weather load relationship Simulations nicely smooth out bumps of historical weather record Simulations provide for new extreme values to exceed historic record Validating Relationship Validate by capturing the weather load relationship in the historical period and simulated backcast The structural state space modeling captures the changes in shape with changes in load 28 ascend analytics
Load Validation Load Confidence Intervals Confidence intervals for hourly load by month at the mean, 5 th, and 95 th percentiles Alignment of simulations with historical data August January Daily Load Shape Morning and evening peak during cold months Single afternoon peak during warm months 29 ascend analytics
Spot Price Simulations Spot Price Simulations Daily historical market prices for Mid-C Heavy compared to NWE load Low correlation between NWE load and power prices. Mid-C February Confidence Intervals Hourly P10-Mean-P90 Confidence Intervals for February Mid-C power prices Historical in red, simulation in blue Good distributional agreement between simulated and historical data 30 ascend analytics
Site A: Hourly wind and load Wind generation declines as system load peaks Wind doesn t blow when you need it to Average of hourly load and wind generation Wind Capacityy Factor 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 3/31 0:00 3/31 6:00 3/31 12:00 3/31 18:00 800 700 600 500 400 300 200 100 - Load MW Gordon Butte Wind NWE Load 31 ascend analytics
Site B: Hourly wind and load Diversifying wind portfolio can make the wind blow when you need it to Adding more wind farms in the vicinity of wind rather than just one wind farm in the place where wind is most likely Average of hourly load and wind generation Wind Capacityy Factor 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 3/31 0:00 3/31 6:00 3/31 12:00 3/31 18:00 Judith Gap Wind NWE Load 800 700 600 500 400 300 200 100 - Load MW 32 ascend analytics
Wind Generation Validation Simulated wind production matches the historical behavior of NWE wind assets Wind is scaled to meet 15% of load in 2015 onwards 80% 70% Monthly capacity factor 60% 50% 40% 30% 20% 10% 0% 1 2 3 4 5 6 7 8 9 10 11 12 Month Historical data Mean P5 P95 33 ascend analytics
Solar Generation, Load, and Market Conditions Simulation of load, solar and market prices Position exposure Cost of supply exposure Solar Validation PowerSimm Modeling of Palo Verde Prices 34 ascend analytics
Hourly Portfolio Attributes PowerSimm simulated all assets for each hour of the 30 year study horizon, for each simulation iteration 35 ascend analytics
What s Next? Renewable Integration Studies Portfolio Integration of Hydros Optimal Capacity Expansion Planning 36 ascend analytics
Renewable Integration Examine Impacts of Renewable Additional regulation and load following requirements Cost of renewable integration Calculate CPS2 scores Regulation Used to meet the minute-by-minute system imbalances Correct for area control errors (ACE) Contingency Reserves Comprised of multiple products based on response time Spinning Reserves Non-Spinning Reserves 37 ascend analytics
Joint Thermal & Hydro Optimization of Energy & Ancillaries 38 ascend analytics
Deterministic capacity expansion optimization: Best athlete The best expansion plan for each scenario is akin to the best athlete for each sport. Which is the best plan for overall scenarios? Different conditions yield different optimal plans Objective: Find the cheapest capacity expansion plan that satisfies the constraints Cheapest is often defined as the net present value of the revenue requirements Constraints : Reliability, RPS requirements, ancillary services Result: Optimal capacity expansion of generation resources and conservation options to minimize revenue requirements subject constraints. Best swimmer/runner/cyclist is a sports analogy referring to the best expansion plan from a deterministic run (single forecast) of simulated weather, load, and renewables. Michael Phelps Best Swimm er Ryan Hall Best Runner Chris Froome Matt Biondi Alberto Salazar Tejay Van Garderen Ryan Lochte Steve Prefontaine Taylor Phinney Gary Hall, Jr. Galen Rupp Chris Horner Best Cyclist 39 ascend analytics
Stochastic capacity expansion optimization: Best triathlete PowerSimm Planner Result Best Expansion Plan for All Future Simulated States Requirements Robust simulations Advanced optimization of energy and ancillary services Advantages Simplifies decision choices Captures full tail of cost distribution Accounts for multiple future conditions Dave Scott Michael Phelps Ryan Hall Chris Froome Best Triathlete 40 ascend analytics
Thank You! U.S. Offices Headquarters: Boulder, CO 1877 Broadway Street Suite 706 Boulder, CO 80302 (303) 415-1400 Other Offices: Oakland, CA Providence, MA Bozeman, MT Gary Dorris President O:303.415.0311 gdorris@ascendanalytics.com 41 ascend analytics