Rebecca Widiss and Kevin Porter Exeter Associates, Inc. SPSC
Transcription
Rebecca Widiss and Kevin Porter Exeter Associates, Inc. SPSC
Variable Generation Forecasting in the West Rebecca Widiss and Kevin Porter Exeter Associates, Inc. SPSC Webinar December 18, 2013 Today’s Webinar Project Overview VG Forecasting Basics Costs & Metrics Forecasting Uses Forecasting Practices Solar Forecasting Lessons Learned & Next Steps Photo Credit: Peter Haas via Wikimedia Commons http://commons.wikimedia.org/wiki/File%3APower_Lines_DSC1993w.jpg Project Overview Project History ▪ Project an update of a report that was done for the SPSC, NREL and DOE in 2011 and published by NREL in 2012: www.nrel.gov/docs/fy12osti/54457.pdf ▪ Intent was to interview several operating entities (OEs) in the West about their approach to variable generation (VG) forecasting ▪ Funding provided by the U.S. Department of Energy’s Office of Electricity Delivery and Energy Reliability, technical support provided by NREL 4 Current Project ▪ Interviews with thirteen operating entities ▪ Twenty-eight questions in the survey covering: • • • • • Needs assessment Forecast timeframes, scopes, types, and sources Probabilistic forecasting Incorporating curtailments and outages Data collection requirements and challenges • • • • • • The impact of FERC Order • 764 or sub-hourly scheduling • System operator training Control-room integration • Forecast costs Cost-benefit analysis Reductions to operating reserve requirements Accuracy measurements Utility-scale and “Behind-theMeter” solar forecasting Coordination between operating entities ▪ Three deliverables: report, summary table, and webinar ▪ We gratefully acknowledge the continued support of DOE and technical assistance of NREL 5 Interviewees ▪ Alberta Electric System Operator (AESO) ▪ Arizona Public Service (APS) ▪ Bonneville Power Administration (BPA) ▪ California Independent System Operator (CAISO) ▪ Glacier Wind ▪ Idaho Power ▪ Portland General Electric (PGE) ▪ Puget Sound Electric (PSE) ▪ Sacramento Municipal Utility District (SMUD) ▪ Southern California Edison (SCE) ▪ Turlock Irrigation District (Turlock) ▪ Xcel Energy ▪ Pacific Gas and Electric (PG&E) 6 VG Levels and Forecasting Practices Glacier Wind Idaho Power PG&E* PGE PSE n/a** 1,759 n/a*** 2,140 4,328 5,660 399 669 n/a 550 823 0 2,700 137**** 2,215 6 3,263 0 2-3 n/a 2 0.5 150 1,400 3 390+ 2008 2009 2004 2009 2011 2000s 2007 2007 2011 1980s 2009 2008 X X X X X X X X X X X To come X Operating Entity AESO APS BPA Average Load 8,604 4,500 6,000 Wind Capacity 1,088 290 4,516 Solar Capacity 0 481 2010 X Year Forecasting Began Wind Forecast Solar Forecast X CAISO 21,57935,781 X *PG&E, SMUD, and SCE also receive PIRP forecasts from CAISO **Glacier Wind has no load SMUD* SCE* Xcel Turlock Energy 1,200 13,000*** 245-336 X X 4,000 To come ***Load, wind, and solar capacity for SCE and PG&E included in CAISO totals ****Turlock’s Tuolumne Wind Project is in BPA’s service area ▪ Variable generation is rising swiftly ▪ SCE had 383 MW of solar online in 2011; that number has more than tripled ▪ In December 2013, the CO PUC approved Xcel’s acquisition of 450 MW of wind and 170 MW of solar at prices equal to or below the company’s cost of conventional generation 7 VG Forecasting Basics Common Forecasts and Their Applications Type Content Weather Situational Provides severe weather alerts Awareness Primary Use Important because storms can lead to rapid changes in VG Intra-day Typically provides 10-minute power Helps grid operators anticipate values for next few hours near-term changes in VG and take appropriate action Day-ahead Provides hourly power values for the next few days Often used in the unit commitment process Nodal Aggregates forecasts for each node or transmission delivery point Helpful in transmission congestion planning Persistence Simply assumes that current output Useful for very short-term levels will continue decisions (<1 hour) Ensemble Aggregates output values from two or more forecasts Helps compensate for the error inherent in any forecast 9 The Basic Steps to Creating a Wind Forecast ▪ Large-scale Numeric Weather Prediction (NWP) models predict weather conditions ▪ Run by agencies such as NOAA; used for a wide variety of applications ▪ Tend to have limited spatial resolution ▪ Local weather data gathered to supplement NWP ▪ Statistical models used to account for subtle influences of local terrain Image credit: National Center for Atmospheric Research (http://www.mmm.ucar.edu/prod/rt/wrf/wrf20/2013121112/wsp.hr18.png) 10 A Few Notes on Solar Forecasting ▪ At a very early stage, where wind forecasting was 10 years ago ▪ Clouds, water vapor, and aerosols affect how much solar radiation reaches the earth (insolation) ▪ Hour-ahead forecasts tend to rely on statistical models, satellite images of water vapor channels, and sky imagers ▪ Day-ahead forecasts tend to use physical models that simulate atmospheric processes Photo credit: Hong Kong Observatory (http://www.hko.gov.hk/wxinfo/aws/imager/sky.htm) 11 Improving Accuracy ▪ Gathering data from more sites, especially at hub heights ▪ Improving model formulas, training models with historic data ▪ Shortening scheduling intervals and updating forecasts more frequently ▪ Improving staff understanding of and access to forecasts Plot of System-Wide Wind Forecast Error Versus Forecast Time Horizon, with Error Expressed as Mean Absolute Error as a Percentage of Installed Wind MW 12 Costs & Metrics Costs ▪ Wind forecast costs have gone down dramatically since 2011 ▪ In 2011, one of the few OEs willing to share hard numbers was paying $1,000 per plant per month ▪ In 2013, several respondents confirmed a ballpark cost of $300$400 per month per wind plant ▪ Declining costs foster a “shopping” mentality Action Operating Entity Using multiple vendors • • • • • Running trial with multiple vendors • BPA • Glacier Wind • SMUD BPA Glacier Wind PGE PSE SCE 14 Cost-Benefit Analysis ▪ There has been a striking shift away from C-B analysis since 2011 ▪ Half of interviewees say it simply is not worth the effort—VG forecasting is so clearly beneficial: ▪ It’s a no-brainer – AESO ▪ Anyone with large amounts of renewables should forecast – PG&E ▪ One day of less hedging probably takes care of the yearly cost of forecasting – PGE ▪ Xcel tracks the value of a 1% reduction in forecast Mean Average Error by market Region Annual Value Public Service Company of CO $1,300,000 Southwestern Public Service Co. $250,000 ▪ SCE says its ROI significant, PG&E plans formal C-B analysis, SMUD interested 15 Accuracy ▪ Almost every respondent says accuracy is on the rise—a few shared their track records: Operating Entity 2012 Report Current Report AESO Day-ahead MAPE: 13.0% Day-ahead MAPE: 12.8% CAISO Day-ahead MAE: <15% Day-ahead MAE: <10% Glacier Wind Hour-ahead MAE: 10% better than persistence, at best Hour-ahead MAE: 20-25% better than persistence Idaho Power Day-ahead MAE: 12% Day-ahead MAE: 13% SCE Day-ahead RMSE: 13-20% Day-ahead RSME: 8-13% ▪ Interviewed OEs credit gains to improved forecasting techniques, seasoned vendors, experienced users, trained models, and growing portfolio size 16 Reserve Requirements ▪ Many interviewees believe that VG forecasting has led to lower operating reserve requirements (than if such forecasting were not put into place) ▪ Some anecdotal evidence for this: ▪ AESO buying the same reserves as five years ago, while its wind portfolio has doubled ▪ But no interviewee provided formal quantification The Damsel of the Holy Grail Image credit: Dante Gabriel Rossetti via Wikimedia Commons (http://tinyurl.com/lv26jen) 17 Forecasting Uses Core Applications of Forecasting ▪ Almost every OE interviewed has expanded forecasting uses since 2011 ▪ Biggest change: the use of forecasting in forward unit commitment has more than doubled! Forward Unit Commitment Intra-Day Unit Commitment Planning Resources Hydro, Coal, or Gas Storage Mgmt. Gen./Trans. Outage Planning 0 2 4 6 8 10 12 19 Additional Applications of Forecasting ▪ Unique circumstances lead to unique uses: ▪ In 2014, wind generators will be allowed to use AESO’s short-term forecast to bid into its energy market ▪ BPA uses wind forecasting to meet non-power objectives (e.g. irrigation, flood control, fisheries preservation) for the federal hydro facilities it manages ▪ Because its wind project is outside its service area, Turlock uses wind forecasts for trading, marketing, and optimizing schedules Image credit: Tom Patterson, NPS via Wikimedia Commons (http://tinyurl.com/pwlvvw8) ▪ If high wind is predicted, Xcel Energy will reduce day-ahead gas purchases 20 Forecasting Practices Types and Timeframes ▪ Most of the surveyed OEs use a suite of forecasts… ▪ Persistence (12 OEs) ▪ Numerical Weather Prediction (12) ▪ Statistical (12) ▪ Weather Situational (8) ▪ Ramp (2) Spanning multiple timeframes… 14 12 10 8 6 4 2 0 22 Scopes ▪ … With broader geographic coverage than in 2011 Individual Plant ▪ “Region” is a self-defined term ▪ CAISO looking across Western Interconnection ▪ SCE looking to develop forecasts by wind region (e.g. Tehachapi, San Gorgonio) Utility / Balancing Area ▪ PGE views the forecast BPA provides as regional Commercial Pricing Node Region 0 5 10 15 ▪ Xcel forecasts for commercial pricing nodes within MISO 23 Data Collection ▪ Few changes since 2011 in types of wind data being gathered: Operating Entity Glacier Idaho Wind Power AESO APS BPA CAISO PG&E PGE SMUD SCE Turlock Xcel Wind Speed/Direction X X X X X X X X X X X X Temperature X X X X X X X X X X X X Barometric Pressure X X X X X X X X X X X Turbine Location X X X X X X X X X X X X Turbine Power Output X X X X X X X X X X X X Turbine Availability X X * * X X X X * X X X Turbine Outage X * * * X X X X X X X Turbine Power Curve X X ** X ** X X X X ** * Receives info from wind plant in total ** Derives an empirical wind curve ▪ AESO, BPA sped up data transmission from generators ▪ Interviewees experimenting with Doppler radar, Lidar, extreme temperature notifications, etc. 24 Probabilistic Forecasting Operating Entity AESO APS BPA CAISO Glacier Wind Idaho Power PG&E PGE PSE SMUD Turlock Xcel Use of Ensemble Forecasts X X Discontinued Use of Confidence Intervals X X To come X X X To come To come X To come In Trial X X X X X X X Confidence Interval Range 10% and 90% ▪ Ensemble forecasts and confidence intervals (CIs) popular in theory ▪ Challenge lies in interpreting them X Considering 90% ▪ AESO uses likeliest value, not CI ▪ SMUD looks for forecast “unanimity” ▪ Xcel working on new way to combine uncertainties from multiple forecasts 90% 80% 80% 80% 75% ▪ BPA using “Super Forecast” methodology to select best forecast for each plant for every hour 25 Curtailments and Outages What info do you incorporate into your forecasts? Just outages/ availability Neither ▪ Nearly every OE interviewed uses turbine outage/ availability information to project full (potential) value of wind plants ▪ If curtailment info available, it is used for calculating metrics and for statistical training of forecasts O/A & Curtailments 26 Control Room Integration ▪ VG forecast displays are nearly universal ▪ Typically automated feeds, often alongside current weather and generation ▪ Roles and protocols changing ▪ CAISO combined load and VG forecasting ▪ Xcel has forecast-based protocols for curtailment, cycling, unit decommitment Photo credit: Design-Build Institute of America (http://www.dbia.org/awards/Pages/2012-Award-Winners.aspx) 27 Control Room Integration (cont’d) ▪ 1/3+ of respondents integrating forecast values into operations tools: ▪ AESO – values fed into dispatch decision and short-term adequacy tools ▪ APS – values fed into dispatch and scheduling tools ▪ CAISO – developing ramping tool to gauge near-term needs using load, VG forecast, resource commitments, etc. ▪ BPA – values fed into Real-time Reserves Requirement Tool Image credit: Jorge Maturana via Wikimedia Commons (http://tinyurl.com/mesm6uq) ▪ Idaho Power – day-ahead forecast fed into EMS ▪ PGE – working on tool to facilitate tighter scheduling periods 28 Staff Training and Familiarity ▪ Formal training is rare ▪ APS summer prep session Are operators familiar with their VG forecasts? ▪ CAISO 6-wk orientation, biannual course ▪ More often, staff coach one another informally ▪ Three respondents have meteorologists on staff to help interpret forecasts (Idaho Power, PSE, Xcel Energy) Yes Getting there ▪ CAISO looking to add a meteorologist 29 Solar Forecasting Utility-Scale Solar Forecasting ▪ At a very early stage ▪ Unlike wind, some opting to develop in-house solar forecasts (APS, Idaho Power, CAISO, SCE) ▪ No consensus on how to do so, nor how easy it will be ▪ Data requirements are still fairly minimal ▪ Combo of: solar insolation, availability, weather info, back panel temp, system characteristics, power output ▪ SMUD testing four different commercial providers Photo credit: Chandra Marsono via flickr (http://tinyurl.com/k3rscnj) 31 “Behind-the-Meter” Solar Forecasting ▪ If solar forecasting is in its infancy, DG forecasting is at an even more rudimentary stage ▪ Significant interest and concern expressed during interviews ▪ Locating DG systems is Step 1 ▪ APS mining in-house records ▪ CAISO working with CEC and Clean Power Research Is DG solar forecasting a priority? Imminent need Eventual need Not yet a concern ▪ PG&E using data from CA Solar Initiative 32 Lessons & Next Steps Advice and Preferences Advice for Those Starting Out: Preferences for the Region: ▪ Start early ▪ Interviewees split on sharing forecasts ▪ Set realistic expectations ▪ Buy/create what you need ▪ Put data requirements into contracts ▪ Competitive advantage issue ▪ Question of how useful it would be ▪ Also split on the value of guidelines ▪ Use several performance metrics ▪ Fear of “one-size-fits-all” rules ▪ Evolve with your forecast ▪ Yet interest in universal resource adequacy standard for reserves 34 Trends, Challenges, Recommendations Common Trends Common Challenges Possible Next Steps ▪ Increasing accuracy of, and confidence in, wind forecasts ▪ Forecasting the timing and size of ramps ▪ Use of forecasts in short-term dispatch ▪ Handling uncertainty ▪ Integrate forecasting with faster scheduling and balancing area coordination ▪ Rapidly decreasing wind forecast costs ▪ Early development of solar forecasts ▪ Integration of forecasts into operational tools and processes ▪ Forecasting for DG solar ▪ Further improving forecast accuracy ▪ Further integrating forecasts into grid operations ▪ Reduce confidence intervals? ▪ Improve national forecast models 35 Questions? Rebecca Widiss Kevin Porter [email protected] [email protected] 410-992-7500 410-992-7500 Photo credit Dennis Brekke via flickr (http://tinyurl.com/mbvdzkh) 36