Wind & Solar Energy Prediction: Challenges Of Opportunities

Wind & Solar Energy Prediction:Challenges of OpportunitiesSource: AMECMr. William P. Mahoney IIIDeputy Director, Research Applications LaboratoryNational Center for Atmospheric Research, Boulder, Colorado, USA

Mother Nature Is Not Always Kind11

Setting the Context - ScaleSource: Terra Magnetica22

Weather & Solar Energy Related Industry IssuesFinancingMaintenanceCostsVariability isCostlyPrediction ErrorsCostlyExtreme WeatherImpacts Wind energy resource estimates at wind farm sites areover-estimated on average Wind turbines are failing faster than predicted (up to40% earlier) Wind & solar power variability complicate powerintegration and load balancing across the grid – requiresreserves Wind energy prediction has typical errors of 10-15%(flat terrain) to 15-25% (complex terrain) Wind turbines are not designed to handle extremeweather conditions (shear, ice, snow, high wind, etc.).More representative weather datasets are needed forturbine design33

Overarching Wind EnergyScience Challenges Boundary layer meteorology (0 to 200 m above ground)is not well understood nor is this layer well measured The wind energy industry greatly under appreciates thecomplexity of the airflow in this layer The wind industry has historically assumed lessturbulence and more wind with height above the groundImage source: Wind Measure International44

Overarching Solar EnergyScience ChallengeImprove prediction needed of: Cloud lifecycleAerosolsJet ContrailsSurface conditions (snow/ice melt on solar devices) Cloud and precipitation processes are highly complexand operate on very small scales (10s to 100s ofmeters) Weather models greatly over simplify cloud physicsproperties and precipitation processes. Jet contrails can spread into a cirrus deck and are notpredicted by any models55

Current Meteorological Shortfalls forWind Energy Lack of wind, temperature, and stability measurementsbetween 10 and 200 m above ground Weather models not optimized for wind energy predictionand modeling across scales Need improved data assimilation techniques to takeadvantage of wind farm and other local observations Dearth of vertical observations offshore Lack of understanding of complex flows near the Earth’ssurface Ice and snow accretion and deposition prediction66

Current Meteorological Shortfalls forSolar Energy Lack of surface irradiance measurement covering a wide rangeof climates Lack of global water vapor measurements at high resolution(horizontal and vertical) Weather models not optimized for solar energy prediction;modeling across scales (synoptic to cloud scale) Inadequate data assimilation techniques to take advantage ofsolar farm solar and other local cloud observations Lack of full understanding of cloud physics and precipitationprocesses77

Examples of Complexity88

Predicting Inversions –Wind Decoupling60-80 mWarmer AirCold AirNocturnal inversion – Denver 19 September 201015 degree C difference over 1500 ftobservation10 m99

Low-Level Jets of High Wind (U.S. Midwest)Lidar (laser radar)measured windvelocitytoward lidar10 ms-1 ribbon of high speed airHeight(km)Distance (km)Courtesy, Robert Banta, NOAALow-level jet streams can damage wind generators10

Wind Variability at Turbine Height Can be SubstantialCourtesy Ned Patton, NCAR11

Influence of Stability on Low-Level FlowHorizontal slices of vertical velocityNear neutralStrongly unstableCourtesy Ned Patton, NCAR12

Wake Effects of Turbine ArraysTurbine wakes result inpower loss, turbulence,wind shear and overallwear and tear on theturbines drive trainsCourtesy Branko Kosovic, NCAR13

Wind Shear vs. Turbine EfficiencyKnowledge of the wind profile is important for wind to powerconversion – Shear across blades can reduce efficiency by up to 20%!(Lundquist and Wharton, 2009)Cold Front Wind RampLow-level Jet Wind Ramp(T. Aguilar, 2010)14

Wind Energy Ramps – Colliding Gust FrontsCollidingthunderstormgust fronts inTexasMahoney 198815

Wind Energy Ramp Events8/03/09 771mw up-ramp from 20:10 - 22:10 followed by a 738mw down-ramp from 22:40 - 00:50900800800 MW in 2 hrs.700cold front600500400Ridgecrestsmall thunderstormsSpring CanyonCedar Creek300Logan/Peetz TableColorado Green/Twin onnequin-100TIME16

Wind Energy Ramp NowcastingPredicting wind energyramp events using arapid cycle, highresolution weathermodel and Dopplerradar data.Animation of the Variational Doppler Radar Analysis System(VDRAS) covering eastern Colorado wind farms. Wind vectors andDoppler radar reflectivity are shown.17

Complex Flows – Offshore WindFor offshoreapplications itis important tocapture windand waveinteractionsMoving wavesPeter Sullivan, NCARWaves generate their own wind field that persists to hub height18

Hurricane Flow CharacterizationComplexitiesWRF Hurricane SimulationLarge-Eddy Simulation (LES)190 ft (62 m) resolutionResolving turbulence scalesHow do wind turbinesrespond to hurricanes,typhoons and USANor’easters?(Rich Rotunno, NCAR19

Icing Accretion and SnowDepositionIcing prediction and its impact onturbine performance is a criticalresearch topic.Source unknown4D aviation icing product, NCAR20

Wind Energy Prediction – Data FlowPredict windspeed at turbineheightPredict wind energyof each turbine usingmanufacturer orempirical powercurve algorithmsPredict electricalconnection nodepower by adding upgeneration capacityof each turbine usingpower curve data21

Optimizing Prediction by Blending TechnologiesEach technology has its own ‘sweet spot’ with respect to prediction skill.22

Research in Complex FlowsGlobalRegional4103Grid Cell Size [m] 10107Domain Size [m]106Adopted from Mike Robinson (DOE/NREL)23

Research in Complex FlowsGlobalRegionalLocal4103Grid Cell Size [m] 10107Domain Size [m]101106104Adopted from Mike Robinson (DOE/NREL)24

Research in Complex FlowsGlobalRegionalLocalTurbineBlade4103Grid Cell Size [m] 10107Domain Size [m]100101106104103Adopted from Mike Robinson (DOE/NREL)25

Fully-Coupled CFD/CSD for Turbine/Platform Interactionwith the Atmosphere and OceanObjective: To create a state-of-theart High- Performance Computing“Cyber Wind Facility” for therenewable energy industry andresearchers.Mesoscale Weather DataCourtesy Jim Brasseur, Penn State26

Atmospheric Science Research to SupportWind and Solar Energy Multi-year field experiments (on and off-shore)Boundary layer meteorology (complex flow)Cloud physics & precipitation processes (icing, snow, etc.)Turbulence characteristics and predictionComputational science (improve efficiency)Land surface condition predictionOcean dynamics (waves, currents)Aerodynamic studies related to turbine designMulti-scale modeling (global to millimeter scales)Future climate modeling (effects on wind/solar resources)27

Thank Youmahoney@ucar.edu28

Wind Energy Prediction – User Interface29

Xcel Energy Wind Energy Prediction SystemHardware – Deterministic Modeling System Four Dell 2950 servers: 2 CPUs (8 cores) at 2.66GHz, 8-16GBRAM, 2-4TB RAID5 Sixty-one Dell 1950 servers: 2 CPUs (8 cores) at 2.66GHz,4or8GB RAM, 250GB, Myrinet card One 64 port gigabit switch for network access to storage and useraccess to cluster One 64 port Myrinet switch for high speed MPI Network attached, 8-10TB of storage for model processing andoutput. Installed within 2 or 3 new Dell full height (42U) racks, with peripheralssuch as:8 port KVM switch, Dell 15” LCD console panel, power distributionunits, UPS 3000, all associated cabling.30

Xcel Energy Wind Energy Prediction SystemHardware – Ensemble Modeling SystemThree Dell 2950 servers: 2 CPUs (8 cores) at 2.66GHz, 16GBRAM, 2-4TB RAID5 Forty-two Dell 1950 servers: 2 CPUs (8 cores) at 2.66GHz,4or8GB RAM, 250GB One 64 port gigabit switch for network access to storage and useraccess to cluster Network attached, 12-16TB of storage, with potential ability toscale to 40TB Installed within 2 new Dell full height (42U) racks, with peripheralssuch as:8 port KVM switch, Dell 15” LCD console panel, power distributionunits, UPS 3000, all associated cabling.31

Weather & Solar Energy Related Industry Issues . Predicting wind energy ramp events using a rapid cycle, high-resolution weather model and Doppler radar data. . generation capacity of each turbine using power curve data . Optimizing Prediction by Blending Technologies