Photos placed in horizontal position with even amount of white space between photos and header Simulating Solar Power Plant Variability for Grid Studies: A Wavelet-based Variability Model Matthew Lave 1,2, Joshua Stein 1, Jan Kleissl 2, Abraham Ellis 1, Clifford Hansen 1, Yusuke Miyamoto 3 mlave@sandia.gov UWIG Solar User Group Fall 2011, Maui, HI 1 Sandia National Laboratories; 2 University of California, San Diego, 3 Kandenko, Ibaraki, Japan Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy s National Nuclear Security Administration under contract DE-AC04-94AL85000. 1
Wouldn t it be nice to be able to determine how much of a reduction in variability will occur in transitioning from a GHI point sensor to an entire power plant for any plant? In order to address how to integrate PV into the grid, we need to have an understanding of the variability. How does plant size (footprint and capacity) affect the reduction in variability? What is the difference between central and distributed plants? How does this relationship vary geographically (coastal vs. inland, by latitude, etc.)? To answer these questions, a solar power variability model is needed. 2
Wavelet Variability Model (WVM) The WVM presented here is a method to estimate aggregated PV plant output variability given only a single point sensor measurement Universal: works for plants at any location, with any arrangement of PV modules Uses a wavelet decomposition by timescale to account for different variability reductions (VRs) at different timescales Can adjust PV density to simulate a distributed plant, a central plant, or combinations of both 3
Two test cases: Distributed PV Plant Ota City, Japan 2MW rooftop PV ~550 houses with PV Varying tilt/azimuth Central PV Plant Copper Mountain, NV 48MW thin film PV Fixed tilt 4
Wavelet Decomposition Start with normalized GHI from a point sensor (blue) and normalized power output of all of Ota City (green). Normalized means 1 = clear. Normalized GHI and power output look similar, but if we zoomed in further, we would notice the power output is smoother than the GHI. To examine this, we break these down into fluctuations at each timescale, from 2s to 1.1h.
Normalized GHI/Power 1.1h 34m 17m 8.5m 4m 2m 1m 32s 16s 8s 4s 2s Wavelet Decomposition = Strong reduction in fpi at short timescales: lots of geographic smoothing Little reduction in fpi at long timescales: little geographic smoothing 6
Determining the Variability Reduction 7
Ota City, Japan Lanai, HI UC San Diego, CA Copper Mountain, NV (limited data) Alamosa, CO (limited data) 8
Wavelet Variability Model (WVM) Model Inputs PV Footprint Point Sensor Timeseries Model Outputs Variability Reduction at Each Timescale PV Plant Capacity or Density Plant Areal Average Irradiance* *can convert to power using the Sandia Photovoltaic Array Performance Model, or a simple linear multiplier if not all inputs to the Performance Model are known. 9
Input: PV Footprint Input area of interest by drawing one or many polygons on a Google Map Ota City Copper Mountain scale comparison 10
Input: Timeseries and PV Density Timeseries Input point sensor timeseries, latitude, longitude, and UTC offset from *.mat file Timeseries resolution determines simulation resolution: 1-sec data in -> 1-sec data out 2MW Central Plant PV Density If plant size (MW) is known, program will calculate PV density Otherwise, specify central or distributed 2MW Distributed Plant 11
Validation of WVM Highly variable days were chosen to test the WVM 12
Compare Fluctuation Power Index Ota City Copper Mountain Goal of WVM is to reproduce power content of fluctuations. WVM simulation (P norm,sim ) matches the actual power (P norm ) well, and is a strong improvement over the GHI point sensor (GHI norm ). Simulation does not match actual as well at Copper Mountain (48MW) as at Ota City (2MW) 13
Compare Ramp Rates Simulated RRs (blue) compare well to actual power RRs (red) at all timescales. Copper Mountain Ota City The WVM simulation is much better at matching RRs than GHI at short timescales. For Ota City: through 30s. For Copper Mountain: through 60s. 14
Next Steps 15
Acknowledgements Yusuke Miyamoto and Eichi Nakashima from Kandeko for providing and supporting Ota City data. David Jeon, Leslie Padilla, and Shiva Bahuman from Sempra Energy for providing and supporting Copper Mountain data. 16
17