18
Copyright © 2016 Clean Power Research, L.L.C v052814 Quantifying Uncertainty with Satellite-to-Ground Tuning Adam Kankiewicz Sandia/EPRI 5 th PV Performance Modeling Workshop May 9 th 2016

1 1 kankiewicz_sandia_epri_pv_perf_wrk_shp_presentation_2016

Embed Size (px)

Citation preview

Copyright © 2016 Clean Power Research, L.L.Cv052814

Quantifying Uncertainty with Satellite-to-Ground Tuning

Adam KankiewiczSandia/EPRI 5th PV Performance Modeling Workshop

May 9th 2016

Copyright © 2016 Clean Power Research, L.L.C.

Presentation Outline

2

Motivation

Ground data

Satellite irradiance modeling

How and why we tune satellite data

Tuning uncertainty

Key takeaways

Copyright © 2016 Clean Power Research, L.L.C.

3

www.solaranywhere.com

Motivation: I Have X Amount of Ground Data. Can You Perform a (Viable) Tuning Study???

How does length or time period of ground data influence tuning study uncertainty?

GH

I (W

/m2 )

Ground-based Solar Resource Monitoring

Necessary to understand local variability effects

Ground truth for tuning process

Have to place into long term reference frame for proper resource context!

Image courtesy of GroundWork Renewables, Inc.

Long Term Resource Reference Frame

Satellite data provides the consistent, long term reference frame needed to derive reliable

estimates of P50, P90, variability, etc.

Satellite Data Modeling

Clear sky and cloudy sky errors need to be independently targeted in any solar satellite data tuning process!

Clear Sky Irradiance Radiative Transfer Model

+Cloudy Sky Irradiance

Cloud Modulation

Satellite Ground Tuning Methodology

Measure-correlate-predict (MCP) and Model Output Statistics (MOS) corrections often ignore individual satellite irradiance errors

Clear sky tuning

Cloudy sky tuning

Satellite DNI/DHI Rebalancing

Time of Day4 8 12 16 20

DNI

DHI

GHI OriginalRebalanced

GHI = COS(Z)*DNI + DHI

Not rebalancing data can improperly skew POAI calculations in energy simulations (PVsyst, SAM, etc.)

4 8 12 16 20Time of Day

GHI

Study Methodology Data inputs:

• Hourly averaged irradiance ground data (14 SURFRAD and ISIS sites)• Hourly averaged SolarAnywhere irradiance data

The satellite-to-ground tuning process is applied to fixed segments (1-24 months) of ground data which are rolling by a one month interval over 5 years

The tuning results are applied to 5 years of satellite data and residual error metrics are calculated

10

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

3 month example

Individual Site Results: Albuquerque, NM

11

Seasonal influences can be dramatic with less than a year of data Error results stabilize significantly with year + timeframes

Individual Site Results: Goodwin Creek, MS

12

Seasonal influences can be dramatic with less than a year of data Error results stabilize significantly with year + timeframes

Individual Site Results: Penn State, PA

13

Seasonal influences can be dramatic with less than a year of data Error results stabilize significantly with year + timeframes

Overall Results

14

Similar trends at all locations

Overall Results

15

Decreasing envelope of uncertainty with increased month selection independent of location

Key Takeaways

Seasonal impacts can be amplified with less than a year of ground data

We can provide uncertainty for tuning studies based on X amount of ground data

See further results presented at IEEE PVSC

16

The information herein is for informational purposes only and represents the current view of Clean Power Research, L.L.C. as of the date of this presentation. Because Clean Power Research must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Clean Power Research, and Clean Power Research cannot guarantee the accuracy of any information provided after the date of this presentation. CLEAN POWER RESEARCH, L.L.C. MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.

Thank you

Skip DiseSolarAnywhere Prod. Manager

[email protected]

Adam KankiewiczSolar Research Scientist [email protected]

Please feel free to contact us for any details or clarification related to presentation

Tom StaplesSenior Account Executive [email protected]

Impact of Non-Average Years on Tuning

18

Ground site Year Residual MBE Variance from long term annual average

Penn State

2010 0.67% 1.48%2011 0.55% -4.16%2012 -1.52% 1.46%2013 0.39% -2.29%

Goodwin Creek

2010 0.01% 4.40%2011 0.21% 1.15%2012 0.17% 1.90%2013 1.25% -4.33%

Albuquerque

2010 -0.59% 0.15%2011 0.97% -0.84%2012 1.11% -3.91%2013 0.93% -1.39%

CPR tuning is not affected by above or below average solar irradiance years