Improved prediction of site spectral impact Dr Benjamin Duck, Dr Chris Fell
16 June 2015
ENERGY FLAGSHIP
CSIRO PV Outdoor Research Facility
4th PVPMC Workshop – Cologne – 21st October 2015
The problem of spectrum
4th PVPMC Workshop – Cologne – 21st October 2015
• OBJECTIVE
– Method for determining the impact that changes in the spectral irradiance distribution has on PV for both validation (historical) and forecasting (predictive) at arbitrary locations
• METRICS
– Spectral mismatch factor
– Spectral impact factor
• PROBLEM
– Detailed spectral information is typically unavailable
– Existing models have difficulty making accurate short timescale predictions
• QUESTION
– Can we improve on existing methods for predicting spectral impact based on commonly available data
Impact of changes to spectrum
4th PVPMC Workshop – Cologne – 21st October 2015
c-Si measured daily SIF
c-Si measured mismatch
• Impact depends on timescale
– Instantaneous > 25%
– Affected by instantaneous cloud cover
– Daily > 10%
– Days with constant clouds are rare
– High air mass compensation
– Yearly = 1%
– Averaging reduces impact significantly
• Directly measured mismatch (spectroradiometer)
• In-directly measured mismatch (short circuit current)
Measuring spectral mismatch
4th PVPMC Workshop – Cologne – 21st October 2015
• Directly measured results agree reasonably with indirect results from module Isc values.
• Variation in data is well matched
• Seasonal changes are duplicated
• Small differences may be due to • Low light level performance
• Unacounted for seasonal changes
Measuring spectral mismatch II
4th PVPMC Workshop – Cologne – 21st October 2015
CIGS
c-Si
CdTe
• Without direct measurements spectral mismatch must be estimated
• Commonly used methods are:
– Sandia array performance model (SAM)
– CREST ‘utilization factor’ (PVsyst)
• Underlying assumption: Spectrum at air mass 1.5 = AM1.5
Linking spectral mismatch to air mass
4th PVPMC Workshop – Cologne – 21st October 2015
King, Boyson and Kratochvil, Photovoltaic Array Performance Model, Sandia
National Laboratories report SAND2004-3535, Albuquerque, NM (2004)
Betts, Gottschalg and Infield, Spectral irradiance correction for PV system yield
calculations,19 th European Photovoltaic Solar Energy Conference, Paris (2004)
Comparing measurements to predictions
4th PVPMC Workshop – Cologne – 21st October 2015
CdTe
CIGS
• Season independent
offset
• Daily SIF variation is not
matched
• Small average season
independent offset
• Large daily SIF variation
not matched
•Out of season behaviour
not explained
Offset in measured Pmp data
• Data corrected to 1000 W/m2 and 25 °C shows an offset
PVSC 42 – 16th June 2015
Spectrum at air mass 1.5 ≠ AM1.5 reference
• Data at 1000 W/m2, 25 °C and air mass = 1.5 shows an offset
4th PVPMC Workshop – Cologne – 21st October 2015
Clear skies vs Cloudy skies
• Data when skies are not clear does not follow simple air mass model
• CREST model attempts to capture this using clearness index
• Original form uses bandgap based windowing of spectrum data.
• Apply results from modified model using true spectral response.
PVSC 42 – 16th June 2015
Spectral mismatch estimation – CREST
• CREST uses a functional relationship between Air Mass, a clearness index (kt
*) and the useful fraction of incident irradiance.
• Useful fraction is calculated using a spectral windowing technique
• Improvement is found by calculating spectral mismatch using the true spectral response (WUF) rather than the useful fraction.
• Adds ability to account for cloud cover
• Coefficients are found by fitting a surface to the spectral mismatch data.
4th PVPMC Workshop – Cologne – 21st October 2015
𝑓 𝐴𝑀, 𝑘𝑡∗ =𝑊𝑈𝐹𝑚𝑒𝑎𝑠𝑊𝑈𝐹𝑟𝑒𝑓= 𝑨 ∶ 𝑷 𝑘𝑡
∗ 𝑸 𝐴𝑀
𝑸 𝐴𝑀 = 𝐴𝑀𝑚𝑚
0
𝑷 𝑘𝑡∗ = 𝑘𝑡
∗𝑛𝑛
0
• Captures variations due to cloudy skies as well as seasonal changes.
Modified CREST model predictions
4th PVPMC Workshop – Cologne – 21st October 2015
c-Si measured
c-Si modelled (CREST)
Modified CREST model predictions
4th PVPMC Workshop – Cologne – 21st October 2015
CdTe
CIGS
• Site spectral offset is accounted for.
• Variation in data due to cloudy conditions is replicated.
• Site dependence is implicit due to fits to specific site data.
• Possible to use the site spectral offset as a scaling factor?
Comparison of impact
4th PVPMC Workshop – Cologne – 21st October 2015
Module Type
Model MBE daily SIF MAE daily SIF
RMSE daily SIF
Standard Deviation
CdTe SANDIA 0.030 0.031 0.044 0.032
CREST -0.010 0.014 0.019 0.016
CREST-WUF 0.001 0.009 0.012 0.012
c-Si SANDIA -0.003 0.021 0.027 0.026
CREST -0.014 0.016 0.021 0.016
CREST-WUF -0.002 0.007 0.011 0.010
CIGS SANDIA -0.017 0.024 0.026 0.020
CREST -0.015 0.017 0.021 0.015
CREST-WUF -0.001 0.007 0.010 0.010
Comparison of impact
• Hourly absolute resource estimate error statistics
• Adjusted Sandia has had site spectral offset applied.
• Improvement of using the modified CREST is clear.
4th PVPMC Workshop – Cologne – 21st October 2015
Unmodified CREST Adjusted Sandia Modified CREST
• Surface shape is different from direct measurements
• Cloudy data not as consistently captured. Diffuse contribution?
Modified CREST using Isc data
4th PVPMC Workshop – Cologne – 21st October 2015
CIGS measured
CIGS modelled
Modified CREST using Isc data
4th PVPMC Workshop – Cologne – 21st October 2015
CdTe
CIGS
• Equivalent results to direct spectral measurement observed.
• Out of season differences are still present.
• Results are stable with subsampling of dataset and within module type.
Clear sky model dependence
4th PVPMC Workshop – Cologne – 21st October 2015
• Surface result is dependent upon modelling of kt*
• Requires consistent GHI model to be adopted
Frequency of air mass and kt* data
4th PVPMC Workshop – Cologne – 21st October 2015
• Most data is below air mass = 3.0
• Highest concentration of data for clear sky days
• Lots of cloudy days but surface data is scattered
• kt* is not continuous
• kt* > 1 suggests limitations of GHI
model
Other variations
4th PVPMC Workshop – Cologne – 21st October 2015
• Out of season variation is not captured
• Changes to air mass dependence for clear skies
c-Si measured mismatch for a clear sky day in each season
• Propose combining the modified CREST surface and site spectral offset to lead to superior predictions
• The relationship between site location and climate and the site spectral offset is unclear
• Added measurements are also needed to determine clearness surface dependence on site and offset properties
• Standardisation of measurement and fitting methods.
Is general characterisation possible
4th PVPMC Workshop – Cologne – 21st October 2015
Acknowledgements
• This work was conducted with support from • The CSIRO Energy Flagship program
• The Australian Renewable Energy Agency (ARENA)
4th PVPMC Workshop – Cologne – 21st October 2015