05 2017 05-04-clear sky models g-kimball

05 2017 05-04-clear sky models g-kimball

• 1.
  Confidential | ©2016 SunPower Corporation Clear sky irradiance and temperature models for mitigating sensor drift in PV system degradation analysis Gregory M. Kimball[1], Dirk C. Jordan[2], Chris Deline [2] [1] Sunpower Corporation, 77 Rio Robles, San Jose, USA [2] National Renewable Energy Laboratory, 15013 Denver West Parkway, Golden, CO 80401, USA PV Performance Modeling and Monitoring Workshop 8th PVPMC, 2017-05-09
• 2.
  2Confidential | ©2016 SunPower Corporation | Agenda 1. Introducing clear sky models for PV degradation analysis 2. Examples of clear sky normalization 3. Static vs dynamic clear sky models
• 3.
  3Confidential | ©2016 SunPower Corporation | Degradation assessment, method PRSTC PRSTC metric uses irradiance and temperature sensors to normalize power data • Normalize: – Use sensor irradiance and temperature to model expected performance 𝑃𝑅 𝑆𝑇𝐶 = 𝑃𝐴𝐶 𝑘𝑊 𝑃𝑆𝑇𝐶,𝑟𝑎𝑡𝑒𝑑 ∗ 𝐼𝑟𝑟𝑎𝑑𝑖𝑎𝑛𝑐𝑒 𝑃𝑂𝐴 𝑊 𝑚2 1000 𝑊 𝑚2 ∗ (1 + γ 𝑡𝑒𝑚𝑝𝑐𝑜 ∗ 𝑇𝑐𝑒𝑙𝑙 − 25 °𝐶 ) PRSTC PRSTC Weekly medians 15-minute data, filtered
• 4.
  4Confidential | ©2016 SunPower Corporation | Degradation assessment, method PRCS PRCS instead uses clear sky models to normalize power data 𝑃𝑅 𝐶𝑆 = 𝑃𝐴𝐶 𝑘𝑊 𝑃𝑆𝑇𝐶,𝑟𝑎𝑡𝑒𝑑 ∗ 𝐶𝑙𝑒𝑎𝑟 𝑆𝑘𝑦 𝐼𝑟𝑟𝑎𝑑𝑖𝑎𝑛𝑐𝑒 𝑃𝑂𝐴 𝑊 𝑚2 1000 𝑊 𝑚2 ∗ (1 + γ 𝑡𝑒𝑚𝑝𝑐𝑜 ∗ 𝑇𝑐𝑙𝑒𝑎𝑟 𝑠𝑘𝑦 𝑐𝑒𝑙𝑙 − 25 °𝐶 ) • Normalize: – Use clear sky irradiance and temperature to model expected performance PRCS PRCS Weekly medians 15-minute data, filtered (Sensor data still used for clear sky filtering) 1. Dirk C. Jordan, Chris Deline, Sarah R. Kurtz, Gregory M. Kimball, Mike Anderson, “Robust PV Degradation Methodology and Application”, PVSC, 2017.
• 5.
  5Confidential | ©2016 SunPower Corporation | Clear sky irradiance model 𝑃𝑅_𝐶𝑆 = 𝑃𝐴𝐶 𝑘𝑊 𝑃𝑆𝑇𝐶,𝑟𝑎𝑡𝑒𝑑 ∗ 𝐶𝑙𝑒𝑎𝑟 𝑆𝑘𝑦 𝐼𝑟𝑟𝑎𝑑𝑖𝑎𝑛𝑐𝑒 𝑃𝑂𝐴 𝑊 𝑚2 1000 𝑊 𝑚2 ∗ (1 + γ 𝑡𝑒𝑚𝑝𝑐𝑜 ∗ 𝑇𝑐𝑙𝑒𝑎𝑟 𝑠𝑘𝑦 𝑐𝑒𝑙𝑙 − 25 °𝐶 ) Modeled and measured irradiance • Clear sky irradiance models report the expected solar resource under clear conditions • Transposition of the data converts to plane-of-array (POA) irradiance • PVLIB provides an open-source clear sky model 1. W. F. Holmgren, R. W. Andrews, A. Lorenzo, J. S. Stein. “PVLIB Python 2015”. 42nd IEEE Photovoltaics Specialists Conference, 2015. 2. J. S. Stein, W. F. Holmgren, J. Forbes, C. W. Hansen. “PVLIB: Open Source Photovoltaic Performance Modeling Functions for Matlab and Python”. 43rd IEEE Photovoltaics Specialists Conference, 2016.
• 6.
  6Confidential | ©2016 SunPower Corporation | Clear sky temperature model • We introduce clear sky temperature models to report the expected solar cell temperature under clear conditions • NEO provides average ambient day and night temperature based on climate models • Cell temperature is a function of ambient temperature and irradiance 𝑃𝑅_𝐶𝑆 = 𝑃𝐴𝐶 𝑘𝑊 𝑃𝑆𝑇𝐶,𝑟𝑎𝑡𝑒𝑑 ∗ 𝐶𝑙𝑒𝑎𝑟 𝑆𝑘𝑦 𝐼𝑟𝑟𝑎𝑑𝑖𝑎𝑛𝑐𝑒 𝑃𝑂𝐴 𝑊 𝑚2 1000 𝑊 𝑚2 ∗ (1 + γ 𝑡𝑒𝑚𝑝𝑐𝑜 ∗ 𝑇𝑐𝑙𝑒𝑎𝑟 𝑠𝑘𝑦 𝑐𝑒𝑙𝑙 − 25 °𝐶 ) Source data from Nasa Earth Observatory, derived from MODIS Available at: https://github.com/kwhanalytics/rdtools/tree/clearsky_temperature
• 7.
  7Confidential | ©2016 SunPower Corporation | Agenda 1. Introducing clear sky models for PV degradation analysis 2. Examples of clear sky normalization 3. Static vs dynamic clear sky models
• 8.
  8Confidential | ©2016 SunPower Corporation | Sensor drift and shift, example 1 However, sensor drift compromises PRSTC! PRSTC PRSTC shows drift to more positive values Clear sky index shows decreasing values. Temperature/Expected remains steady. 𝐶𝑙𝑒𝑎𝑟 𝑠𝑘𝑦 𝑖𝑛𝑑𝑒𝑥 = 𝐼𝑟𝑟 𝑃𝑂𝐴 𝑊 𝑚2 𝐶𝑙𝑒𝑎𝑟 𝑆𝑘𝑦 𝐼𝑟𝑟 𝑃𝑂𝐴 𝑊 𝑚2 𝑡𝑒𝑚𝑝 𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑 = 𝑇𝑐𝑒𝑙𝑙 (𝐾) 𝑇𝑐𝑙𝑒𝑎𝑟 𝑠𝑘𝑦 𝑐𝑒𝑙𝑙 (𝐾) 1. Mike Anderson, Zoe Defreitas. "A SunPower Fleet-Wide System Degradation Study using Year-over-Year Performance Index Analysis", SunPower white paper, 2012. 2. Mike Anderson, Zoe Defreitas, et al., “A System Degradation Study of 445 Systems using Year-over-Year Performance Index Analysis”, PVMRW, 2013.
• 9.
  9Confidential | ©2016 SunPower Corporation | Sensor drift and shift, example 1 PRCS trades precision for accuracy. PRCS PRSTC PRSTC shows median degradation rate of +0.6 %/yr PRCS shows median degradation rate of -0.9 %/yr
• 10.
  10Confidential | ©2016 SunPower Corporation | Sensor drift and shift, example 2 And sensor shifts are just as bad. PRSTC PRSTC shows shifts in 2013, 2015, with PR>1 in 2016 Clear sky index shows decreasing values in 2015- 2016. Temperature/Expected shifts down in 2013- 2015. 𝐶𝑙𝑒𝑎𝑟 𝑠𝑘𝑦 𝑖𝑛𝑑𝑒𝑥 = 𝐼𝑟𝑟 𝑃𝑂𝐴 𝑊 𝑚2 𝐶𝑙𝑒𝑎𝑟 𝑆𝑘𝑦 𝐼𝑟𝑟 𝑃𝑂𝐴 𝑊 𝑚2 𝑡𝑒𝑚𝑝 𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑 = 𝑇𝑐𝑒𝑙𝑙 (𝐾) 𝑇𝑐𝑙𝑒𝑎𝑟 𝑠𝑘𝑦 𝑐𝑒𝑙𝑙 (𝐾)
• 11.
  11Confidential | ©2016 SunPower Corporation | Sensor drift and shift, example 2 PRCS trades precision for stability.PRCS PRSTC PRSTC shows shifts in 2013, 2015, with PR>1 in 2016 PRCS shows consistent behavior throughout its history
• 12.
  12Confidential | ©2016 SunPower Corporation | Agenda 1. Introducing clear sky models for PV degradation analysis 2. Examples of clear sky normalization 3. Static vs dynamic clear sky models
• 13.
  13Confidential | ©2016 SunPower Corporation | Clear sky irradiance models Model name Data source Model type Spatial resolution (°) Temporal resolution Size Linke SoDa, Ineichen static 0.16 Monthly 20 MB NEO NEO/Modis, Solis dynamic 0.05 Monthly 80 MB/yr Ecmwf ECMWF, Solis dynamic 1 3 hr 130 MB/yr Ecmwf month ECMWF, Solic dynamic 1 Monthly 3 MB/yr 2012 2013 2014 2015 2016 2012 2013 2014 2015 2016 IrradianceGHI(W/m2) MSE(clearskyindexvs1.0) 2012 2013 2014 2015 2016 2012 2013 2014 2015 2016 IrradianceGHI(W/m2) MSE(clearskyindexvs1.0)
• 14.
  14Confidential | ©2016 SunPower Corporation | Irradiance model performance Static models and dynamic models show similar errors versus sensor data MSE (sensor vs model) irradiance ~480 sites evaluated 90% of sites in the US
• 15.
  15Confidential | ©2016 SunPower Corporation | Clear sky temperature models Model name Data source Model type Spatial resolution (°) Temporal resolution Size png NEO/Modis static 0.05 Monthly day/night 22 MB neo hdf NEO/Modis static 0.5 Monthly day/night 6 MB Ecmwf skt ECMWF dynamic 1 3 hr 90 MB/yr Neo dynamic NEO/Modis dynamic 0.05, 1 Monthly day/night 22 MB + 0.2 MB/yr Ecmwf hdf5 ECMWF static 1 Monthly day/night 3 MB 2012 2013 2014 2015 2016 2012 2013 2014 2015 2016 2012 2013 2014 2015 2016 2012 2013 2014 2015 2016 Ambienttemp(degC) Ambienttemp(degC) MSE(sensorvsmodel) MSE(sensorvsmodel)
• 16.
  16Confidential | ©2016 SunPower Corporation | Temperature model performance temperature MSE (sensor vs model) Static models and dynamic models show similar errors versus sensor data ~480 sites evaluated 90% of sites in the US
• 17.
  17Confidential | ©2016 SunPower Corporation | Conclusion • High quality models are available for clear sky irradiance (thank you PVLIB). We introduce simple models of clear sky ambient temperature (thanks to Nasa, ECMWF). • Using clear sky models, PV system performance can be analyzed without the effects of sensor drift and degradation. • For degradation analysis, simple static models appear to perform as well as more complex dynamic models. The PRCS metric prevents poor sensors from looking like AMAZING performance.
• 18.
  Confidential | ©2016 SunPower Corporation Thank you!
• 19.
  19Confidential | ©2016 SunPower Corporation | Dynamic clear sky model components • The dynamic clear sky model uses monthly average data from satellite sources to account for the effect of aerosols, water vapor, and temperature anomaly. • The model generates values for clear sky irradiance (W/m2) and clear sky temperature (°C) that vary from year-to-year.
• 20.
  20Confidential | ©2016 SunPower Corporation | Sensors with nominal behavior PRCS has lower precision but…… PRCS PRSTC PRSTC shows median degradation rate of -1.3 %/yr with σYoY of 3.3 %/yr PRCS shows median degradation rate of -1.3 %/yr with σYoY of 5.1 %/yr