P50/P90 Analysis of a Solar Photovoltaic Plant in METU NCC Using the Empirical Method

Gören D., Taylan O.

2020 2nd International Conference on Photovoltaic Science and Technologies (PVCon), Ankara, Turkey, 30 November - 02 December 2020, pp.1-8

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/pvcon51547.2020.9757770
  • City: Ankara
  • Country: Turkey
  • Page Numbers: pp.1-8
  • Middle East Technical University Affiliated: Yes


Large investments in renewable energy systems have some financial risks, and energy exceedance probabilities are required to find the expected range of energy yield values. In this study, P50/P90 analysis was done for a 1 MWp solar photovoltaic (PV) power plant in METU NCC using the System Advisor Model (SAM) software. Three different transposition models were used, which are the isotropic sky-diffuse, HDKR, and Perez models. Photovoltaic Geographical Information System-Surface Solar Radiation Dataset Heliosat (PVGISSARAH) database was used as the long-term satellite-based data, which includes the time period 2005-2016 with hourly time resolution, and three different typical meteorological year (TMY) datasets were used to compare the results with P50 values. Actual energy yield data of the years 2016 and 2017 were also compared. Interannual variability of global horizontal irradiation (GHI) was found as about 2.21%. According to the comparison of TMY and P50 values, it was found that average TMY underpredicted the annual GHI by 5.15%, and the energy yield values by 6.77%, 6.86%, and 6.83% for the isotropic, HDKR, and Perez models, respectively. Normal distribution assumption was also compared with the empirical method, and P50 value was underpredicted by 0.43% and 0.93% on average for GHI and energy yield, respectively. On the other hand, Normal CDF overpredicted P90 values than the empirical CDF by 2.61% and 3.93% for GHI and the energy yield, respectively. The overall uncertainty of energy yield was found as about 7.08%. The levelized cost of electricity (LCOE) also varied by 3.50% on average, and the non-exceedance probability values were obtained with respect to the energy yield.