Calculation of CSP Yields with Probabilistic Meteorological Data Sets: A Case Study in Brazil

Abstract

Abstract Current practice for yield prognosis of solar thermal power plants simulates the average annual performance by using a typical meteorological year data set (TMY). This represents the long-term average or a 50 % probability of exceedance(P50) ofdirect normal irradiance (DNI) at the project's location. For more conservative risk evaluation it is common practice to calculate the 90 % yield exceedance level (P90) by estimating the uncertainty of the long-term DNI which depends on data set uncertainty and inter-annual variability.A simple approach to calculate the P90 yieldis to assume a normal distribution for this uncertaintyand a direct 1:1 relation of DNI averages to the yields. However, since the relation of DNI to energy yield is actually not linear it becomes more and more popular to calculate it from annual meteorological data sets (MY90), which are representing P90 DNI averages at a realistic distribution of actual values in the same time resolution as the P50 TMY. Applying such MY90 data sets still has the shortcoming that they are synthetic, whilereal years of data should lead to more realistic yields. Thus, this paper proposes the use of multiple years of weather inputto realistically include the annual variability of DNI. To also represent the effect of the data set uncertainty,thetime-series are modifiedin such away that the annual DNI values follow a normal distribution with a 1-sigma width equivalent to the diagnosed data set uncertainty. The impact of this probabilistic approach on the energy yield of a CSP project is shown for thesite Bom Jesus da Lapa in Brazil.Since the estimation of a realistic uncertainty of the long-term DNI at this location was challenging, several uncertainties between 3-9 % were assumed that could possibly be inherent in such a data set. Using theseassumed data set uncertainties and theinter-annual variability of the data set, the deviations to the long-term meanof energy yield are shown for the current practice approaches and the new method. Hence for a data set uncertainty of 5 %the very basic risk analysis results in a single-year P90 yield11.1 % below P50,while using a MY90 single year data set is resulting in a P90 yield 9.7 % below P50. The probabilistic approach introduced here is leading to P90 yields 8.5 % below P50.