Long term yield predictions (LTYP) are a prerequisite for business decisions on long term invest-ments into photovoltaic (PV) power plants. The preparation of a LTYP report typically relies on numerical modelling and prediction of the expected electrical yield, based on experience with previous PV power plants, laboratory measurements and more or less the whole knowledge gained in the PV community over the past years and decades. However, though PV system model-ling has been performed for decades, not much effort has been spent on a comprehensive inves-tigation of the uncertainties related to this task. This report tries to collect some insights into the field of uncertainties of several technical aspects of PV system yield prediction and assessment.
The first main section lists typical measurements, dealing either with a PV system component’s properties or with PV system performance. It covers the uncertainties related to the most im-portant measurands in PV solar energy:
Uncertainty in irradiance measurements is in part related to the instruments, and in part to the measurement practices. While existing handbooks and guidelines may help to reduce operational issues, the uncertainties related to the instruments themselves are more difficult to minimize or reduce. However, if a certain non-ideal behaviour of an instrument is systematic and known, then a systematic correction can be applied to reduce or remove its influence on the measurement. In some cases manufacturers already supply information about temperature dependency and/or non-linearity. In the current PVSENSOR project, a wide range of instruments were characterized, and many systematic errors were identified and quantified. With such knowledge of instrument operating conditions it will be possible to quantify each systematic source of measurement error.
STC power measurement of PV modules and the estimation of its uncertainties is a topic that gained attention in recent years and saw remarkable improvements. For a full uncertainty as-sessment, it is important that stability issues are considered in addition to pure measurement uncertainty. There are laboratories with a profound knowledge on their uncertainty budget, typi-cally those with the smallest overall uncertainties (down to ±1.6%), laboratories where the refer-ence cell or reference module calibration dominates the uncertainty budget, and laboratories where apparently uncertainties were not analysed in detail.
System testing looks at the performance of the complete conversion chain of a PV system. The determination of the observed performance ratio PR is rather easy, including an assessment of its uncertainty. The determination of the expected PR (as a quality requirement) is the major issue, as the PR depends on the system design and changes with the system’s operating conditions. Despite this potential weakness, a PR test can form a valuable tool during the commissioning of a PV system.
The second major section of this report investigates several of the modelling steps for gains and losses in a PV system, again along a similar list:
Irradiation data derived from satellite images are increasingly used as input for long-term yield estimations and as reference yield for monitoring and business reporting. Several authors have evaluated the quality of satellite-based irradiance data in the past, typical normalized root mean square errors for satellite-based irradiation reported in literature are situated between 4% to 8% for monthly and 2% to 6% for annual irradiation values.
Solar irradiation at the Earth’s surface is not stable over time for all locations on earth but may undergo significant long-term variations for particular regions, which is referred to as “global dimming and brightening”. Consequently, also related uncertainties may not be considered to be negligible. In the presence of long-term trends, the question for solar resource assessments is no longer “what is the ‘true’ climatological value?”, but “what is the best predictor for the next 20 years?”. A suitable estimator should be a recent time period, that is long enough to filter the influence of single years with high anomalies, but which is short enough, to minimize the influ-ence of past trends. Using irradiance data for the 10 most recent years is proposed as a good compromise to fulfil these conditions.
The DC energy yield of a PV module depends on module characteristics as well as operating condi-tions. With respect to uncertainties, the different influencing effects (irradiance level, angle of incidence, operating temperature, etc.) are typically represented by one individual factor per ef-fect. The influences are assumed to be independent. Furthermore, these factors are often used in integrated form, e.g. over one year. The calculation of the influencing factors and the uncertainty estimation in detail is described and discussed in the main text.