Asymmetric uncertainties in offshore wind turbine availability: comparison with real operational wind farm data
Marie-Anne Cowan, Gemma Daron
DNV GL, Bristol, UK
A critical aspect of the energy production assessment of an offshore wind farms is the definition of the technical energy losses and their associated uncertainties. Defining the uncertainty in the energy yield estimate is crucial to understanding the level of risk in a project and feeds into the economic analysis of a prospective project.
System availability is often the largest energy loss of an offshore wind farm, behind wake losses, and one of the largest sources of uncertainty in an energy production assessment. By their very nature, offshore wind farm technical losses are extremely project specific and depend upon the project characteristics and system design. A number of risks associated with turbine availability, such as turbine reliability, are best accounted for as uncertainties. Turbine availability has a higher downside risk compared to potential upside gains, lending itself to a non-normal distribution truncated to 100%.
This work investigates how the turbine availability uncertainty should be treated as an asymmetric uncertainty and compares the modelled assumption with real operational wind farm data. The outcome of this work is a more realistic prediction of the expected energy production of the wind farm and better understanding of actual turbine availability losses that are currently being achieved.
An asymmetric distribution has been used to model wind farm availability. In order to compare with real data, DNV GL has gathered wind farm availability information from a range of offshore wind projects from different geographical locations and operators. These data were reviewed to analyse the actual industry behaviour around the European offshore wind industry and used to derive a distribution. The resulting distribution was compared with the modelled distribution in order to validate the assumption of asymmetry to account for the upside and downside risks.
Furthermore, DNV GL has undertaken a high level review of actual production data from operational European offshore wind farms compared to pre-construction energy yield estimates in order to draw conclusions regarding typical distributions observed in real world data.
The review of turbine availability data revealed an asymmetric distribution with years with low availability observed. A reasonable fit was observed between the actual turbine availability data and the modelled distribution.
The review of the operational data from European offshore wind farms resulted in a total of 100 individual wind farm years from 25 European offshore wind farms with operational periods varying from 1 year to 10 years. The resulting distribution of wind farm performance indicates a bigger ‘downside' risk than ‘upside' gain, further supporting the above finding. A Monte Carlo analysis was also undertaken considering an asymmetric distribution for turbine availability in conjunction with assumptions made for distributions of other technical losses to model the energy distribution of a "typical" offshore project for comparison.
The results of the study confirm that modelling turbine availability as an asymmetric uncertainty more accurately reflects the observed availability distribution from operational European wind farms and that the modelled distribution assumed provides a reasonable fit to these data.
Furthermore, when reviewing the overall wind farm performance, the study reveals that the distribution of actual wind farm performance data is more accurately described by an asymmetric distribution, indicating a bigger ‘downside' than ‘upside'. The observed distribution of actual performance data and the modelled energy distribution derived for a "typical" offshore wind farm as part of a Monte Carlo losses and uncertainty model also indicated a reasonable fit, further supporting the above finding.
This presentation will provide delegates with an insight into the availability distributions that are currently being observed in real operational wind farm data and with evidence to support the assumption that the uncertainty in this technical loss is best described by assuming a non-normal distribution to capture the asymmetry in the downside risks and upside gains.
Given the resolution of the data available and the high level nature of the work, there are some limitations noted with this work, which highlights the need for more data sharing within the industry - particularly on major events such as cable failures. Therefore delegates will be encouraged to think about how they could contribute to greater data sharing.