The importance of doing it right: long-term correction

Introduction

Since CREYAP 1 it has been known that LT (long-term) correction is one of the big contributors to the uncertainty of projects. In the past LT correction has been hampered by poor quality of LT data, which made it difficult to optimise the MCP (measure-correlate-predict) processes. However, with the improved quality and accessibility of meso-scale data the door has been opened for more sophisticated approaches to perform LT correction. With higher quality data in hand it becomes possible to analyse the performance of the commonly used MCP processes and optimise their use, which is necessary because there can be significant differences in the resulting LT corrected wind speeds.
The irregular weather behaviours we experience globally adds extra relevance as we see very unusual “wind years” which are very sensitive to the optimal choice of MCP process.


Approach

As a benchmark the exceptionally poor wind year of 2014 in Turkey has been used, a very unusual year both regarding wind speeds and wind directions. Through a structured approach using several meso-scale data sets and MCP processes we quantify their consequences on the calculated AEP (annual energy prediction) and site-suitability. We demonstrate how to reduce the range of uncertainty that can be made.

Main body of abstract

Each of the commonly used and industry-accepted MCP processes has its strengths and weaknesses which becomes very obvious when applying several methods on the same site. One model might indicate an upward correction of the on-site data whereas another might show less increase or maybe a downward correction, and for another site the results might be the other way around.

The question now is how to choose the best model for corrections given a certain site. Which parameters can you use to make the right choice of model. It is common understanding that the wind speed correlation is one of the key factors to decide which LT data set is most suitable. We have found that a whole range of parameters and considerations are necessary in order to choose the “right” dataset and the “right” method of applying the dataset.
A total of seven sites have been analysed in various terrain, all equipped with an IEC-compliant measurement mast of a measurement height of 80m which has been measuring for a minimum of one year. The anemometry consists of First Class, MEASNET calibrated cup anemometers. The recovery rate of the data is above 95%. The installation of two top anemometers allowed to remove any potential shadowing effect due to the tower and lightning finials.
Systematically the data from all sites have been LT corrected with 2 meso-scale products, in most cases also with MERRA data. The data sets are well correlated with the LT data. Wind Index, linear regression (with residuals) and Matrix MCP processes have been used. Despite the excellent correlation we note huge differences in the resulting energy correction factors, on average 13%, in extreme cases up to 25% depending on MCP process. The reason is believed to be the atypical wind rose for the year with on-site data, which some methods do not take into account.
The decision which MCP process to choose is not only affecting the AEP but can also lead to the wrong choice of IEC class wind turbine.


Conclusion

We show that MCP processes need to focus on more aspects than correlation of wind speed. Wind direction, diurnal and seasonal distributions of wind speed COMBINED with direction needs to be evaluated to choose the optimal method.


Learning objectives
Guidelines for a more qualified choice of data sets and methods are presented as a first step towards a best-practice guideline for the industry to reduce the uncertainty of LT corrections.