1. Lidar Measurement Accuracy under Complex Wind Flow in Use for Wind Farm Projects
Matthieu Boquet, Mehdi Machta, Jean-Marc Thevenoud

2. Agenda Remote sensors in the framework of a wind farm project
Assessment of accuracy and uncertainty: statistical approach
Application on WINDCUBE™ v2 lidar data sets

3. Assessing the wind resource for a wind farm project
The future production of a wind farm depends on:
 the wind across the site and over project life
 the performance of the turbine array
Determining the future production (mean value and its uncertainty) drives the financing of the project
 The better the analysis (i.e. the lower the uncertainty) the more attractive the financial terms
Use measurements, know-how, theories, models, experience…

4. The need for measurements
Vertical data at hub height
Horizontal data cross the site ?

5. Application of remote sensor
Remote sensors measure up to and above hub height
Remote sensors are portable and can therefore be easily moved to measure at several locations
They allow a more secure assessment of the wind resource...
…But only if the remote sensor is providing accurate wind data with a high confidence level!
Before applying one technology rather than another, the instrument’s accuracy and uncertainty need to be well understood and assessed.

6. Instrument comparison with a calibrated cup
WINDCUBE® v2 lidar
Reference calibrated cup anemometer
Y = X (m/s)
R2 = 0.998
Mean deviation = 0.5%
Standard deviation of deviation = 2.1%
Will you get same bias (accuracy) and scatter (uncertainty) on your site of application?
Yes, if the technology is robust
Yes, if you were lucky and have the exact same measurement conditions on the site of application than during this test
No, otherwise…

7. Accuracy needs to be thoroughly assessed
By technological knowledge:
Technology used: radio, acoustic, light…
Device configuration: emission, reception…
Device conception: material, electronic…
Signal processing: algorithms…
By empirical analysis:
Considered as a black box
Accuracy assessed under a wide (exhaustive?) variety of conditions
Approach of IEC * standard for wind turbine power curve measurements
* Currently under revision

8. Statistical approach
Knowing the device technology, all potentially influential parameters need to be listed (according to knowledge, literature, experience, etc.):
Wind shear, wind veer, turbulence intensity…
Atmospheric stability, rain…
Aerosol concentration, ambient noise…
These parameters need to be measured during the accuracy/uncertainty assessment test
The relationship between wind speed deviations (between the remote sensor and the calibrated cup) and the parameters are assessed

9. A classifier is used to assess these relationships
error distribution during sensitivity test
Test site conditions during test +
Real error (m.s-1)
Classifier is found (decision tree, etc.)
Estimated error (m.s-1)
General site conditions
Classifier

10. Statistical approach
One classifier is built per data set using a statistical methodology
One data set of variables is built, following the probability distributions defined for every variable and thus can be considered representative for most sites of application
The method of building the classifiers is valid if they give same RSD error distribution for the same general data set
This error distribution allows for estimation of the RSD mean bias and uncertainty

11. Remarks on the statistical approach
The classifier needs to take into account dependency between variables
It determines the most influential parameters on the deviations in the data set
Be careful with variables known to be influential which are not present during the test!
Parameter variations on site of application could be also addressed as input to the classifier

12. Application to WINDCUBE® v2 lidar
Here classifier is a decision tree bag
It estimates system error based on atmospheric and instrumental parameters
The classifier predicts the instrument error!

13. Error distribution for several data sets (taken under various conditions)
Estimated error (m.s-1)
Classifiers are found for various data sets and applied to reference site:
results of estimated accuracy are coherent among the data sets

14. Influential parameters and final accuracy
The most influential parameters are found to be wind shear and flow inclination (expected from theory)
Precipitation, turbulence intensity, aerosol concentration are found to have an inconsequential influence on the accuracy
Accuracy is defined as the mean error and is here found to be ~1%
Uncertainty is defined as the standard deviation of the error and is here found to be lower than 2%
Results are coherent with validations against met masts, as seen in slide 6:
Y = X (m/s)
R2 = 0.998
Mean deviation = 0.5%
Standard deviation of deviation = 2.1%

15. Conclusion
The accuracy of a remote sensor technology needs to be assessed according to environmental parameters in order to ensure that the measurement will not differ from a validation test site to the application site
WINDCUBE® Lidar has shown high robustness toward surrounding environmental variables and its accuracy has been assessed in a large variety of sites and conditions
The instrument’s measured wind speed values (and the uncertainty associated) can be used for assessment of the wind resource on wind farm sites, resulting in a high lidar RoI