OATAO - Open Archive Toulouse Archive Ouverte Open Access Week

Sensitivity analysis of a filtering algorithm for wind lidar measurements

Rieutord, Thomas. Sensitivity analysis of a filtering algorithm for wind lidar measurements. PhD, Mathématiques Appliquées, Institut National Polytechnique de Toulouse, 2017

[img]
Preview
(Document in English)

PDF (Première partie : chapitres 1 et 2) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
10MB
[img]
Preview
(Document in English)

PDF (Deuxième partie : chapitres 2 et 3, bibliographie) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
6MB

Abstract

Wind energy industry and airport safety are in need of atmospheric observations. Remote sensors, such as lidars, are well proven and common technology to provide wind measurements in the first hundreds of meters of altitude. However, acquisition abilities of lidars are polluted by measurement noise. Using non-linear filtering techniques, we took part at the development of an algorithm improving wind and turbulence estimations. The process is based on a representation of the atmosphere with fluid particles. It uses a stochastic Lagrangian model of turbulence and a genetic selection filtering technique. Its efficiency depends of the setting of various parameters. Their values were fixed experimentally during the development phase. But their influence has never been assessed. This work addresses this question with a variance-based sensitivity analysis. New estimators of Sobol indices, using penalized regression have been tested. These estimators ensure the lowest Sobol indices automatically go to zero so the overall interpretation is simplified. The sensitivity analysis allows to reduce the system from 5 outputs and 9 inputs to 3 inputs (number of particles, real observation noise, observation noise given to the filter) and 2 outputs (wind spectrum slope, root-mean-squared error on wind). With this reduced system we determined a procedure to correctly set the most important parameters. The observation noise given to the filter is well set when the wind spectrum slope has the expected value of -5/3. Once it is set correctly, the error on wind is minimum and its expression is known.

Item Type:PhD Thesis
Uncontrolled Keywords:
Institution:Université de Toulouse > Institut National Polytechnique de Toulouse - INPT (FRANCE)
Laboratory name:
Research Director:
Gamboa, Fabrice and Dabas, Alain
Statistics:download
Deposited By: Thèse INPT
Deposited On:19 Jan 2018 11:38

Repository Staff Only: item control page