Rare Events Early Detection in Wind Turbines
Loading...
Journal Title
Journal ISSN
Volume Title
Sähkötekniikan korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Author
Date
2022-12-12
Department
Major/Subject
Autonomous Systems
Mcode
ELEC3055
Degree programme
Master's Programme in ICT Innovation
Language
en
Pages
60+17
Series
Abstract
This thesis approaches the challenge of detecting rare events, specifically wind gusts, in the context of wind turbines operations. The goal by predicting wind speeds in the short horizon is to foresee gusts and enable enhanced control responses. While wind turbulence remains elusive, wind gusts are characterized as events where wind speed reaches at least 8 m/s and the variation between the peak and lull of the speed is beyond 4.5 m/s. Wind gusts impact turbines integrity by causing motor fatigue and tower/blade extreme loads. Such impacts can cause electric grid instability. Therefore, by being able to predict wind gusts occurrence and intensity, it is possible to reduce extreme loads, resulting in reduced component sizing, and naturally wind turbine costs while increasing product competitiveness. The pipeline developed here is the first combined pipeline for combined wind speed prediction and wind gusts detection. It consists of two consecutive stages: prediction and detection. In the prediction stage, a Long Short Term Memory network is used to predict wind speeds over a horizon range of 0.05 seconds to 0.5 seconds. In the detection stage, wavelet transformation is applied to the predicted wind speed to detect gust-like profiles by using generic wavelets (Morlet & DOG2) that can be adapted according to installation site characteristics. Each pipeline stage is benchmarked against multiple methods, representing the industry standard. The data set used is from a mast-measurement at the Hamburg Weather Mast in Germany, containing wind speed measurements sampled at 0.05 s for the duration of about 8.3 s. The method has shown that LSTM is the best performing prediction algorithm, when evaluated according to the Mean Squared Error and Mean Absolute Error metrics compared to other DL algorithms as well as ARIMA, the standard statistical technique for time series forecasting. Additionally, wavelet transformation has shown robust performance with minimal false positive detections compared to CUSUM, an industry standard for wind gusts detection. The two wavelets are combined with a 2-point derivative test that evaluates the wind acceleration to minimize error. The combination of the individual thresholds for each method result in a different control strategy recommendation.Description
Supervisor
Raisch, JorgThesis advisor
Ahmed Attia, SidKeywords
wind gust, wavelet, turbine, LSTM, forecasting, detection
