Aspects of Statistical Trends in High-Rate Wind Measurements
Authors
Oxvig, Christian Schou ; Pedersen, Patrick Steffen
Term
4. term
Publication year
2013
Submitted on
2013-06-06
Pages
104
Abstract
This thesis investigates low-frequency trends in high-rate wind speed measurements (1–10 Hz) and their implications for wind turbine load estimation. Because turbine design often relies on 10-minute statistics (mean and variability), we analyze how slow variations with periods longer than 1–2 minutes affect these statistics and the models used. Heterogeneous raw measurements are first harmonized into a consistent database. We distinguish between insignificant, linear, and periodic trends and develop an automatic classification method. A well-established wind speed simulation model is extended with a trend component. We show that an ideal low-pass filter can remove the trend, after which the residual can be described by the original model; in practice, detrending each wind speed component with an IIR filter is recommended. Applying the designed filter to measurements and comparing classification results before and after processing demonstrates that significant trends are effectively removed. The 10-minute mean wind speed remains unchanged, while 10-minute variability is reduced. The work is supported by the Wind Analysis Framework (WAF) Python package, which enables reproduction of the results.
Dette speciale undersøger lavfrekvente trends i højrate vindhastighedsmålinger (1–10 Hz) og deres betydning for estimering af laster på vindmøller. Da vindmølleprojektering ofte bygger på 10-minutters statistik (middel og spredning), analyseres hvordan langsomme variationer med perioder over 1–2 minutter påvirker disse statistikker og de anvendte modeller. De heterogene rå målinger harmoniseres først til en ensartet database. Vi skelner mellem ubetydelige, lineære og periodiske trends og udvikler en metode til automatisk klassifikation. En veletableret model til simulering af vindhastigheder udvides med en trendkomponent. Det vises, at et ideelt lavpasfilter kan fjerne trenden, hvorefter restsignalet kan beskrives af den oprindelige model; i praksis anbefales detrending af hver vindhastighedskomponent med et IIR-filter. Det designede filter anvendes på målingerne, og en omfattende sammenligning af klassifikationsresultater viser, at betydelige trends fjernes effektivt. 10-minutters middelvind er uændret, mens 10-minutters spredning reduceres. Arbejdet understøttes af Python-pakken Wind Analysis Framework (WAF), som muliggør reproduktion af resultaterne.
[This apstract has been generated with the help of AI directly from the project full text]