Come meet the poster presenters to ask them questions and discuss their work
Check the programme for our poster viewing moments. For more details on each poster, click on the poster titles to read the abstract.
PO095: Investigating underperformance by modelling power, relying only on limited SCADA data
Mateo Obregon, Performance Data Scientist, BayWa r.e.
Wind turbine generators (WTG) are designed to produce as much power as possible for wind speeds within a cutin to rated range. For greater wind speeds, modern turbines adjust blade pitch and rotor speed to limit production to this rated power. In effect, cutin, rated and cutout wind speeds define different wind speed regions for predicting output power. The problem of modelling power output from a WTG is essentially reduced to modelling power within the (cutin, rated) wind speed range (Wang et al., 2019). The power ~ (wind speed)3 model is a third-degree polynomial regression. Khalfallah and Koliub (2007) find that piecewise polynomial regression produce better fits because boundary points (i.e., power at cutinand rated) are poorly modelled by the former models. Furthermore, spline methods ensure that contiguous areas in the piecewise regression are differentiably continuous. These models give an expected power for any given wind speed. However, WTG SCADA comprises of averaged wind speed and power over 10 minute frames. Many factors can modify the windspeed-to-power relationship during these 10 minutes (changes in wind direction, wind speed, air density, icing or component overheating, to name a few; see Wang et al., 2019). While observed power can reasonably deviate from expected power, identifying unexpected deviations from modelled power offers an avenue to detect potential WTG underperformance problems. The difference from modelled power –the residual unexplained error– must thus be classified as either normal or underperformance. In this presentation we review several explorations we have carried out into classifying SCADA power into normal or unexplained by analysing power residuals from polynomial regressions and how to partition the (cutin, rated) wind speed range. We justify our choice of methods to classify underperformance from limited SCADA data. References Khalfallah, M. G. and Koliub, A. M.: Suggestions for improving wind turbines power curves, 209, 221–229, https://doi.org/10.1016/j.desal.2007.04.031, 2007. Wang, Y., Hu, Q., Li, L., Foley, A. M., and Srinivasan, D.: Approaches to wind power curve modeling: A review and discussion, 116, 109422, https://doi.org/10.1016/j.rser.2019.109422, 2019.