SC²S Colloquium - February 2, 2018

From Sccswiki
Revision as of 12:26, 17 January 2018 by Bakhtiar (talk | contribs) (Created page with "{| class="wikitable" |- | '''Date:''' || February 2, 2018 |- | '''Room:''' || 02.07.023 |- | '''Time:''' || 3:00 pm, s.t. |- |} == Daniel Lehmberg: Estimating Potential Power...")
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search
Date: February 2, 2018
Room: 02.07.023
Time: 3:00 pm, s.t.

Daniel Lehmberg: Estimating Potential Power Supply of an Offshore Wind Farm using Machine Learning

Wind power is becoming increasingly important in the energy mix of Germany. Because wind farms are highly flexible in regulating the power output they can also contribute to stabilizing the electricity grid - in particular with minute control energy. A requirement for participation in the control energy market is to accurately estimate the "potential power", which is the value a wind farm can produce in prevailing environmental conditions. This thesis focuses on the "potential power" method for which a model architecture is proposed that uses Machine Learning methods with "Gradient Boosted Trees" as the main model. An "Feature Selection Rule" is given to evaluate whether input data it can be used without biasing estimations during supply phases of control energy when an error cannot be measured. Fifteen months of data are available from the offshore wind farm DanTysk for evaluation. The important input factors and variables including their transformations are described. This includes the wind farm characteristics, such as wake effects, and properties of the sensor measurements. Results show that the proposed model can meet the accuracy requirements for the current pilot phase. Further improvements to reduce the error to also meet the stricter requirements for the final phase are still needed.