SC²S Colloquium - September 28, 2017

From Sccswiki
Jump to navigation Jump to search
Date: September 28, 2017
Room: 02.07.023
Time: 1:00 pm, s.t.

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

So far in 2017 an average of 16% of electrical power in Germany has been produced by wind power [1] - with further expansion planned. This means that wind power can be considered an important factor for grid stability in the future. Since December 2015 transmission network operators (TNO) in Germany have pursued a pre-qualification of wind farms for the minute reserve market - a short term market which helps to balance supply and demand of power in a time horizon of 15 minutes to 2 hours. Currently the process to allow pre-qualification of wind farms is in a pilot phase (until end of 2018), which aims to examine practical aspects and provide insights into a satisfactory regulation accuracy. In contrast to conventional power production, the more volatile production of renewable energy generally requires different methods and procedures. The preferred method (discussed in [2]) is for the wind farm provider to estimate the 'potential power', defined as "the power the wind farm would have produced if not regulated due to curtailment". This work focuses on a data driven approach to estimate the potential power with machine learning techniques. Currently the estimation is carried out with the xgboost algorithm. Data for training and testing is available from a German offshore wind farm for a period of 15 months with data point time intervals of 10 minutes.

-[1] https://www.energy-charts.de/energy_pie_de.htm

-[2] https://www.energiesystemtechnik.iwes.fraunhofer.de/content/dam/iwes-neu/energiesystemtechnik/de/Dokumente/Studien-Reports/20140822_Abschlussbericht_rev1.pdf