Berndt, Jonas (2018). On the predictability of exceptional error events in wind power forecasting —an ultra large ensemble approach—. PhD thesis, Universität zu Köln.

[img]
Preview
PDF
TheseJonasBerndt.pdf - Accepted Version

Download (20MB) | Preview

Abstract

Exceptional error events in wind power forecasting impose a major obstacle to today’s reliable power supply. The predictability of such error events is fundamentally restricted by the underlying weather forecast, resting on limitations of state-of-the-art numerical prediction systems. This work aims to identify such imminent forecast errors applying a probabilistic approach. To this end, the standard sizes of meteorological ensembles are increased from O(10) to an ultra large ensemble size of O(1000) members to accomplish an improved approximation of the probability density function. For this purpose, a novel approach of an ensemble control system named ESIAS-met has been developed on a Petaflop architecture. Further, an increased ensemble size favors the application of nonlinear data assimilation techniques based on the particle filter, while imposing the challenge of growing computational expenses of a resampling step within the particle filter algorithm. ESIAS-met presents a computationally efficient solution to the problem by realizing a parallel execution of the ensemble. Performance measurements demonstrate strong scalability of the system with up to 4096 members. Moreover, the computational expenses of a particle filter resampling step are shown to become independent of the ensemble size. The ESIAS-met system is further applied to investigate the benefit of an increased ensemble size on the predictability of recent exceptional error events. The analysis reveals, that despite the large ensemble size, the forecast error is only represented by single outliers. Higher order moments prove to provide a robust measure of the proper direction of forecast error and assess their likelihood of appearance. It is shown, that at least O(100) ensemble members are needed to resolve the higher order moments sufficiently well. Hence, the results achieved in this work yield important potential for future warning capabilities of exceptional error events.

Item Type: Thesis (PhD thesis)
Creators:
CreatorsEmailORCID
Berndt, Jonasjonas.berndt@gmx.deUNSPECIFIED
URN: urn:nbn:de:hbz:38-90982
Subjects: Natural sciences and mathematics
Earth sciences
Uncontrolled Keywords:
KeywordsLanguage
Numerical weather predictionEnglish
Wind power forecastingEnglish
Ensemble forecastingEnglish
Faculty: Faculty of Mathematics and Natural Sciences
Divisions: Faculty of Mathematics and Natural Sciences > Institut für Geophysik und Meteorologie
Language: English
Date: 2018
Date of oral exam: 2 March 2018
Referee:
NameAcademic Title
Elbern, HendrikPD Dr.
Shao, YapingProf. Dr.
Refereed: Yes
URI: http://kups.ub.uni-koeln.de/id/eprint/9098

Downloads

Downloads per month over past year

Export

Actions (login required)

View Item View Item