ENBIS: European Network for Business and Industrial Statistics

A Bayesian approach to boost wind parameter estimation by fusing historical and sampling data

23 September 2009, 11:45 – 12:00

Abstract

Submitted by

Antonio Lepore

Authors

Pasquale Erto, Antonio Lanzotti and Antonio Lepore

Affiliation

University of Naples Federico II, Dept. of Aerospace Engineering, ADDRESS: P.le Tecchio n. 80, 80125 Napoli (Italy)

Abstract

As is known, electrical energy production by renewable energy sources has a strategic role in industry today. Particularly, wind energy has proven to be one of the most competitive and fastest growing energy sources in the last decade. Currently the problem of evaluating the site-specific wind potential is faced using wind atlases and on-site anemometric long-term monitoring.
Usually information published on the wind atlases traces a territorial map of the wind speeds in correspondence with different heights. Undoubtedly this document represents a good starting-point for a preliminary site analysis, but it fails to give reliable customized information to support the evaluation and, eventually, the activation of potential investments. On the one hand, the direct anemometric on-site monitoring gives reliable information intrinsically because it is directly taken from the site. However, on the other hand, it is costly and lengthy because wide temporal windows (of at least one year) are required to characterize the site-specific wind properties accurately. In order to integrate both sources of knowledge a Bayesian model is implemented by a Markov chain Monte Carlo (MCMC) approach. The proposed methodology mathematically combines the prior information (e.g. fluid-dynamic assessment) with sampling data, in order to furnish robust and timely posterior information.
The prior distribution elicitation is carried out by using standard commercial software which generally support wind site potential experts and by adopting a practical approach suggested by technological considerations. In such a way prior information on wind features can be better incorporated in a more familiar\easier way into prior distributions for the Weibull model parameters adopted.
Simulated data and real sampling data, collected from a southern Italian site, are analysed in order to establish optimal size for monitoring window and the best filtering strategy for highly correlated anemometric data.

Return to programme