Table of Contents Author Guidelines Submit a Manuscript
Journal of Renewable Energy
Volume 2014 (2014), Article ID 986830, 15 pages
http://dx.doi.org/10.1155/2014/986830
Research Article

Comparison and Optimization of Neural Networks and Network Ensembles for Gap Filling of Wind Energy Data

1Department of Forest Ecosystems and Society, Oregon State University, Corvallis, OR 97331, USA
2Department of Geography, Ruhr University Bochum, 44780 Bochum, Germany

Received 28 January 2014; Revised 15 April 2014; Accepted 16 April 2014; Published 26 May 2014

Academic Editor: Shuhui Li

Copyright © 2014 Andres Schmidt and Maya Suchaneck. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Wind turbines play an important role in providing electrical energy for an ever-growing demand. Due to climate change driven by anthropogenic emissions of greenhouse gases, the exploration and use of sustainable energy sources is essential with wind energy covering a significant portion. Data of existing wind turbines is needed to reduce the uncertainty of model predictions of future energy yields for planned wind farms. Due to maintenance routines and technical issues, data gaps of reference wind parks are unavoidable. Here, we present real-world case studies using multilayer perceptron networks and radial basis function networks to reproduce electrical energy outputs of wind turbines at 3 different locations in Germany covering a range of landscapes with varying topographic complexity. The results show that the energy output values of the turbines could be modeled with high correlations ranging from 0.90 to 0.99. In complex terrain, the RBF networks outperformed the MLP networks. In addition, rare extreme values were better captured by the RBF networks in most cases. By using wind meteorological variables and operating data recorded by the wind turbines in addition to the daily energy output values, the error could be further reduced to more than 20%.