Wind Speed Prediction
As a graduate research assistant at the University of Georgia I developed machine learning-based predictive models of wind speed from multiple, multivariate climatological data time series for the Georgia Automated Environmental Monitoring Network (AEMN). A sampling of learning algorithms from regression trees, support vector regression, Bayesian processes, and lazy learners were applied to both inter- and intra-location prediction tasks [1]. The primary findings of my work were that Bayesian regression ensembles produced higher accuracy models than other best-in-class learners. Biased resampling of the data demonstrated performance differences were primarily due to the highly kurtotic, skewed distribution of wind speed.
References
- [1]P. Knox, S. Venkadesh, K. Nadig, A. Champion, G. Hoogenboom, W. D. Potter, and R. McClendon, “Using Climatological Data and Neural Networks to Predict Temperature, Dewpoint and Wind for Agricultural Applications.” Poster presented at the 2012 Southeast Climate Consortium Conference, April 9–11, 2012.