Yiannis Kamarianakis - Regime-switching forecast combinations: a 2-stage scheme for wind-farm
energy outputs

About the speaker

Yiannis Kamarianakis, PhD, is a Principal Researcher at IACM FORTH. His research focuses on statistical space-time models, which are motivated by transportation and environmental applications. He has more than 60 publications in scientific journals, book chapters, conference proceedings, technical reports etc. Dr. Kamarianakis has worked on the development of IBM's Traffic Prediction Tool, has extensive experience in applied statistics and data-science-related activities; in 2013 he received a prize in TRB’s forecasting competition and was a member of IBM’s teams that achieved 2 runners up positions in IEEE data mining competitions (2010). Currently, Dr. Kamarianakis serves as Associate Editor for IEEE Transactions on Intelligent Transportation Systems and as a member of the Editorial Advisory board for Transportation Research Part C: Emerging Technologies.

Abstract

This work computes medium term forecasts (12-36 hours ahead) of wind farm energy production, using numerical weather prediction outputs. The first stage of the procedure evaluates alternative models, such as random forests, extreme gradient boosting, polynomial regressions with numerous predictors estimated with elastic-net and lad-lasso, in terms of their accuracy in, a) downscaling wind speeds at the wind farm locations, and b) forecasting energy production. In the second stage, selected energy production forecasts are combined, with weights that depend on the levels of forecasted wind speed. A regime-switching combination scheme based on a new Smooth Transition regression model is discussed.