A generalized dynamical model for wind speed forecasting

The Weibull distribution is commonly used to model wind speed data, mainly due to its good fit to asymmetric positive variables. Several proposals have extended this approach to accommodate realistic features of wind data such as nonstationary behavior due to changes in atmospheric regimes. The present work considers wind speed modeling over time through the dynamic Weibull and Gamma state space models. Properties of both models are presented and filtering, smoothing and prediction equations are analytically obtained. Efficient simulation of scenarios is obtained through the beta prime distribution, which allows fast online forecasts of wind speed. The models are compared regarding fit and predictive performance for the analysis of two wind speed datasets in different regions of Brazil. Results indicate that the dynamic Gamma model is competitive with the Weibull model for wind prediction.

Automated summary

The study models wind speed using dynamic Weibull and Gamma state space models and derives filtering, smoothing, and prediction equations. Efficient simulation is achieved through the beta prime distribution for fast online wind speed forecasting. Results suggest that the dynamic Gamma model competes well with the Weibull model in wind speed prediction.