Forecast of daily output energy of wind turbine using sARIMA and nonlinear autoregressive models

Abstract

Forecast models for wind speed and wind turbine power generation are valuable support tools for operators of Control Energy Center. In this work, a year of daily energy output of a wind turbine is analyzed. The original time series was separated into a high-power sample and a low-power sample. High-power sample has a seasonal pattern while low-power sample does not. Afterward, a sARIMA model was produced for high-power sample forecast, with a good performance, while for low-power sample any ARIMA model defeated persistence model; thus, a couple of nonlinear autoregressive artificial neural networks are proposed. Mean absolute error and mean square error are reported and demonstrate that the sARIMA model can predict satisfactorily high-power sample, even with limited data, while to forecast low-power sample, it is necessary to use a neural networks approach and all data available to produce accurate forecasts. In each case, a normalized comparison with persistence model is also reported. Finally, a method which uses previous data of daily output energy and forecasted future wind speed values from a numeric weather prediction model is presented to objectively identify whether the current time is in a high-power or low-power regime to choose the ad hoc daily output energy forecast model.