Combined Forecast Model for Wind Generation in Brazilian Monthly Dispatch Scheduling
Keywords:
Seasonal ARIMA, Combined Forecast, Wind Power, Holt-Winters, Energy Planning, Dispatch SchedulingAbstract
The fast growing of wind generation in Brazil has brought benefits to the country's energy supply, especially under critical hydrological conditions, as observed between 2014 and 2017. However, deviations in wind generation predictions may lead to the depletion of reservoirs and expensive thermal generation. In order to improve the accuracy of wind generation forecasts for the monthly dispatch scheduling, four different models based on exponential smoothing (Holt-Winters) and ARIMA (Box \& Jenkins) approach were proposed for Rio do Fogo wind farm, located in the Northeast of Brazil. The results show that the variance-weighted combination of an ARIMA (1,0,0) x (1,1,1)${}_{12}$ and a multiplicative Holt-Winters model provides the best fit for the studied series. In addition, the mean absolute percentage error found is 14\% lower than the one provided by the current forecast model (moving averages). Thus, it is possible to apply the ARIMA/Holt-Winters variance-weighted combination model with greater accuracy for forecasting the monthly generation of Rio do Fogo wind farm, during the dispatch planning carried out by the Brazilian independent system operator (ONS for short).