Comparative studies on different time series models for wind power generation forecasting

Integration of wind to existing energy sources requires understanding of its intermittent behavior which can be addressed by accurate forecasts. This paper opts to identify time-series models most appropriate for wind power generation forecasting. From a wind turbine dataset containing measurements of weather and mechanical conditions, wind speed has the highest correlation with active power generation, hence used for training and evaluation of the forecasting models. The stationarity test shows that the time-series wind speed data is suitable for forecasting. Utilizing these wind speed measurements, a 30-day forecast can be generated with satisfactory accuracy. Among the different forecasting models evaluated using MAE, RMSE, and R² scores, XGBoost emerges as the most suitable model, providing the best fit to the data used.